Not Applicable.
Not Applicable.
This disclosure relates generally to computer networking and more particularly to database system and operation.
Computing devices are known to communicate data, process data, and/or store data. Such computing devices range from wireless smartphones, laptops, tablets, personal computers (PC), work stations, and video game devices, to data centers that support millions of web searches, stock trades, or on-line purchases every day. In general, a computing device includes a central processing unit (CPU), a memory system, user input/output interfaces, peripheral device interfaces, and an interconnecting bus structure.
As is further known, a computer may effectively extend its CPU by using “cloud computing” to perform one or more computing functions (e.g., a service, an application, an algorithm, an arithmetic logic function, etc.) on behalf of the computer. Further, for large services, applications, and/or functions, cloud computing may be performed by multiple cloud computing resources in a distributed manner to improve the response time for completion of the service, application, and/or function.
Of the many applications a computer can perform, a database system is one of the largest and most complex applications. In general, a database system stores a large amount of data in a particular way for subsequent processing. In some situations, the hardware of the computer is a limiting factor regarding the speed at which a database system can process a particular function. In some other instances, the way in which the data is stored is a limiting factor regarding the speed of exception. In yet some other instances, restricted co-process options are a limiting factor regarding the speed of execution.
The data storage systems 3 store existing data. The existing data may originate from the data gathering devices or other sources, but the data is not real time data. For example, the data storage system stores financial data of a bank, a credit card company, or like financial institution. The data system 2-N processes queries 5-N regarding the data stored in the data storage systems to produce responses 6-N.
Data system 2 processes queries regarding real time data from data gathering devices and/or queries regarding non-real time data stored in the data storage system 3. The data system 2 produces responses in regard to the queries. Storage of real time and non-real time data, the processing of queries, and the generating of responses will be discussed with reference to one or more of the subsequent figures.
Each of the sub-systems 11, 12, 13, 15, and 16 include a plurality of computing devices; an example of which is discussed with reference to one or more of
In an example of operation, the parallelized data input sub-system 11 receives a data set (e.g., a table) that includes a plurality of records. A record includes a plurality of data fields. As a specific example, the data set includes tables of data from a data source. For example, a data source includes one or more computers. As another example, the data source is a plurality of machines. As yet another example, the data source is a plurality of data mining algorithms operating on one or more computers.
As is further discussed with reference to
The parallelized data input sub-system: 11 processes a table to determine how to store it. For example, the parallelized data input sub-system 11 divides the data set into a plurality of data partitions. For each partition, the parallelized data input sub-system. It divides it into a plurality of data segments based on a segmenting factor. The segmenting factor includes a variety of approaches dividing a partition into segments. For example, the segment factor indicates a number of records to include in a segment. As another example, the segmenting factor indicates a number of segments to include in a segment group. As another example, the segmenting factor identifies how to segment a data partition based on storage capabilities of the data store and processing sub-system. As a further example, the segmenting factor indicates how many segments for a data partition based on a redundancy storage encoding scheme.
As an example of dividing a data partition into segments based on a redundancy storage encoding scheme, assume that it includes a 4 of 5 encoding scheme (meaning any 4 of 5 encoded data elements can be used to recover the data). Based on these parameters, the parallelized data input sub-system 11 divides a data partition into 5 segments: one corresponding to each of the data elements).
The parallelized data input sub-system 11 restructures the plurality of data segments to produce restructured data segments. For example, the parallelized data input sub-system II restructures records of a first data segment of the plurality of data segments based on a key field of the plurality of data fields to produce a first restructured data segment. The key field is common to the plurality of records. As a specific example, the parallelized data input sub-system 11 restructures a first data segment by dividing the first data segment into a plurality of data slabs (e.g., columns of a segment of a partition of a table). Using one or more of the columns as a key, or keys, the parallelized data input sub-system 11 sorts the data slabs. The restructuring to produce the data slabs is discussed in greater detail with reference to
The paralyzed data input sub-system 11 also generates storage instructions regarding how sub-system 12 is to store the restructured data segments for efficient processing of subsequently received queries regarding the stored data. For example the storage instructions include one or more of: a naming scheme, a request to store, a memory resource requirement a processing resource requirement, an expected access frequency level, an expected storage duration a required maximum access latency time. and other requirements associated with storage, processing, and retrieval of data.
A designated computing device of the parallelized data store, retrieve, and/or process sub-system 12 receives the restructured data segments and the storage instructions. The designated computing device (which is randomly selected, selected it a round robin manner, or by default) interprets the storage instructions to identify resources (e.g., itself, its components, other computing devices, and/or components thereof) within the computing device's storage cluster. The designated computing device then divides the restructured data segments of a segment group of a partition of a table into segment divisions based on the identified resources and/or the storage instructions. The designated computing device then sends the segment divisions to the identified resources for storage and subsequent processing in accordance with a query. The operation of the parallelized data store. retrieve, and/or process sub-system 12 is discussed in greater detail with reference to
The parallelized query and response sub-system 13 receives queries regarding tables (e.g., data sets) and processes the queries prior to sending them to the parallelized data store, retrieve, and/or process sub-system 12 for execution. For example, the parallelized query and response sub-system 13 generates an initial query plan based on a data processing request (e.g., a query) regarding a data set (e.g., the tables). Sub-system 13 optimizes the initial query plan based on one or more of the storage instructions, the engaged resources, and optimization functions to produce an optimized query plan.
For example, the parallelized query and response sub-system 13 receives a specific query no. 1 regarding the data set no. 1 (e.g., a specific table). The query is in a standard query format such as Open Database Connectivity (ODBC). Java Database Connectivity (JDBC), and/or SPARK. The query is assigned to a node within the parallelized query and response sub-system 13 for processing. The assigned node identifies the relevant table, determines where and how it is stored, and determines available nodes within the parallelized data store, retrieve, and/or process sub-system 12 for processing the query.
In addition, the assigned node parses the query to create an abstract syntax tree. As a specific example, the assigned node converts an SQL (Structured Query Language) statement into a database instruction set. The assigned node then validates the abstract syntax tree. If not valid, the assigned node generates a SQL exception, determines an appropriate correction, and repeats. When the abstract syntax tree is validated, the assigned node then creates an annotated abstract syntax tree. The annotated abstract syntax tree includes the verified abstract syntax tree plus annotations regarding column names, data type(s), data aggregation or not, correlation or not, sub-query or not, and so on.
The assigned node then creates an initial query plan from the annotated abstract syntax tree. The assigned node optimizes the initial query plan using a cost analysis function (e.g., processing time, processing resources, etc.) and/or other optimization functions. Having produced the optimized query plan, the parallelized query and response sub-system 13 sends the optimized query plan to the parallelized data store, retrieve, and/or process sub-system 12 for execution. The operation of the parallelized query and response sub-system 13 is discussed in greater detail with reference to
The parallelized data store, retrieve, and/or process sub-system 12 executes the optimized query plan to produce resultants and sends the resultants to the parallelized query and response sub-system 13. Within the parallelized data store, retrieve. and/or process sub-system 12 a computing device is designated as a primary device for the query plan (e.g., optimized query plan) and receives it. The primary device processes the query plan to identify nodes within the parallelized data store, retrieve, and/or process sub-system 12 for processing the query plan. The primary device then sends appropriate portions of the query plan to the identified nodes for execution. The primary device receives responses from the identified nodes and processes them in accordance with the query plan.
The primary device of the parallelized data store, retrieve, and/or process sub-system 12 provides the resulting response (e.g., resultants) to the assigned node of the parallelized query and response sub-system 13. For example, the assigned node determines whether further processing is needed on the resulting response (e.g., joining. filtering, etc.). If not, the assigned node outputs the resulting response as the response to the query (e.g., a response for query no. 1 regarding data set no. 1). If, however further processing is determined, the assigned node further processes the resulting response to produce the response to the query. Having received the resultants, the parallelized query and response sub-system 13 creates a response from the resultants for the data processing request.
As will be described in greater detail with reference to one or more subsequent figures a computing device includes a plurality of nodes and each node includes a plurality of processing core resources. Each processing core resource is capable of executing at least a portion of an administrative operation independently. This supports lock free and parallel execution of one or more administrative operations.
The administrative sub-system IS functions to store metadata of the data set described with reference to
The parallelized ingress sub-system: 24 includes a plurality of ingress data sub-systems 25-1 through 25-p that each include a local communication resource of local communication resources 26-1 through 26-p and a plurality of computing devices 18-1 through 18-n. A computing device executes an ingress data processing function (e.g., 28-1) to receive streaming data regarding a table via a wide area network 22 and processing it for storage as generally discussed with reference to
In general, the bulk data processing function is geared towards receiving data of a table in a bulk fashion (e.g., the table exists and is being retrieved as a whole, or portion thereof). The ingress data processing function is geared towards receiving streaming data from one or more data sources (e.g., receive data of a table as the data is being generated). For example, the ingress data processing function is geared towards receiving data from a plurality of machines in a factory in a periodic or continual manner as the machines create the data.
Processing resources of the parallelized data store, retrieve, &/or process sub-system: 12 processes the components of the optimized plan to produce results components 32-1 through 32-n. The computing device of the Q&R sub-system 13 processes the result components to produce a query response.
The Q&R sub-system 13 allows for multiple queries regarding one or more tables to be processed concurrently. For example a set of processing core resources of a computing device (e.g., one or more processing core resources) processes a first query and a second set of processing core resources of the computing device (or a different computing device) processes a second query.
As will be described in greater detail with reference to one or more subsequent figures a computing device includes a plurality of nodes and each node includes multiple processing core resources such that a plurality of computing devices includes pluralities of multiple processing core resources A processing core resource of the pluralities of multiple processing core resources generates the optimized query plan and other processing core resources of the pluralities of multiple processing core resources generates other optimized query plans for other data processing requests. Each: processing core resource is capable of executing at least a portion of the Q & R function. In an embodiment, a plurality of processing core resources of one or more nodes executes the Q & R function to produce a response to a query. The processing core resource is discussed in greater detail with reference to
In an embodiment, the parallelized data store, retrieve, and/or process sub-system 12 includes a plurality of storage clusters 35-1 through 35-z. Each storage cluster includes a corresponding local communication resource 26-1 through 26-z and a number of computing devices 18-1 through 18-5. Each computing device executes an input, output, and processing (IO &P) processing function 34-1 through 34-5 to store and process data.
The number of computing devices in a storage cluster corresponds to the number of segments (e.g., a segment group) in which a data partitioned is divided. For example, if a data partition is divided into five segments a storage cluster includes five computing devices. As another example, if the data is divided into eight segments, then there are eight computing devices in the storage clusters.
[01SS] To store a segment group of segments 29 within a storage cluster, a designated computing device of the storage cluster interprets storage instructions to identify computing devices (and/or processing core resources thereof) for storing the segments to produce identified engaged resources. The designated computing device is selected by a random selection a default selection, a round-robin selection, or any other mechanism for selection.
The designated computing device sends a segment to each computing device in the storage cluster, including itself. Each of the computing devices stores their segment of the segment group. As an example five segments 29 of a segment group are stored by five computing devices of storage cluster 35-1. The first computing device 18-1-1 stores a first segment of the segment group: a second computing device 18-2-1 stores a second segment of the segment group; and so on. With the segments stored, the computing devices are able to process queries (e.g., query components from the Q&R sub-system 13) and produce appropriate result components.
While storage cluster 35-1 is storing and/or processing a segment group, the other storage clusters 35-2 through 35-n are storing and/or processing other segment groups. For example, a table is partitioned into three segment groups. Three storage clusters store and/or process the three segment groups independently. As another example, four tables are independently stored and/or processed by one or more storage clusters. As yet another example, storage cluster 35-1 is storing and/or processing a second segment group while it is storing/or and processing a first segment group.
In an embodiment, each node is capable of operating independently of the other nodes. This allows for large scale parallel operation of a query request, which significantly reduces processing time for such queries. In another embodiment, one or more node function as co-processors to share processing requirements of a particular function, or functions.
The disk memory 38 includes a plurality of memory interface modules 43-1 through 43-0 and a plurality of memory devices 42-1 through 42-n (e.g., non-volatile memory). The memory devices 42-1 through 42-n include, but are not limited 10. solid state memory, disk drive memory, cloud storage memory, and other non-volatile memory. For each type of memory device. a different memory interface module 43-1 through 43-n is used. For example, solid state memory uses a standard, or serial. ATA (SATA). variation, or extension thereof, as its memory interface. As another example, disk drive memory devices use a small computer system interface (SCSI), variation, or extension thereof, as its memory interface.
In an embodiment, the disk memory 38 includes a plurality of solid state memory devices and corresponding memory interface modules. In another embodiment, the disk memory 38 includes a plurality of solid state memory devices, a plurality of disk memories, and corresponding memory interface modules.
The network connection 41 includes a plurality of network interface modules 46-1 through 46-n and a plurality of network cards 47-1 through 47-n. A network card includes a wireless LAN (WLAN) device (e.g., an IEEE 802.11 n or another protocol), a LAN device (e.g., Ethernet), a cellular device (e.g., CDMA), etc. The corresponding network interface modules 46—through: 46-n include a software driver for the corresponding network card and a physical connection that couples the network card to the central processing module 39 or other component(s) of the node.
The connections between the central processing module 39, the main memory 40, the disk memory 38, and the network connection 41 may be implemented in a variety of ways. For example, the connections are made through a node controller (e.g., a local version of the computing device controller hub 36). As another example, the connections are made through the computing device controller hub 36.
The main memory 40 is divided into a computing device (CD) 56 section and a database (DB) 51 section. The database section includes a database operating system (OS) area 52, a disk area 53, a network area 54, and a general area 55. The computing device section includes a computing device operating system (OS) area 57 and a general area 58. Note that each section could include more or less allocated areas for various tasks being executed by the database system.
In general, the database OS 52 allocates main memory for database operations Once allocated, the computing device OS 57 cannot access that portion of the main memory 40. This supports lock free and independent parallel execution of one or more operations.
The database overriding operating system (DB OS) 61 includes custom DB device management 69, custom DB process management 70 (e.g., process scheduling and/or inter-process communication & synchronization), custom DB file system management 71, custom DB memory management 72, and/or custom security 73. In general, the database overriding OS 61 provides hardware components of a node for more direct access to memory, more direct access to a network connection, improved independency, improved data storage, improved data retrieval, and/or improved data processing than the computing device OS.
In an example of operation, the database overriding OS 61 controls which operating system, or portions thereof, operate with each node and/or computing device controller hub of a computing device (e.g., via OS select 75-1 through 75-n when communicating with nodes 37-1 through 37-n and via OS select 75-m when communicating with the computing device controller bub 36). For example device management of a node is supported by the computer operating system, while process management. memory management, and file system management are supported by the database overriding operating system. To override the computer OS, the database overriding OS provides instructions to the computer OS regarding which management tasks will be controlled by the database overriding OS. The database overriding OS also provides notification to the computer OS as to which sections of the main memory it is reserving exclusively for one or more database functions, operations, and/or tasks. One or more examples of the database overriding operating system are provided in subsequent figures.
The database system 10 can be implemented as a massive scale database system that is operable to process data at a massive scale. As used herein a massive scale refers to a massive number of records of a single dataset and/or many datasets, such: as millions, billions, and/or trillions of records that collectively include many Gigabytes, Terabytes, Petabytes, and/or Exabytes of data. As used herein, a massive scale database system refers to a database system operable to process data at a massive scale. The processing of data at this massive scale can be achieved via a large number, such as hundreds, thousands, and/or millions of computing devices 18. nodes 37, and/or processing core resources 48 performing various functionality of database system 10 described herein in parallel, for example, independently and/or without coordination.
Such processing of data at this massive scale cannot practically be performed by the human mind. In particular, the human mind is not equipped to perform processing of data at a massive scale. Furthermore, the human mind is not equipped to perform hundreds, thousands, and/or millions of independent processes in parallel, within overlapping time spans. The embodiments of database system 10 discussed herein improves the technology of database systems by enabling data to be processed at a massive scale efficiently and/or reliably.
In particular, the database system 10 can be operable to receive data and/or to store received data at a massive scale. For example, the parallelized input and/or storing of data by the database system 10 achieved by utilizing the parallelized data input sub-system 11 and/or the parallelized data store, retrieve, and/or process sub-system 12 can cause the database system 10 to receive records for storage at a massive scale, where millions, billions, and/or trillions of records that collectively include many Gigabytes. Terabytes. Petabytes, and/or Exabytes can be received for storage, for example, reliably, redundantly and/or with a guarantee that no received records are missing in storage and/or that no received records are duplicated in storage. This can include processing real-time and/or near-real time data streams from one or more data sources al a massive scale based on facilitating ingress of these data streams in parallel. To meet the data rates required by these one or more real-time data streams, the processing of incoming data streams can be distributed across hundreds, thousands, and/or millions of computing devices 18. nodes 37, and/or processing core resources 48 for separate, independent processing with minimal and/or no coordination. The processing of incoming data streams for storage at this scale and/or this data rate cannot practically be performed by the human mind. The processing of incoming data streams for storage at this scale and/or this data rate improves database system by enabling greater amounts of data to be stored in databases for analysis and/or by enabling real-time data to be stored and utilized for analysis. The resulting richness of data stored in the database system can improve the technology of database systems by improving the depth and/or insights of various data analyses performed upon this massive scale of data.
Additionally, the database system 10 can be operable to perform queries upon data at a massive scale. For example, the parallelized retrieval and processing of data by the database system 10 achieved by utilizing the parallelized query and results sub-system 13 and/or the parallelized data store, retrieve, and/or process sub-system 12 can cause the database system 10 to retrieve stored records at a massive scale and/or to and/or filter, aggregate, and/or perform query operators upon records at a massive scale in conjunction with query execution, where millions, billions, and/or trillions of records that collectively include many Gigabytes. Terabytes, Petabytes, and/or Exabytes can be accessed and processed in accordance with execution of one or more queries at a given time, for example, reliably, redundantly and/or with a guarantee that no records are inadvertently missing from representation in a query resultant and/or duplicated in a query resultant. To execute a query against a massive scale of records in a reasonable amount of any such as a small number of seconds, minutes, or hours, the processing of a given query can be distributed across hundreds, thousands, and/or millions of computing devices 18, nodes 37, and/or processing core resources 48 for separate. independent processing with minimal and/or no coordination. The processing of queries at this massive scale and/or this data rate cannot practically be performed by the human mind. The processing of queries at this massive scale improves the technology of database systems by facilitating greater depth and/or insights of query resultants for queries performed upon this massive scale of data.
Furthermore, the database system 10 can be operable to perform multiple queries concurrently upon data al a massive scale. For example, the parallelized retrieval and processing of data by the database system 10 achieved by utilizing the parallelized query and results sub-system 13 and/or the parallelized data store, retrieve, and/or process sub-system 12 can cause the database system 10 to perform multiple queries concurrently, for example, in parallel, against data at this massive scale, where hundreds and/or thousands of queries can be performed against the same, massive scale dataset within a same time frame and/or in overlapping time frames. To execute multiple concurrent queries against a massive scale of records in a reasonable amount of time such as a small number of seconds, minutes, or hours, the processing of a multiple queries can be distributed across hundreds. thousands, and/or millions of computing devices 18, nodes 37, and/or processing core resources 48 for separate, independent processing with minimal and/or no coordination. A given computing devices 18, nodes 37, and/or processing core resources 48 may be responsible for participating in execution of multiple queries at a same time and/or within a given time frame, where its execution of different queries occurs within overlapping time frames. The processing of many concurrent queries at this massive scale and/or this data rate cannot practically be performed by the human mind. The processing of concurrent queries improves the technology of database systems by facilitating greater numbers of users and/or greater numbers of analyses to be serviced within a given time frame and/or over time.
As an example, the table is regarding a fleet of vehicles. Each row represents data regarding a unique vehicle. The first column stores a vehicle ID, the second column stores make and model information of the vehicle. The third column stores data as to whether the vehicle is on or off. The remaining columns store data regarding the operation of the vehicle such as mileage. gas level, oil level, maintenance information, routes taken, etc.
With the third column selected as the key column, the other columns of the segment are to be sored based on the key column. Prior to being sorted, the columns are separated to form data slabs. As such, one column is separated out to form one data slab.
Before the sorted data slabs are stored in the data & parity section, or concurrently with storing in the data & parity section, the sorted data slabs of a segment are redundancy encoded. The redundancy encoding may be done in a variety of ways. For example, the redundancy encoding is in accordance with RAID 5. RAID 6, or RAID 10. As another example, the redundancy encoding is a form of forward error encoding (e.g., Reed Solomon, Trellis, etc.). As another example, the redundancy encoding utilizes an erasure coding scheme. An example of redundancy encoding is discussed in greater detail with reference to one or more of
The manifest section stores metadata regarding the sorted data slabs. The metadata includes one or more of, but is not limited to, descriptive metadata, structural metadata, and/or administrative metadata. Descriptive metadata includes one or more of, but is not limited to, information regarding data such as name, an abstract, keywords, another, etc. Structural metadata includes one or more of, but is not limited to, structural features of the data such as page size page ordering, formatting, compression information, redundancy encoding information logical addressing information, physical addressing information, physical to logical addressing information, etc. Administrative metadata includes one or more of, but is not limited to, information that aids in managing data such as file type, access privileges, rights management, preservation of the data, etc.
The key column is stored in an index section. For example a first key column is stored in index #0. If a second key column exists, it is stored in index #1. As such, for each key column it is stored in its own index section. Alternatively, one or more key columns are stored in a single index section.
The statistics section stores statistical information regarding the segment and/or the segment group. The statistical information includes one or more of, but is not limited, to number of rows (e.g., data values) in one or more of the sorted data slabs, average length of one or more of the sorted data slabs, average row size (e.g., average size of a data value), etc. The statistical information includes information regarding raw dam slabs, raw panty data, and/or compressed data slabs and parity data.
This illustration of query execution plan 2405 illustrates the flow of execution of a given query by utilizing a subset of nodes across some or all of the levels 2410. In this illustration. nodes 37 with a solid outline are nodes involved in executing a given query. Nodes 37 with a dashed outline are other possible nodes that are not involved in executing the given query, bra could be involved in executing other queries in accordance with their level of the query execution plan in which they are included.
Each of the nodes of IO level 2416 can be operable to, for a given query, perform the necessary row reads for gathering corresponding rows of the query. These row reads can correspond to the segment retrieval to read some or all of the rows of retrieved segments determined to be required for the given query. Thus, the nodes 37 in level 2416 can include any nodes 37 operable to retrieve segments for query execution from its own storage or from storage by one or more other nodes, to recover segment for query execution via other segments in the same segment grouping by utilizing the redundancy error encoding scheme: and/or to determine which exact set of segments is assigned to the node for retrieval to ensure queries are executed correctly.
IO level 2416 can include all nodes in a given storage cluster 35 and/or can include some or all nodes in multiple storage clusters 35, such as all nodes in a subset of the storage clusters 35-1-35-2 and/or all nodes in all storage clusters 35-1-35-z. For example, all nodes 37 and/or all currently available nodes 37 of the database system 10 can be included in level 2416. As another example IO level 2416 can include a proper subset of nodes in the database system, such as some or all nodes that have access to stored segments and/or that are included in a segment set 35. In some cases. nodes 37 that do not store segments included in segment sets, that do not have access to stored segments, and/or that are not operable to perform row reads are not included at the IO level, but can be included at one or more inter levels 2414 and/or root level 2412.
The query executions discussed herein by nodes in accordance with executing queries at level 2416 can include retrieval of segments; extracting some or all necessary rows from the segments with some or all necessary columns; and sending these retrieved rows to a node at the next level 2410. H-J as the query resultant generated by the node 37. For each node 37 at IO level 2416, the set of raw rows retrieved by the node 37 can be distinct from rows retrieved from all other nodes, for example, to ensure correct query execution. The total set of rows and/or corresponding columns retrieved by nodes 37 in the 10 level for a given query can be dictated based on the domain of the given query, such as one or more tables indicated in one or more SELECT statements of the query, and/or can otherwise include all data blocks that are necessary to execute the given query.
Each inner level 2414 can include a subset of nodes 37 in the database system 10. Each level 2414 can include a distinct set of nodes 37 and/or some or more levels 2414 can include overlapping sets of nodes 37. The nodes 37 at inner levels are implemented, for each given query, to execute queries in conjunction with operators for the given query. For example, a query operator execution flow can be generated for a given incoming query, where an ordering of execution of its operators is determined. and this ordering is utilized to assign one or more operators of the query operator execution flow to each node in a given inner level 2414 for execution. For example, each node at a same inner level can be operable to execute a same set of operators for a given query, in response to being selected to execute the given query, upon incoming resultants generated by nodes at a directly lower level to generate its own resultants sent to a next higher level. In particular, each node at a same inner level can be operable to execute a same portion of a same query operator execution flow for a given query. In cases where there is exactly one inner level, each node selected to execute a query at a given inner level performs some or all of the given query's operators open the raw rows received as resultants from the nodes at the IO level, such as the entire query operator execution flow and/or the portion of the query operator execution flow performed upon data that has already been read from storage by nodes at the IO level. In some cases, some operators beyond row reads are also performed by the nodes at the IO level. Each node at a given inner level 2414 can further perform a gather function to collect, union, and/or aggregate resultants sent from a previous level, for example in accordance with one or more corresponding operators of the given query.
The root level 2412 can include exactly one node for a given query that gathers resultants from every node at the top-most inner level 2414. The node 37 at root level 2412 can perform additional query operators of the query and/or can otherwise collect, aggregate, and/or union the resultants from the top-most inner level 2414 to generate the final resultant of the query, which includes the resulting set of rows and/or one or more aggregated values, in accordance with the query, based on being performed on all rows required by the query. The root level node can be selected from a plurality of possible root level nodes, where different root nodes are selected for different queries. Alternatively, the same root node can be selected for all queries.
As depicted in
In some cases, the IO level 2416 always includes the same set of nodes 37, such as a full set of nodes and/or all nodes that are in a storage cluster 35 that stores data required to process incoming queries. In some cases, the lowest inner level corresponding to level 2410. H-1 includes at least one node from the IO level 2416 in the possible set of nodes. In such cases, while each selected node in level 2410.H-1 is depicted to process resultants sent from other nodes 37 in
The node 37 at root level 2412 can be fixed for all queries, where the set of possible nodes at root level 2412 includes only one node that executes all queries at the root level of the query execution plan. Alternatively, the root level 2412 can similarly include a set of possible nodes, where one node selected from this set of possible nodes for each query and where different nodes are selected from the set of possible nodes for different queries. In such cases, the nodes at inner level 2410.2 determine which of the set of possible root nodes to send their resultant to. In some cases, the single node or set of possible nodes at root level 24.12 is a proper subset of the set of nodes at inner level 2410.2, and/or is a proper subset of the set of nodes at the IO level 2416. In cases where the root node is included at inner level 2410.2, the root node generates its own resultant in accordance with inner level 2410.2, for example, based on multiple resultants received from nodes at level 2410.3, and gathers its resultant that was generated in accordance with inner level 2410.2 with other resultants received from nodes at inner level 2410.2 to ultimately generate the final resultant in accordance with operating as the root level node.
In some cases where nodes are selected from a set of possible nodes at a given level for processing a given query, the selected node must have been selected for processing this query at each lower level of the query execution tree. For example, if a particular node is selected to process a node at a particular inner level, it must have processed the query to generate resultants at every lower inner level and the IO level. In such cases, each selected node at a particular level will always use its own resultant that was generated for processing at the previous, lower level, and will gather this resultant with other resultants received from other child nodes at the previous, lower level. Alternatively, nodes that have not yet processed a given query can be selected for processing at a particular level, where all resultants being gathered are therefore received from a set of child nodes that do not include the selected node.
The configuration of query execution plan 2405 for a given query can be determined in a downstream fashion, for example, where the tree is formed from the root downwards. Nodes at corresponding levels are determined from configuration information received from corresponding parent nodes and/or nodes at higher levels, and can each send configuration information to other nodes, such as their own child nodes, at lower levels until the lowest level is reached. This configuration information can include assignment of a particular subset of operators of the set of query operators that each level and/ot each node will perform for the query. The execution of the query is performed upstream in accordance with the determined configuration, where IO reads are performed first, and resultants are forwarded upwards until the root node ultimately generates the query result.
As used herein, execution of a particular query by a particular node 37 can correspond to the execution of the portion of the particular query assigned to the particular node in accordance with full execution of the query by the plurality of nodes involved in the query execution plan 2405. This portion of the particular query assigned to a particular node can correspond to execution plurality of operators indicated by a query operator execution flow 2433. In particular, the execution of the query for a node 37 al an inner level 2414 and/or root level 2412 corresponds to generating a resultant by processing all incoming resultants received from nodes at a lower level of the query execution plan 2405 that send their own resultants to the node 37. The execution of the query for a node 37 at the IO level corresponds to generating all resultant data blocks by retrieving and/or recovering all segments assigned to the node 37.
Thus, as used herein, a node 37's full execution of a given query corresponds to only a portion of the query's execution across all nodes in the query execution plan 2405. In particular a resultant generated by an inner level node 37's execution of a given query may correspond to only a portion of the entire query result, such as a subset of rows in a final result set, where other nodes generate their own resultants to generate other portions of the full resultant of the query. In such embodiments, a plurality of nodes at this inner level can fully execute queries on different portions of the query domain independently in parallel by utilizing the same query operator execution flow 2433 Resultants generated by each of the plurality of nodes at this inner level 2414 can be gathered into a final result of the query, for example, by the node 37 at root level 2412 if this inner level is the top-most inner level 2414 or the only inner level 2414. As another example, resultants generated by each of the plurality of nodes at this inner level 2414 can be further processed via additional operators of a query operator execution flow 2433 being implemented by another node at a consecutively higher inner level 2414 of the query execution plan 2405, where all nodes at this consecutively higher inner level 2414 all execute their own same query operator execution flow 2433.
As discussed in further detail herein, the resultant generated by a node 37 can include a plurality of resultant data blocks generated via a plurality of partial query executions. As used herein a partial query execution performed by a node corresponds to generating a resultant based on only a subset of the query input received by the node 37. In particular, the query input corresponds to all resultants generated by one or more nodes at a lower level of the query execution plan that send their resultants to the node. However, this query input can correspond to a plurality of input data blocks received over time, for example, in conjunction with the one or more nodes at the lower level processing their own input data blocks received over time to generate their resultant data blocks sent to the node over time. Thus, the results generated by a node's full execution of a query can include a plurality of resultant data blocks, where each resultant data block is generated by processing a subset of all input data blocks as a partial query execution upon the subset of all data blocks via the query operator execution flow 2433.
As illustrated in
Each segment 2424 stored in memory drive 2425 can be generated as discussed previously in conjunction with
Thus, in addition to performing the first stage of query execution by being responsible for row reads, nodes 37 can be utilized for database storage, and can each locally store a set of segments in its own memory drives 2425. In some cases, a node 37 can be responsible for retrieval of only the records stored in its own one or more memory drives 2425 as one or more segments 2424. Executions of queries corresponding to retrieval of records stored by a particular node 37 can be assigned to that particular node 37. In other embodiments, a node 37 does not use its own resources to store segments. A node 37 can access its assigned records for retrieval via memory resources of another node 37 and/or via other access to memory drives 2425, for example, by utilizing system: communication resources 14.
The query processing module 2435 of the node 37 can be utilized to read the assigned by first retrieving or otherwise accessing the corresponding redundancy-coded segments 2424 that include the assigned records its one or more memory drives 24425. Query processing module 2435 can include a record extraction module 2438 that is then utilized to extract or otherwise read some or all records from these segments 2424 accessed in memory drives 2425, for example, where record data of the segment is segregated from other information such as parity data included in the segment and/or where this data containing the records is converted into row-formatted records from the column-formatted row data stored by the segment. Once the necessary records of a query are read by the node 37, the node can further utilize query processing module 2435 to send the retrieved records all al once, or in a stream as they are retrieved from memory drives 2425, as data blocks to the next node 37 in the query execution plan 2405 via system communication resources 14 or other communication channels.
Note that the embodiments of node 37 discussed herein can be configured to execute multiple queries concurrently by communicating with nodes 37 in the same or different tree configuration of corresponding query execution plans and/or by performing query operations upon data blocks and/or read records for different queries. In particular, incoming data blocks can be received from other nodes for multiple different queries in any interleaving order, and a plurality of operator executions upon incoming data blocks for multiple different queries can be performed in any order, where output data blocks are generated and sent to the same or different next node for multiple different queries in any interleaving order. IO level nodes can access records for the same or different queries any interleaving order. Thus, at a given point in time, a node 37 can have already begun its execution of at least two queries, where the node 37 has also not yet completed its execution of the at least two queries.
A query execution plan 2405 can guarantee query correctness based on assignment data sent to or otherwise communicated to all nodes at the IO level ensuring that the set of required records in query domain data of a query, such as one or more tables required to be accessed by a query, are accessed exactly one time: if a particular record is accessed multiple times in the same query and/or is not accessed, the query resultant cannot be guaranteed to be correct. Assignment data indicating segment read and/or record read assignments to each of the set of nodes 37 at the 10 level can be generated, for example, based on being mutually agreed upon by all nodes 37 at the IO level via a consensus protocol executed between all nodes at the 10 level and/or distinct groups of nodes 37 such as individual storage clusters 35. The assignment data can be generated such that every record in the database system and/or in query domain of a particular query is assigned to be read by exactly one node 37. Note that the assignment data may indicate that a node 37 is assigned to read some segments directly from memory as illustrated in
Assuming all nodes 37 read all required records and send their required records to exactly one next node 37 as designated with the query execution plan 2405 for the given query, the use of exactly one instance of each record can be guaranteed Assuming all inner level nodes 37 process all the required records received from the corresponding set of nodes 37 in the IO level 2416, via applying one or more query operators assigned to the node in accordance with their query operator execution flow 2433, correctness of their respective partial resultants can be guaranteed. This correctness can further require that nodes 37 at the same level intercommunicate by exchanging records in accordance with JOIN operations as necessary, as records received by other nodes may be required to achieve the appropriate result of a JOIN operation. Finally, assuming the root level node receives all correctly generated partial resultants as data blocks from its respective set of nodes at the penultimate, highest further level 2414 as designated in the query execution plan 2405, and further assuming the root level node appropriately generates its own final resultant, the correctness of the final resultant can be guaranteed.
In some embodiments, each node 37 in the query execution plan can monitor whether it has received all necessary data blocks to fulfill its necessary role in completely generating its own resultant to be sent to the next node 37 in the query execution plan A node 37 can determine receipt of a complete set of data blocks that was sent from a particular node 37 at an immediately lower level, for example, based on being numbered and/or have an indicated ordering in transmission from the particular node 37 at the immediately lower level, and/or based on a final data block of the set of data blocks being tagged in transmission from the particular node 37 at the immediately lower level to indicate it is a final data block being sent. A node 37 can determine the required set of lower level nodes from which it is to receive data blocks based on its knowledge of the query execution plan 2405 of the query A node 37 can thus conclude when a complete set of data blocks has been received each designated lower level node in the designated set as indicated by the query execution plan 2405. This node 37 can therefore determine itself that all required data blocks have been processed into data blocks sent by this node 37 to the next node 37 and/or as a final resultant if this node 37 is the root node. This can be indicated via lagging of its own last data block, corresponding to the final portion of the resultant generated by the node, where it is guaranteed that all appropriate data was received and processed into the set of data blocks sent by this node 37 in accordance with applying its own query operator execution flow 2433.
In some embodiments, if any node 37 determines it did not receive all of its required data blocks, the node 37 itself cannot fulfill generation of its own set of required data blocks. For example, the node 37 will not transmit a final data block tagged as the “last” data block in the set of outputted data blocks to the next node 37, and the next node 37 will thus conclude there was an error and will not generate a full set of data blocks itself. The root node, and/or these intermediate nodes that never received all their data and/or never fulfilled their generation of all required data blocks, can independently determine the query was unsuccessful. In some cases, the root node, upon determining the query was unsuccessful, can initiate re-execution of the query by re-establishing the same or different query execution plan 2405 in a downward fashion as described previously, where the nodes 37 in this re-established query execution plan 2405 execute the query accordingly as though it were a new query. For example, in the case of a node failure that caused the previous query to fail, the new query execution plan 2405 can be generated to include only available nodes where the node that failed is not included in the new query execution plan 2405.
While
The shuffle node sets 2485 can be utilized to enable transfer of information between nodes, for example, in accordance with performing particular operations in a given query that cannot be performed in isolation. For example, some queries require that nodes 37 receive data blocks from its children nodes in the query execution plan for processing, and that the nodes 37 additionally receive data blocks from other nodes at the same level 2410. In particular, query operations such as JOIN operations of a SQL query expression may necessitate that some or all additional records that were access in accordance with the query be processed in tandem to guarantee a correct resultant, where a node processing only the records retrieved from memory by its child IO nodes is not sufficient.
In some cases, a given node 37 participating in a given inner level 2414 of a query execution plan may send data blocks to some or all other nodes participating in the given inner level 2414, where these other nodes utilize these data blocks received from the given node to process the query via their query processing module 2435 by applying some or all operators of their query operator execution flow 2433 to the data blocks received from the given node. In some cases, a given node 37 participating in a given inner level 2414 of a query execution plan may receive data blocks to some or all other nodes participating in the given inner level 2414, where the given node utilizes these data blocks received from the other nodes to process the query via their query processing module 2435 by applying some or all operators of their query operator execution flow 2433 to the received data blocks.
This transfer of data blocks can be facilitated via a shuffle network 2480 of a corresponding shuffle node set 2485. Nodes in a shuffle node set 2485 can exchange data blocks in accordance with executing queries, for example, for execution of particular operators such as JOIN operators of their query operator execution flow 2433 by utilizing a corresponding shuffle network 2480. The shuffle network 2480 can correspond to any wired and/or wireless communication network that enables bidirectional communication between any nodes 37 communicating with the shuffle network 2480. In some cases, the nodes in a same shuffle node set 2485 are operable to communicate with some or all other nodes in the same shuffle node set 2485 via a direct communication link of shuffle network 2480, for example, where data blocks can be routed between some or all nodes in a shuffle network 2480 without necessitating any relay nodes 37 for touting the data blocks In some cases, the nodes in a same shuffle set can broadcast data blocks.
In some cases, some nodes in a same shuffle node set 2485 do not have direct links via shuffle network 2480 and/or cannot send or receive broadcasts via shuffle network 2480 to some or all other nodes 37. For example, at least one pair of nodes in the same shuffle node set cannot communicate directly. In some cases, some pairs of nodes in a same shuffle node set can only communicate by routing their data via at least one relay node 37. For example, two nodes in a same shuffle node set do not have a direct communication link and/or cannot communicate via broadcasting their data blocks. However, if these two nodes in a same shuffle node set can each communicate with a same third node via corresponding direct communication links and/or via broadcast. this third node can serve as a relay node to facilitate communication between the two nodes. Nodes that are “further apart” in the shuffle network 2480 may require multiple relay nodes.
Thus, the shuffle network 2480 can facilitate communication between all nodes 37 in the corresponding shuffle node set 2485 by utilizing some or all nodes 37 in the corresponding shuffle node set 2485 as relay nodes, where the shuffle network 2480 is implemented by utilizing some or all nodes in the nodes shuffle node set 2485 and a corresponding set of direct communication links between pairs of nodes in the shuffle node set 2485 to facilitate data transfer between any pair of nodes in the shuffle node set 2485. Note that these relay nodes facilitating data blocks for execution of a given query within a shuffle node sets 2485 to implement shuffle network 2480 can be nodes participating in the query execution plan of the given query and/or can be nodes that are not participating in the query execution plan of the given query. In some cases, these relay nodes facilitating data blocks for execution of a given query within a shuffle node sets 2485 are strictly nodes participating in the query execution plan of the given query. In some cases, these relay nodes facilitating data blocks for execution of a given query within a shuffle node sets 2485 are strictly nodes that are not participating in the query execution plan of the given query.
Different shuffle node sets 2485 can have different shuffle networks 2480. These different shuffle networks 2480 can be isolated, where nodes only communicate with other nodes in the same shuffle node sets 2485 and/or where shuffle node sets 2485 are mutually exclusive. For example, data block exchange for facilitating query execution can be localized within a particular shuffle node set 2485, where nodes of a particular shuffle node set 2485 only send and receive data from other nodes in the same shuffle node set 2485, and where nodes in different shuffle node sets 2485 do not communicate directly and/or do not exchange data blocks at all. In some cases, where the inner level includes exactly one shuffle network, all nodes 37 in the inner level can and/or must exchange data blocks with all other nodes in the further level via the shuffle node set via a single corresponding shuffle network 2480.
Alternatively, some or all of the different shuffle networks 2480 can be interconnected, where nodes can and/or must communicate with other nodes in different shuffle node sets 2485 via connectivity between their respective different shuffle networks 2480 to facilitate query execution. As a particular example, in cases where two shuffle node sets 2485 have at least one overlapping node 37, the interconnectivity can be facilitated by the at least one overlapping node 37, for example, where this overlapping node 37 serves as a relay node to relay communications from at least one first node in a first shuffle node sets 2485 to at least one second node in a second first shuffle node set 2485. In some cases, all nodes 37 in a shuffle node set 2485 can communicate with any other node in the same shuffle node set 2485 via a direct link enabled via shuffle network 2480 and/or by otherwise not necessitating any intermediate relay nodes. However, these nodes may still require one or more relay nodes, such as nodes included in multiple shuffle node sets 2485, to communicate with nodes in other shuffle node sets 2485, where communication is facilitated across multiple shuffle node sets 2485 via direct communication links between nodes within each shuffle node set 2485.
Note that these relay nodes facilitating data blocks for execution of a given query across multiple shuffle node sets 2485 can be nodes participating in the query execution plan of the given query and/or can be nodes that are not participating in the query execution plan of the given query. In some cases, these relay nodes facilitating data blocks for execution of a given query across multiple shuffle node sets 2485 are strictly nodes participating in the query execution plan of the given query. In some cases, these relay nodes facilitating data blocks for execution of a given query across multiple shuffle node sets 2485 are strictly nodes that are not participating in the query execution plan of the given query.
In some cases a node 37 has direct communication links with its child node and/or parent node, where no relay nodes are required to facilitate sending data to parent and/or child nodes of the query execution plan 2405 of
For example a user types or otherwise indicates a query for execution via interaction with a computing device associated with and/or communicating with an external requesting entity. The computing device generates and transmits a corresponding query request 2515 for execution via the database system 10, where the corresponding query resultant 2526 is transmitted back to the computing device, for example, for storage by the computing device and/or for display to the corresponding user via a display device.
As illustrated in
In some cases, the operator flow generator module 2514 implements au optimizer to select the query operator execution flow 2517 based on determining the query operator execution flow 2517 is a most efficient and/or otherwise most optimal one of a set of query operator execution flow options and/or that arranges the operators in the query operator execution Dow 2517 such That the query operator execution flow 2517 compares favorably to a predetermined efficiency threshold. For example, the operator flow generator module 2514 selects and/or arranges the plurality of operators of the query operator execution flow 2517.10 implement the query expression in accordance with performing optimizer functionality, for example, by perform a deterministic function upon the query expression to select and/or arrange the plurality of operators in accordance with the optimizer functionality. This can be based on known and/or estimated processing times of different types of operators. This can be based on known and/or estimated levels of record filtering that will be applied by particular filtering parameters of the query. This can be based on selecting and/or deterministically utilizing a conjunctive normal form and/or a disjunctive normal form to build the query operator execution Row 2517 from the query expression. This can be based on selecting a determining a first possible serial ordering of a plurality of operators to implement the query expression based on determining the first possible serial ordering of the plurality of operators is known to be or expected to be more efficient than at least one second possible serial ordering of the same or different plurality of operators that implements the query expression. This can be based on ordering a first operator before a second operator in the Query operator execution flow 2517 based on determining executing the first operator before the second operator results in more efficient execution than executing the second operator before the first operator. For example, the first operator is known to filter the set of records open which the second operator would be performed to improve the efficiency of performing the second operator due to being executed upon a smaller set of records than if performed before the first operator. This can be based on other optimizer functionality that otherwise selects and/or arranges the plurality of operators of the query operator execution flow 2517 based on other known, estimated, and/or otherwise determined criteria.
A query execution module 2504 of the query processing system 2510 can execute the query expression via execution of the query operator execution flow 2517 to generate a query resultant. For example, the query execution module 2504 can be implemented via a plurality of nodes 37 that execute the query operator execution flow 2517. In particular, the plurality of nodes 37 of a query execution plan 2405 of
The query execution module 2504 can execute the determined query operator execution flow 2517 by performing a plurality of operator executions of operators 2520 of the query operator execution flow 2517 in a corresponding plurality of sequential operator execution steps. Each operator execution step of the plurality of sequential operator execution steps can correspond to execution of a particular operator 2520 of a plurality of operators 2520-1-2520-M of a query operator execution flow 2433.
In some embodiments, a single node 37 executes the query operator execution flow 2517 as illustrated in
A single operator execution by the query execution module 2504, such as via a particular node 37 executing its own query operator execution flows 2433, by executing one of the plurality of operators of the query operator execution flow 2433. As used herein, an operator execution corresponds to executing one operator 2520 of the query operator execution flow 2433 on one or more pending data blocks 2537 in an operator input data set 2522 of the operator 2520. The operator input data set 2522 of a particular operator 2520 includes data blocks that were outputted by execution of one or more other operators 2520 that are immediately below the particular operator in a serial ordering of the plurality of operators of the query operator execution flow 2433. In particular, the pending data blocks 2537 in the operator input data set 2522 were outputted by the one or more other operators 2520 that are immediately below the particular operator via one or more corresponding operator executions of one or more previous operator execution steps in the plurality of sequential operator execution steps. Pending data blocks 2537 of an operator input data set 2522 can be ordered, for example as an ordered queue, based on an ordering in which the pending data blocks 2537 are received by the operator input data set 2522. Alternatively, an operator input data set 2522 is implemented as an unordered set of pending data blocks 2537.
If the particular operator 2520 is executed for a given one of the plurality of sequential operator execution steps, some or all of the pending data blocks 2537 in this particular operator 2520's operator input data set 2522 are processed by the particular operator 2520 via execution of the operator to generate one or more output data blocks. For example, the input data blocks can indicate a plurality of rows, and the operation can be a SELECT operator indicating a simple predicate. The output data blocks can include only proper subset of the plurality of rows that meet the condition specified by the simple predicate.
Once a particular operator 2520 has performed an execution upon a given data block 2537 to generate one or more output data blocks, this data block is removed from the operator's operator input data set 2522. In some cases, an operator selected for execution is automatically executed upon all pending data blocks 2537 in its operator input data set 2522 for the corresponding operator execution step. In this case, an operator input data set 2522 of a particular operator 2520 is therefore empty immediately after the particular operator 2520 is executed. The data blocks outputted by the executed data block are appended to an operator input data set 2522 of an immediately next operator 2520 in the serial ordering of the plurality of operators of the query operator execution flow 2433, where this immediately next operator 2520 will be executed upon its data blocks once selected for execution in a subsequent one of the plurality of sequential operator execution steps.
Operator 2520. 1 can correspond to a bottom-most operator 2520 in the serial ordering of the plurality of operators 2520. 12520.M. As depicted in
Note that in the plurality of sequential operator execution steps utilized to execute a particular query, some or all operators will be executed multiple times, in multiple corresponding ones of the plurality of sequential operator execution steps. In particular, each of the multiple times a particular operator 2520 is executed, this operator is executed on set of pending data blocks 2537 that are currently in their operator input data set 2522, where different ones of the multiple executions correspond to execution of the particular operator upon different sets of data blocks that are currently in their operator queue at corresponding different times.
As a result of this mechanism of processing data blocks via operator executions performed over time, at a given time during the query's execution by the node 37, at least one of the plurality of operators 2520 has an operator input data set 2522 that includes at least one data block 2537. At this given time, one more other ones of the plurality of operators 2520 can have input data sets 2522 that are empty. For example, a given operator's operator input data set 2522 can be empty as a result of one or more immediately prior operators 2520 in the serial ordering not having been executed yet, and/or as a result of the one or more immediately prior operators 2520 not having been executed since a most recent execution of the given operator.
Some types of operators 2520, such as JOIN operators or aggregating operators such as SUM. AVERAGE, MAXIMUM. or MINIMUM operators, require knowledge of the full set of rows that will be received as output from previous operators to correctly generate their output. As used herein, such operators 2520 that must be performed on a particular number of data blocks. such as all data blocks that will be outputted by one or more immediately prior operators in the serial ordering of operators in the query operator execution Dow 2517.10 execute the query, are denoted as “blocking operators” Blocking operators are only executed in one of the plurality of sequential execution steps if their corresponding operator queue includes all of the required data blocks to be executed. For example, some or all blocking operators can be executed only if all prior operators in the serial ordering of the plurality of operators in the query operator execution flow 2433 have had all of their necessary executions completed for execution of the query, where none of these prior operators will be further executed in accordance with executing the query.
Some operator output generated via execution of an operator 2520, alternatively or in addition to being added to the input data set 2522 of a next sequential operator in the sequential ordering of the plurality of operators of the query operator execution Dow 2433, can be sent to one or more other nodes 37 in a same shuffle node set as input data blocks to be added to the input data set 2522 of one or more of their respective operators 2520. In particular, the output generated via a node's execution of an operator 2520 that is serially before the last operator 2520.M of the node's query operator execution flow 2433 can be sent to one or more other nodes 37 in a same shuffle node set as input data blocks to be added to the input data set 2522 of a respective operators 2520 that is serially after the last operator 2520. 1 of the query operator execution flow 2433 of the one or more other nodes 37.
As a particular example, the node 37 and the one or more other nodes 37 in a shuffle node set all execute queries in accordance with the same, common query operator execution flow 2433, for example, based on being assigned to a same inner level 2414 of the query execution plan 2405. The output generated via a node's execution of a particular operator 2520.i this common query operator execution flow 2433 can be sent to the one or more other nodes 37 in a same shuffle node set as input data blocks to be added to the input data set 2522 the next operator 2520 (+1, with respect to the serialized ordering of the query of this column query operator execution flow 2433 of the one or more other nodes 37. For example, the output generated via a node's execution of a particular operator 2520.i is added input data set 2522 the next operator 2520.j+1 of the same node's query operator execution flow 2433 based on being serially next in the sequential ordering and/or is alternatively or additionally added to the input data set 2522 of the next operator 2520, i+1 of the common query operator execution flow 2433 of the one or more other nodes in a same shuffle node set based on being serially next in the sequential ordering.
In some cases, in addition to a particular node sending this output generated via a node's execution of a particular operator 2520.1 to one or more other nodes to be input data set 2522 the next operator 2520.i+1 in the common query operator execution flow 2433 of the one or more other nodes 37, the particular node also receives output generated via some or all of these one or more other nodes' execution of this particular operator 2520.i in their own query operator execution flow 2433 upon their own corresponding input data set 2522 for this particular operator. The particular node adds this received output of execution of operator 2520.i by the one or more other nodes to the be input data set 2522 of its own next operator 2520.1+1.
This mechanism of sharing data can be utilized to implement operators that require knowledge of all records of a particular table and/or of a particular set of records that may go beyond the input records retrieved by children or other descendants of the corresponding node. For example, JOIN operators can be implemented in this fashion, where the operator 2520.i+1 corresponds to and/or is utilized to implement JOIN operator and/or a custom-join operator of the query operator execution flow 2517, and where the operator 2520, i+1 thus utilizes input received from many different nodes in the shuffle node set in accordance with their performing of all of the operators serially before operator 2520,i+1 to generate the input to operator 2520.i+1.
As used herein, a child operator of a given operator corresponds to an operator immediately before the given operator serially in a corresponding query operator execution flow and/or an operator from which the given operator receives input data blocks for processing in generating its own output data blocks. A given operator can have a single child operator or multiple child operators. A given operator optionally has no child operators based on being an IO operator and/or otherwise being a bottommost and/or first operator in the corresponding serialized ordering of the query operator execution flow. A child operator can implement any operator 2520 described herein.
A given operator and one or more of the given operator's child operators can be executed by a same node 37 of a given node 37. Alternatively or in addition, one or more child operators can be executed by one or more different nodes 37 from a given node 37 executing the given operator, such as a child node of the given node in a corresponding query execution plan that is participating in a level below the given node in the query execution plan.
As used herein, a parent operator of a given operator corresponds to an operator immediately after the given operator serially in a corresponding query operator execution flow, and/or an operator from which the given operator receives input data blocks for processing in generating its own output data blocks A given operator can have a single parent operator or multiple parent operators. A given operator optionally has no parent operators based on being a topmost and/or final operator in the corresponding serialized ordering of the query operator execution flow. If a first operator is a child operator of a second operator. the second operator is thus a parent operator of the first operator. A parent operator can implement any operator 2520 described herein.
A given operator and one or more of the given operator's parent operators can be executed by a same node 37 of a given node 37. Alternatively or in addition, one or more parent operators can be executed by one or more different nodes 37 from a given node 37 executing the given operator, such as a parent node of the given node in a corresponding query execution plan that is participating in a level above the given node in the query execution plan.
As used herein a lateral network operator of a given operator corresponds to an operator parallel with the given operator in a corresponding query operator execution flow. The set of lateral operators can optionally communicate data blocks with each other, for example, in addition to sending data to parent operators and/or receiving data from child operators. For example a set of lateral operators are implemented as one or more broadcast operators of a broadcast operation, and/or one or more shuffle operators of a shuffle operation. For example, a set of lateral operators are implemented via corresponding plurality of parallel processes 2550, for example, of a join process or other operation, to facilitate transfer of data such as right input rows received for processing between these operators. As another example, data is optionally transferred between lateral network operators via a corresponding shuffle and/or broadcast operation, for example, to communicate right input rows of a right input row set of a join operation to ensure all operators have a full set of right input rows.
A given operator and one or more lateral network operators lateral with the given operator can be executed by a same node 37 of a given node 37. Alternatively or in addition, one or lateral network operators can be executed by one or more different nodes 37 from a given node 37 executing the given operator lateral with the one or more lateral network operators. For example. different lateral network operators are executed via different nodes 37 in a same shuffle node set 37.
Each node 37 can utilize a corresponding query processing module 2435 to perform a plurality of operator executions for operators of the query operator execution flow 2433 as discussed in conjunction with
In some embodiments a given node 37 can optionally execute one or more operators, for example, when participating in a corresponding query execution plan 2405 for a given query, by implementing some of all features and/or functionality of the operator execution module 3215, for example, by implementing its operator processing module 2435 to execute one or more operator execution modules 3215 for one or more operators 2520 being processed by the given node 37. For example a plurality of nodes of a query execution plan 2405 for a given query execute their operators based on implementing corresponding query processing modules 2435 accordingly.
A given database table 2712 can be stored based on being received for storage, for example, via the parallelized ingress sub-system 24 and/or via other data ingress. Alternatively or in addition, a given database table 2712 can be generated and/or modified by the database system 10 itself based on being generated as output of a query executed by query execution module 2504. such as a Create Table. As Select (CTAS) query or Insert query.
A given database table 2712 can be in accordance with a schema 2409 defining columns of the database table, where records 2422 correspond to rows having values 2708 for some or all of these columns. Different database tables can have different numbers of columns and/or different datatypes for values stored in different columns. For example, the set of columns 2707. IA-2707.CA of schema 2709.A for database table 2712. A can have a different number of columns than and/or can have different datatypes for some or all columns of the set of columns 2707. IB-2707. CB of schema 2709.B for database table 2712.B. The schema 2409 for a given n database table 2712 can denote same or different datatypes for some or all of its set of columns. For example, some columns are variable-length and other columns are fixed-length. As another example, some columns are integers. other columns are binary values, other columns are Strings, and/or other columns are char types.
Row reads performed during query execution, such as row reads performed at the IO level of a query exception plan 2405, can be performed by reading values 2708 for one or more specified columns 2707 of the given query for some or all rows of one or more specified database tables, as denoted by the query expression defining the query to be performed. Filtering, join operations, and/or values included in the query resultant can be further dictated by operations to be performed upon the read values 2708 of these one or more specified columns 2707.
As illustrated in
As illustrated in
Values 2918 of a given row utilized in query execution are thus dispersed across different A given column 2915 can be implemented as a column 2707 having corresponding values 2918 implemented as values 2708 read from database table 2712 read from database storage 2450, for example, via execution of corresponding IO operators. Alternatively or in addition a given column 2915 can be implemented as a column 2707 having new and/or modified values generated during query execution, for example. via execution of an extend expression and/or other operation. Alternatively or in addition, a given column 2915 can be implemented as a new column generated during query execution having new values generated accordingly, for example, via execution of an extend expression and/or other operation. The set of column data streams 2968 generated and/or emitted between operators in query execution can correspond to some or all columns of one or more tables 2712 and/or new columns of an existing table and/or of a new tab to generated during query execution.
Additional column streams emitted by the given operator exception module can have their respective values for the same full set of output rows across for other respective columns. For example, the values across all column streams are in accordance with a consistent ordering, where a first row's values 2918.1.1-2918. 1.C for columns 2915.1-2915.C are included first in every respective column data stream, where a second row's values 2918.2.1-2918.2.C for columns 2915.1-2915.C are included second in every respective column data stream, and so on. In other embodiments, rows are optionally ordered differently in different column streams. Rows can be identified across column streams based on consistent ordering of values, based on being mapped 10 and/or indicating row identifiers, or other means.
As a particular example, for every fixed-length column, a huge block can be allocated to initialize a fixed length column stream, which can be implemented via mutable memory as a mutable memory column stream, and/or for every variable-length: column, another huge block can be allocated to initialize a binary stream, which can be implemented via mutable memory as a mutable memory binary stream. A given column data stream 2968 can be continuously appended with fixed length values to data runs of contiguous memory and/or may grow the underlying huge page memory region to acquire more contiguous runs and/or fragments of memory.
In other embodiments, rather than emitting data blocks with values 2918 for different columns in different column streams, values 2918 for a set of multiple column can be emitted in a same multi-column data stream.
A given operator execution module 3215.A for an operator that is a child operator of the operator executed by operator execution module 3215.B can emit its output data blocks for processing by operator execution module 3215.B based on writing each of a stream of data blocks 2537. 1-2537.K of data stream 2917. A to contiguous or non-contiguous memory fragments 2622 at one or more corresponding memory locations 2951 of query execution memory resources 3045.
Operator execution module 3215. A can generate these data blocks 2537.1-2537.K, of data stream 2917.A in conjunction with execution of the respective operator on incoming data. This incoming data can correspond to one or more other streams of data blocks 2537 of another data stream 2917 accessed in memory resources 3045 based on being written by one or more child operator execution modules corresponding to child operators of the operator executed by operator execution module 3215.A. Alternatively or in addition, the incoming data is read from database storage 2450 and/or is read from one or more segments stored on memory drives, for example, based on the operator executed by operator execution module 3215.A being implemented as au IO operator.
The parent operator execution module 3215.B of operator execution module 3215.A can generate its own output data blocks 2537.1-2537.J of data stream 2917.B based on execution of the respective operator upon data blocks 2537.1-2537.K of data stream 2917.A. Executing the operator can include reading the values from and/or performing operations to filter, aggregate. manipulate, generate new column values from, and/or otherwise determine values that are written to data blocks 25371-2537.J.
In other embodiments, the operator execution module 3215.B does not read the values from these data blocks, and instead forwards these data blocks, for example, where data blocks 2537.1-2537.J include memory reference data for the data blocks 2537.1-2537.K to enable one or more parent operator modules, such as operator execution module 3215.C, to access and read the values from forwarded streams.
In the case where operator execution module 3215.A has multiple parents, the data blocks 2537.1-2537.K of data stream 2917. A can be read, forwarded, and/or otherwise processed by each parent operator execution module 3215 independently it a same or similar fashion. Alternatively or in addition, in the case where operator execution module 3215 B has multiple children. each child's emitted set of data blocks 2537 of a respective data stream 2917 can be read, forwarded, and/or otherwise processed by operator execution module 3215.B in a same or similar fashion.
The parent operator execution module 3215.C of operator execution module 3215.B can similarly read, forward, and/or otherwise process data blocks 2537.1-2537.J of data stream 2917.B based on execution of the respective operator to render generation and emitting of its own data blocks in a similar fashion. Executing the operator can include reading the values from and/or performing operations to filter, aggregate, manipulate. generate new column values from, and/or otherwise process data blocks 2537.1-2537.J to determine values that are written to its own output data. For example, the operator execution module 3215.C reads data blocks 2537.1-2537.K of data stream 2917.A and/or the operator execution module 3215.B writes data blocks 2537.1-2537.J of data stream 2917.B. As another example, the operator execution module 3215.C reads data blocks 2537.1-2537.K of data stream 2917.A, or data blocks of another descendent, based on having been forwarded, where corresponding memory reference information denoting the location of these data blocks is read and processed from the received data blocks data blocks 2537.1-2537.J of data stream 2917.B enable accessing the values from: data blocks 2537.1-2537.K of data stream 2917.A. As another example, the operator execution module 3215.B does not read the values from these data blocks, and instead forwards these data blocks, for example, where data blocks 2537.1-2537.J include memory reference data for the data blocks 2537.1-2537.J to enable one or more parent operator modules to read these forwarded streams.
This pattern of reading and/or processing input data blocks from one or more children for use in generating output data blocks for one or more parents can continue until ultimately a final operator, such as an operator executed by a root level node. generates a query resultant, which can itself be stored as data blocks in this fashion in query execution memory resources and/or can be transmitted to a requesting entity for display and/or storage.
A plurality of records 2422.1-2422.Z, of one or more datasets 2505 to be converted into segments can be processed to generate a corresponding plurality of segments 2424.1-2424.Y. Each segment can include a plurality of column slabs 2610.1-2610.C corresponding to some or all of the C columns of the set of records.
In some embodiments, the dataset: 2505 can correspond to a given database table 2712. In some embodiments, the dataset: 2505 can correspond to only portion of a given database table 2712 (e.g., the most recently received set of records of a stream of records received for the table over time), where other datasets 2505 are later processed to generate new segments as more records are received over time. In some embodiments, the dataset 2505 can correspond to multiple database tables. The dataset 2505 optionally includes non-relational records and/or any records/files/data that is received from/generated by a given data source multiple different data sources.
Each record 2422 of the incoming dataset 2505 can be assigned to be included in exactly one segment 2424. In this example, segment 2424.1 includes at least records 24223 and 2422.7, while segment 2424 includes at least records 2422.1 and 2422.9. All of the Z, records can be guaranteed to be included in exactly one segment by segment generator 2507. Rows are optionally grouped into segments based on a cluster-key based grouping or other grouping by same or similar column values of one or more columns. Alternatively, rows are optionally grouped randomly, in accordance with a round robin fashion or by any other means.
A given row 2422 can thus have all of its column values 2708.1-2708. C included in exactly one given segment 2424. where these column values are dispersed across different column slabs 2610 based on which columns each column value corresponds. This division of column values into different column slabs can implement the column-format of segments described herein. The generation of column slabs can optionally include further processing of each set of column values assigned to each column slab. For example, some or all column slabs are optionally compressed and stored as compressed column slabs.
The database storage 2450 can thus store one or more datasets as segments 2424, for example, where these segments 2424 are accessed during query execution to identify/read values of rows of interest as specified in query predicates, where these identified rows/the respective values are further filtered processed/etc., for example, via operators 2520 of a corresponding query operator execution flow 2517, or otherwise accordance with the query to render generation of the query resultant.
The segment generator 2507 can implement a cluster key-based grouping module 2620 to group records of a dataset 2505 by a predetermined cluster key 2607, which can correspond to one or more columns. The cluster key can be received, accessed in memory, configured via user input, automatically selected based on an optimization, or otherwise determined. This grouping by cluster key can render generation of a plurality of record groups 2625. 1-2625.X.
The segment generator 2507 can implement a columnar rotation module 2630 to generate a plurality of column formatted record data (e.g., column slabs 2610 to be included in respective segments 2424). Each record group 2625 can have a corresponding set of J column-formatted record data 2565.1-2565.J generated, for example, corresponding to J segments in a given segment group.
A metadata generator module 2640 can further generate parity data, index data, statistical data, and/or other metadata to be included in segments in conjunction with the column-formatted record data. A set of X segment groups corresponding to the X record groups can be generated and stored in database storage 2450. For example, each segment group includes I segments, where parity data of a proper subset of segments in the segment group can be utilized to rebuild column-formatted record data of other segments in the same segment group as discussed previously.
In some embodiments, the segment generator 2507 implements some or all features and/or functionality of the segment generator 2517 as disclosed by: U.S. Utility Application Ser. No. 16/985,723, entitled “DELAYING SEGMENT GENERATION IN DATABASE SYSTEMS”. filed Aug. 5, 2020, which is hereby incorporated herein by reference in its entirety and made part of the present U.S. Utility Patent Application for all purposes, U.S. Utility Application Ser. No. 16/985,957 entitled “PARALLELIZED SEGMENT GENERATION VIA KEY-BASED SUBDIVISION IN DATABASE SYSTEMS”, filed Aug. 5, 2020, which is hereby incorporated herein by reference in its entirety and made part of the present U.S. Utility Patent Application for all purposes: and/or U.S. Utility Application Ser. No. 16/985,930, entitled “RECORD DEDUPLICATION IN DATABASE SYSTEMS”. filed Aug. 5, 2020. issued as U.S. Pat. No. 11,321,288 on May 3, 2022, which is hereby incorporated herein by reference in its entirety and made part of the present U.S. Utility Patent Application for all purposes. For example, the database system 10 implements some or all features and/or functionality of record processing and storage system 2505 of U.S. Utility application Ser. No. 16/985,723, U.S. Utility application Ser. No. 16/985,957, and/or U.S. Utility application Ser. No. 16/985,930.
Each IO pipeline 2835 can be generated based on corresponding segment configuration data 2833 for the corresponding segment 2424, such as secondary indexing data for the segment, statistical data/cardinality data for the segment, compression schemes applied to the columns slabs of the segment, or other information denoting how the segment is configured. For example. different segments 2424 have different IO pipelines 2835 generated for a given query based on having different secondary indexing schemes, different statistical data/cardinality data for its values, different compression schemes applied for some of all of the columns of its records, or other differences.
An IO operator execution module 2840 can execute each respective IO pipeline 2835. For example, the IO operator execution module 2840 is implemented by nodes 37 at the IO level of a corresponding query execution plan 2405, where a node 37 storing a given segment 2424 is responsible for accessing the segment as described previously, and thus executes the IO pipeline for the given segment.
This execution of IO pipelines 2835 by IO operator execution module 2840 correspond to executing IO operators 2421 of a query operator execution flow 2517. The output of IO operators 2421 can correspond to output of IO operators 2421 and/or output of IO level. This output can correspond to data blocks that are further processed via additional operators 2520, for example. by nodes at inner levels and/or the root level of a corresponding query execution plan.
Each IO pipeline 2835 can be generated based on pushing some or all filtering down to the IO level, where query predicates are applied via the IO pipeline based on accessing index structures, sourcing values, filtering rows, etc. Each IO pipeline 2835 can be generated to render semantically equivalent application of query predicates, despite differences in how the IO pipeline is arranged/executed for the given segment. For example, an index structure of a first segment is used to identify a set of rows meeting a condition for a corresponding column in a first corresponding IO pipeline while a second segment has its row values sourced and compared to a value to identify which rows meet the condition for example, based on the first segment having the corresponding column indexed and the second segment not having the corresponding column indexed. As another example, the IO pipeline for a first segment applies a compressed column slab processing element to identify where rows are stored in a compressed column slab and to further facilitate decompression of the rows, while a second segment accesses this column slab directly for the corresponding column based on this column being compressed in the first segment and being uncompressed for the second segment.
In some embodiments, the IO pipeline generator module 2834. IO pipeline 2835, and/or IO operator execution module 2840 implements some or all features and/or functionality of the IO pipeline generator module 2834, IO pipeline 2835, and/or JO operator execution module 2840 as disclosed by: U.S. Utility Application Ser. No. 17/303,437, entitled “QUERY EXECUTION UTILIZING PROBABILISTIC INDEXING”, filed May 28, 2021, which is hereby incorporated herein by reference in its entirety and made pan of the present U.S. Utility Patent Application for all purposes. For example, the database system 10 can implement the indexing of segments 2424 and/or IO pipeline generation as execution for accessing segments 2424 during query execution via implementing some or all features and/or functionality as described in U.S. Utility Application Ser. No. 17/303,437.
A given dataset 2505 can have a set of columns 2507.1-2507.C that correspond to various data types. The set of data types of the set of columns 25071-2507 C of one or more datasets 2505 processed and stored by database system 10 include: one or more fixed-length data types (e.g., integers, chars, Boolean values, etc.); one or more variable-length data types (e g, strings files, media data, etc.): one or more array data types, and/or one or more tuple data types. Note that a given tuple data type and/or array data type can be fixed length and/or variable length, for example, based on whether the respective elements within the array or tuple correspond to fixed-length and/or variable length data types and/or whether the respective number of elements is fixed or variable. One or more columns 2507 can optionally be implemented via a same data type (e.g., data set 2505 has multiple integer columns and/or multiple array columns).
A given array column 2712 can include army structures 2818 as its values 2708, where each array structure 2818 includes a plurality of array elements 2709.1-2709.M. Different array structures of different array columns of the same or different dataset can have different numbers of elements M and/or can have different data types as its array elements 2709.
In some embodiments, for a given array column, the array structures 2718 can optionally be required to have a same, fixed number of elements M (e.g., all rows have array structures in array column 2712.A having exactly 8 elements, and have array structures in a different array column 2712. B having exactly IO elements). Alternatively, the array structures 2718 of a given array column do not have this requirement, where different array structures 2718 of the same array column can have different numbers of elements.
In some embodiments, for a given array column, the array structures 2718 can optionally be required to have all elements 2709 having a same, predetermined data type (e.g., all rows have array structures in array column 2712.A with array elements 2709 that are all integers, and have army structures in array column 2712.B with array elements 2709 that are all strings). The predetermined data type can be required to be fixed length and/or can optionally be variable length. Alternatively, the array structures 2718 of a given array column do not have this requirement, where different array elements 2709 of the same array structure can have different data types (e.g. array element 2709. I.A.1 is an integer, and array element 2709. I.A. 2 is a char).
A given tuple column 2713 can include tuple structures 2819 as its values 2708, where each tuple structure 2819 includes a plurality of tuple elements 2739. Different tuple elements 2739 can correspond to different data types. In some embodiments. particular tuple elements of the tuple structures 2819 are assigned particular data types and/or a structured arrangement/number of such tuple elements is fixed for the given tuple column 2713. For example, the first tuple element of every tuple structure 2719 in the column is an integer, and the last tuple element of every tuple structure 2719 in the column is a string. Alternatively, different tuple structures 2719 in the same column can have different configurations and/or have elements of different data types.
In some embodiments, the tuple structures 2819 of a given tuple column 2713 can further include tuple elements 2739 that are implemented as container data types 5037 (e.g., array structures 2718, nested tuple structures 2719, etc.) that themselves include multiple elements 2739, considered sub-elements of the table structure 2719 implementing the values 2708 of the column. The sub-elements of a given element can be of the same or different data type. One or more sub-elements can itself be a further nested container data types 5037 containing its own set of multiple elements of the same or different data type.
The tuple structures of a given tuple column can be required to have a same number of, same data types of, and/or same arrangement of elements/subs-elements 2739. Different tuple structures of different tuple columns of the same or different dataset can have different numbers of, different data types of, and/or different arrangements of elements/subs-elements 2739.
In some embodiments, array column 2712 and/or tuple column 2713 of a dataset 2505 can implement some or all features and/or functionality of the array field 2712 of dataset 2705 as disclosed by: U.S. Utility Application Ser. No. 17/932,727, entitled “UTILIZING ARRAY FIELD DISTRIBUTION DATA IN DATABASE SYSTEMS”, filed Sep. 16, 2022, which is hereby incorporated herein by reference in its entirety and made part of the present U.S. Utility Patent Application for all purposes.
In some embodiments, for fixed-length data, block-level delta-delta compression or other compression, can be implemented by database system 10 in generating segments 2424. Such compression of fixed-length data can be effective for certain data types and patterns, but can be less effective for others. In some embodiments, for variable-length data, per-value compression is or other compression, can be implemented in generating segments 2424. Such compression of variable-length data can be effective for very large values, but can be less effective for smaller values.
These different secondary indexing schemes are then utilized to efficiently accessing the records included in corresponding different segments in conjunction with performing query executions. For example, in order to support various index types, query predicates can be pushed down into the IO operator, where the operator guarantees to return all records that match the predicates it is given, regardless of whether it does a full table scan-and-filter or whether it is able to take advantage of deterministic or probabilistic indexes internally.
This can be advantageous in cases where, as large volumes of incoming data for a given dataset are received over long periods of time, the distribution of the data is not necessarily fixed or known at the onset of storing the corresponding rows and/or is not necessarily constant over time. Rather than applying a same secondary indexing scheme for all segments storing a table/set of rows, secondary indexes can be determined on a segment-by-segment basis, for example, based on changes in data distribution over time that causes different segments to have different local data distributions of values in their respective records. Supporting heterogeneous segments in this manner provides the flexibility needed in long-lived systems. This improves the technology of database systems by enabling improved IO efficiency for each individual segment, where data distribution changes over time are handled via selection of appropriate indexes for different groupings of data received over time.
As illustrated in
The segment generator module 2506 can implement a row data clustering module 2507 to identify and segregate the dataset 2502 into different groups for inclusion in different segment groups and/or individual segments. Note that the segment generator module 2506 can implement a row data clustering module 2507 for generating segments from multiple different datasets with different types of records, records from different data sources, and/or records with different columns and/or schemas, where the records of different datasets are identified and segregated into different segment groups and/or individual segments, where different segments can be generated to include records from different datasets.
The row data clustering module 2507 can be implemented via one or more computing devices 18 and/or via other processing and/or memory resources of the database system 10. The row data clustering module can be implemented to generate segments from rows of records in a same or similar fashion discussed in conjunction with some or all of
Applying the segment generator module 2506 can include selecting and/or generating, for each segment being generated, segment row data 2505 that includes a subset of records 2422 of dataset 2502. Segment row data 2505 can be generated to include the subset of records 2422 of a corresponding segment in a column-based format. The segment row data 2505 can optionally be generated to include panty data such as parity data 2426, where the segment row data 2505 is generated for each segment in a same segment group of multiple segments by applying a redundancy storage encoding scheme to the subset of records 2422 of segment row data 2505 selected for the segments in the segment group as discussed previously.
The segment generator module 2506 can further implement a segment indexing module 2510 that generates secondary indexing data 2545 for a given segment based on the segment row data 2505 of the given segment. The segment indexing module 2510 can optionally further generate indexing data corresponding to cluster keys and/or primary indexes of the segment row data 2505 of the given segment.
The segment indexing module 2510 can generate secondary indexing data 2545 for a given segment as a plurality of secondary indexes that are included in the given segment 2424 and/or are otherwise stored in conjunction with the given segment 2424. For example, the plurality of secondary indexes of a segment's secondary indexing data 2545 can be stored in one or more index sections 0-x of the segment as illustrated in
The secondary indexing data 2545 of a given segment can include one or more sets of secondary indexes for one or more columns of the dataset 2502. The one or more columns of the secondary indexing data 2545 of a given segment can be different from a key column of the dataset 2502, can be different from a primary index of the segment, and/or can be different from the one or more columns of the clustering key utilized by the row data clustering module 2507 identify and segregate the dataset 2502 into different groups for inclusion in different segment groups and/or individual segments.
In some cases, the segment tow data 2505 is formatted in accordance with a column-based format for inclusion in the segment. In some cases, the segment 2424 is generated with a layout in accordance with the secondary indexing data 2545, for example, where the segment row data 2505 is optionally formatted based on and/or in accordance with secondary indexing type of the secondary indexing data 2545. Different segments 2424 with secondary indexing data 2545 in accordance with different secondary indexing types can therefore be generated to include their segment row data 2505 in accordance with different layouts and/or formats.
As segment row data 2505 and secondary indexing data 2545 is generated in conjunction with generating corresponding segments 2424 overtime from the dataset 2502, the segment row data 2505 and secondary indexing data 2545 are sent to a segment Storage system: 2508 for storage. The segment storage system 2508 can be implemented via one or more computing devices 18 of the database system and/or ether memory resources of the database system 10. For example, the segment storage system 2508 can include a plurality of memory drives 24425 of a plurality of nodes 37 of the database system 10. Alternatively or in addition, the segment storage system 2508 can be implemented via computing devices 18 of one or more storage clusters 35. The segment generator module 2506 can send its generated segments to the segment storage system 2508 via system communication resources 14 and/or via other communication resources.
A query execution module 2504 can perform query execution of various queries over time, for example, based on query requests received from and/or generated by client devices, based on configuration information, and/or based on user input. This can include performing queries against the dataset 2502 by performing row reads to the records 2422 of the dataset 2502 included in various segments 2424 stored by the segment storage system 2508. The query execution module 2504 can be implemented by utilizing the parallelized query and results subsystem 13 of
For example, the query execution module 2504 can perform query execution via a plurality of nodes 37 of a query execution plan 2405 as illustrated in
The secondary indexing data 2545 of various segments can be accessed during query executions to enable more efficient row reads of records 2422 included in the segment row data 2505 of the various segments 2424. For example, in performing the row reads at the IO level 2416, the query execution module 2504 can access and utilize the secondary indexing data 2545 of one or more segments being read for the query to facilitate more efficient retrieval of records from segment row data 2505. In some cases, the secondary indexing data 2545 of a given segment enables selection of and/or filtering of rows required for execution of a query in accordance with query predicates or other filtering parameters of the query.
The segment indexing module 2510 can implement a secondary indexing scheme selection module 2530. To further improve efficiency in accessing records 2422 of various segments 2424 in conjunction with execution of various queries, different segments can have their secondary indexing data 2545 generated in accordance with different secondary indexing schemes, where. The secondary indexing scheme is selected for a given segment to best improve and/or optimize the 10 efficiency for that given segment.
In particular, the secondary indexing scheme selection module 2530 is implemented to determine the existence, utilized columns, type, and/or parameters of secondary indexes on a per-segment basis rather than globally. When a segment 2424 is generated and/or written, the secondary indexing scheme selection module 2530 generates secondary indexing scheme selection data 2532 by selecting which index strategies to employ for that segment. The secondary indexing scheme selection data 2532 can correspond to selection of a utilized columns, type, and/or parameters of secondary indexes of the given segments from a discrete and/or continuous set of options indicated in secondary indexing scheme option data 2531.
The selection of each segment's secondary indexing scheme selection data 2532 can be based on the corresponding segment row data 2505, such as local distribution data determined for the corresponding segment row data 2505 as discussed in conjunction with
The secondary indexing scheme selection data 2532 can indicate index types and/or parameters selected for each column. In some embodiments, the secondary indexing scheme selection data 2532 can indicate a revision of the secondary indexing scheme selection module 2530 used to determine the secondary indexing scheme selection data 2532.
The secondary indexing scheme selection data 2532 of a given segment can be utilized to generate corresponding secondary indexing data 2545 for the corresponding segment row data 2505 of the given segment 2424. The secondary indexing data 2545 of each segment is thus generated accordance with the columns, index type, and/or parameters for selected for secondary indexing of the segment by the secondary indexing scheme selection module 2530.
Some or all of the secondary indexing scheme selection data 2532 can be stored as segment layout description data that is mapped to the respective segment. The segment layout description data for each segment can be extractible to identify the index types and/or parameters for each column indexed for the segment, and/or to determine which version of the secondary indexing scheme selection module 2530 was utilized to generate the corresponding secondary indexing scheme selection data 2532. For example, the segment layout description data is stored and/or is extractible in accordance with a JSON format.
The discrete and/or continuous set of options indicated in secondary indexing scheme option data 2531 can include a plurality of indexing ty pes 2532-1-2532-L. Each indexing type 2532-1-2532-L be applied to one column of the dataset 2502 and/or to a combination of multiple columns of the dataset 2502.
In some cases, the set of indexing types 2532-1-2532-L can include one or more secondary index types utilized in database systems. In some cases, the set of indexing types 2532-1-2532-L includes one or more of the following index types:
In some cases, the set of indexing types 2532-1-2532-L can include one or more probabilistic indexing tyes corresponding to a probabilistic indexing scheme discussed in conjunction with
This set of columns to which some or all of the plurality of indexing types 2532-1-2532-L can be selected for application can be indicated in the secondary indexing scheme option data 2531 as dataset schema data 2514, indicating the set of columns 2512-1-2512-C of the dataset 2502 and optionally indicating the datatype of each of the set of columns 2512-1-2512-C. Different datasets 2502 can have different dataset schema data 2514 based on having records that include different sets of data and/or types of data in accordance with different sets of columns.
One or more of the plurality of indexing types 2532-1-2532-L can be further configurable via one or more configurable parameters 2534. Different ones of the plurality of indexing types 2532-1-2532-L can have different sets of and/or numbers of configurable parameters 2534-1-2534-R, based on the parameters that are appropriate to the corresponding indexing type. In some cases, at least one of the configurable parameters 2534 can have its corresponding one or more values selected from a continuous set of values and/or options. In some cases, at least one of the configurable parameters 2534 can have its corresponding one or more values selected from a discrete set of values and/or options. Ranges, sets of valid options, and/or other constraints 10 the configurable parameters 2534 of some or all of the more of the plurality of indexing types 2533 can be indicated in the secondary indexing scheme option data 2531.
In some cases, at least one of the configurable parameters 2534 can correspond to a false-positive tuning parameter of a probabilistic indexing scheme as discussed in conjunction with
The secondary indexing scheme selection module 2530 can determine which columns of the set of columns 2512-1-2512-C will be indexed via secondary indexes for the segment row data 2505 of a given segment by selecting a set of selected columns 2513-1-2513-D as a subset of the set of columns set of columns 2512-1-2512-C. This can include selecting a proper subset of the set of columns 1-C. This can include selecting none of the columns I-C. This can include selection of all of the columns 1-C The selected columns 2513-1-2513-D for the given segment can be indicated in the resulting secondary indexing scheme selection data 2532. Different sets of selected columns 2513-1-2513-D and/or different numbers of selected columns 2513-1-2513-D can be selected by the secondary indexing scheme selection module 2530 for different segments.
The secondary indexing scheme selection module 2530 can further determine which one of more of the set of indexing types 2532-1-2532-L will be utilized for each selected column 2513-1-2513-D. In this example, selected indexing type 2533-1 is selected from the set of indexing types 2532-1-2532-L to index selected column 2513-1, and selected indexing type 2533-D is selected from the set of indexing types 2532-1-2532-L to index selected column 2513-D.
For a given column selected to be indexed a single index type can be selected for indexing the column, as illustrated in this example. In some cases, multiple different index types are optionally selected for indexing the column of a given segment. where a plurality of indexes are generated for the column for each of the multiple different index types.
For a given segment, different selected columns can have same or different ones of the set of indexing types 2532-1-2532-L selected. For example, for a given segment a first indexing type is selected for indexing a first column of the dataset, and a second indexing type is selected for indexing a second column of the dataset.
Different segments with the same set of selected columns 2513-1-2513-D can have the same or different ones of the set of indexing types 2532-1-2532-L selected for the same column. For example a particular column is selected to be indexed for both a first segment and a second segment. A first one of the set of indexing types 2532-1-2532-L is selected to index the particular column for the first segment, and a second one of the set of indexing types 2532-1-2532-L is selected to index the particular column for the second segment. As a particular example a bloom filter is selected to index the particular column for the first segment, and a b-tree is selected to index the given column for the second segment.
The secondary indexing scheme selection module 2530 can further configure the parameters of each selected indexing type 2533-1-2533-D. This can include selecting, for each selected indexing type 2533, a set of one or more selected parameters 2535-1-2535-R, where each selected parameter 2535 is a selected value and/or option for the corresponding configurable parameter 2534 of the corresponding indexing type 2533.
For a given segment, different selected columns can have same ones of the set of indexing ty pes 2532-1-2532-1, selected with the same or different selected parameters. For example, for a given segment, a particular indexing type is selected for indexing a first column of the dataset with a first set of selected parameters 2535-1-2535-R, and the same particular indexing type is selected for indexing a second column of the dataset with a second set of selected parameters 2535-1-2535-R with value that are different from the first set of selected parameters 2535-1-2535-R.
Different segments with the same set of selected indexing types 2533-1-2533-D for the same set of selected columns 2513-1-2513-D with the same or different selected parameters. For example a particular column is selected to be indexed for both a first segment and a second segment via a particular indexing type. A first set of selected parameters 2535-1-2535-R are selected for indexing the particular column via the particular indexing type for the first segment, and a different, second set of selected parameters 2535-1-2535-R are selected for indexing the particular column via the particular indexing type for the second segment.
In some cases, none of the parameters of a given selected indexing type 2533 are configurable, and no parameters values are selected for the given selected indexing type 2533. For example, this given selected indexing type 2533 is applied by the secondary index generator module 2540 to generate the plurality of indexes in accordance with predetermined parameters of the selected indexing type 2533.
As illustrated in
As a result, it can be normal for different segments of the same dataset 2502, such as a same database table, to have secondary index data in accordance with different columns of the dataset, different index types, and/or parameters. Furthermore. it can be advantageous for different segments of the same dataset 2502, such as a same database table, to have different secondary index data when these different segments have different local distribution data. In particular, the different secondary indexing scheme employed for different segments can be selected by the secondary indexing scheme selection module 2530 to leverage particular aspects of their respective local distribution data to improve IO efficiency during row reads.
The local distribution data for given segment row data 2505 can indicate the range, mean, variance, histogram data. probability density function data, and/or other distribution information for values of one or more columns in the set of records included in the given segment row data 2505. The local distribution data for given segment row data 2505 can indicate column cardinality, column range, and/or column distribution of one or more columns of the dataset for records 2422 included in the given segment row data 2505. The local distribution data for given segment row data 2505 can be optionally generated based on sampling only a subset of values included in records of the segment row data 2505, where the local distribution data is optionally probabilistic and/or statistical information. The local distribution data for given segment row data 2505 can be optionally generated based on sampling all values included in records of the segment row data 2505, where the local distribution data indicates the true distribution of the records in the segment. The local distribution data for given segment row data 2505 can optionally be generated as some or all of the statistics section of the corresponding segment, for example, as illustrated in
In some cases, the secondary indexing scheme selection module 2530 can generate the secondary indexing scheme selection data 2532 by performing one or more heuristic functions and/or optimizations. In particular, the selected columns. corresponding selected indexing types, and/or corresponding selected parameters can be selected for a given segment by performing the performing one or more heuristic functions and/or optimizations. The one or more heuristic functions and/or optimizations can generate the secondary indexing scheme selection data 2532 as functions of: the segment row data 2505 for the given segment: local distribution data 2542 determined for the segment row data 2505 for the given segment: user-generated secondary indexing hint data, system-generated secondary indexing hint data, and/or other information.
The one or more heuristic functions and/or optimizations can be configured via user input, can be received from a client device or other computing device, can be automatically generated, and/or can be otherwise determined. For example, a user or administrator can configure the more heuristic functions and/or optimizations via administrative sub-system 15 and/or configuration sub-system 16.
In cases where the one or more heuristic functions and/or optimizations are configured, the one or more heuristic functions and/or optimizations can optionally change over time, for example, based on new heuristic functions and/or optimization functions being introduced and/or based existing heuristic functions and/or optimization functions being modified. In such cases. newer segments generated from more recently received data of the dataset 2502 can have secondary indexing scheme selection data 2532 generated based on applying the more recently updated heuristic functions and/or optimization functions, while older segments generated from older received data of the dataset 2502 can have secondary indexing scheme selection data 2532 generated based on prior versions of heuristic functions and/or optimization functions. In some cases, one or more older segments can optionally be identified for re-indexing by applying the more recently updated heuristic functions and/or optimization functions to generate new secondary indexing scheme selection data 2532 for these older segments, for example, based on application of these more recently updated heuristic functions and/or optimization functions rendering secondary indexing scheme selection data 2532 with more efficient tow reads to these one or more older segments. Such embodiments are discussed in further detail it conjunction with:
The secondary index generator module 2540 can generate indexes for a given segment by indexing each selected column 2513 indicated in the secondary indexing scheme selection data 2532 for the given segment in accordance with the corresponding selected indexing type 2533 indicated in the secondary indexing scheme selection data 2532 for the given segment, and/or in accordance with the parameter selections 2535-1-2535-R indicated in the secondary indexing scheme selection data 2532 for the corresponding selected indexing type 2533. In this example, as D selected columns are indicated to be indexed via selected indexing types 2533-1-2533-D. D sets of secondary indexes 2546-1-2546-D are thus generated via the secondary index generator module. Each set of secondary indexes 2546 indexes the corresponding selected column 2513 via the corresponding selected indexing type 2533 in accordance with the corresponding parameter selections 2535-1-2535-R.
Some or all of the secondary indexing scheme option data 2531 can be configured via user input, can be received from a client device or other computing device, can be automatically generated, and/or can be otherwise determined. For example, a user or administrator can configure the secondary indexing scheme option data 2531 via administrative sub-system 15 and/or configuration sub-system 16.
In cases where the secondary indexing scheme option data 2531 is configured, the secondary indexing scheme option data 2531 can optionally change over time, for example, based on new indexing types being introduced and/or based on the query execution module 2504 being updated to enable access and use of to these new indexing types during row reads of query executions.
In such cases, newest segments generated from more recently received data of the dataset 2502 may have columns indexed via these newer indexing types based on these newer indexing types being available as valid options indicated in the secondary indexing scheme option data 2531 when these newer segments were indexed. Meanwhile, elder segments generated from older received data of the dataset 2502 may have columns indexed via these newer indexing types because they were not yet valid options of the secondary indexing scheme option data 2531 when these older segments were indexed. In some cases, one or more older segments can optionally be identified for re-indexing via these newer indexing types, for example, based on a newly available indexing type being more efficient for IO of these one or more older segments. Such embodiments are discussed in further detail in conjunction with
In some embodiments, the selection and use of various secondary indexing schemes for various segments can be communicated to end-users and/or administrators of the database system 10. For example, an interactive interface displayed on a display device of a client device communicating with the database system 10 can enable users to create a new table as a new dataset: 250 and/or add a column to an existing table: display and/or select whether that a secondary indexing scheme will improve performance for a given query profile: and/or add a new secondary indexing scheme as a new option in the secondary indexing scheme option data. In some cases, for a newly added secondary indexing scheme some or all future segments generated will include secondary indexes on the specified columns where appropriate: some or all future queries that can make use of this index will do so on the segments that contain the new secondary indexing scheme; the number of segments that contain this secondary indexing scheme can be displayed to the end-user. In some embodiments, secondary indexing schemes that are no longer needed can be dropped from consideration as options for future segments.
The segment generator module 2506, segment storage system 2508, and/or query execution module 2504 of
The generation of segments by the segment generator module cannot practically be performed by the human mind. particularly when the database system 10 is implemented to store and perform queries against records at a massive scale as discussed previously. In particular, the human mind is not equipped to perform segment generation and/or segment indexing for millions, billions, and/or trillions of records stored as many Terabytes, Petabytes, and/or Exabytes of data. Furthermore, the human mind is not equipped to distribute and perform segment indexing and/or segment generation as multiple independent processes. such as hundreds, thousands, and/or millions of independent processes, in parallel and/or within overlapping time spans.
The execution of queries by the query execution module cannot practically be performed by the human mind, particularly when the database system 10 is implemented to store and perform queries against records at a massive scale as discussed previously. In particular, the human mind is not equipped to read and/or process millions, billions, and/or trillions of records stored as many Terabytes, Petabytes, and/or Exabytes of records in conjunction with query execution. Furthermore, the human mind is not equipped to distribute and perform record reading and/or processing as multiple independent processes, such as hundreds. thousands, and/or millions of independent processes, in parallel and/or within overlapping time spans.
In various embodiments, a segment indexing module includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, cause the segment indexing module to select a first secondary indexing scheme for a first segment that includes a first plurality of rows from a plurality of secondary indexing options. A first plurality of secondary indexes for the first segment is generated in accordance with the first secondary indexing scheme. The first segment and the secondary indexes for the first segment are stored in memory. A second secondary indexing scheme is selected for a second segment that includes a second plurality of rows from the plurality of secondary indexing options, where the second secondary indexing scheme is different from the first secondary indexing scheme. A second plurality of secondary indexes for the second segment is generated in accordance with the second secondary indexing scheme. The second segment and the secondary indexes for the second segment are stored in memory.
Step 2582 includes generating a first segment that includes a first subset of a plurality of rows of a dataset. Step 2584 includes selecting a first secondary indexing scheme for the first segment from a plurality of secondary indexing options. Step 2586 includes generating a first plurality of secondary indexes for the first segment in accordance with the first secondary indexing scheme. Step 2588 includes storing the first segment and the secondary indexes for the first segment in memory.
Step 2590 includes generating a second segment that includes a second subset of the plurality of rows of the dataset. Step 2592 includes selecting a second secondary indexing scheme for the second segment from a plurality of secondary indexing options. Step 2594 includes generating a second plurality of secondary indexes for the second segment in accordance with the second secondary indexing scheme. Step 2596 includes storing the second segment and the secondary indexes for the second segment in memory. Step 2598 includes facilitating execution of a query against the dataset by utilizing the first plurality of secondary indexes to read at least one row from the first segment and utilizing the second plurality of secondary indexes to read at least one row from the second segment.
In various embodiments, the first segment and the second segment are generated by a segment generator module 2506. In particular, the first segment and the second segment can be generated by utilizing a row data clustering module 2507, and/or the first segment and the second segment are generated as discussed in conjunction with
In various embodiments, the method includes generating first local distribution information for the first segment, where the first secondary indexing scheme is selected for the first segment from a plurality of secondary indexing options based on the first local distribution information. The method can further include generating second local distribution information for the second segment, where the second secondary indexing scheme is selected for the second segment from a plurality of secondary indexing options based on the second local distribution information, and where the second secondary indexing scheme is different from the first secondary indexing scheme based on the second local distribution information being different from the first local distribution information.
In various embodiments, the plurality of secondary indexing options includes a set of secondary indexing options corresponding to different subsets of a set of columns of the database table. The first secondary indexing scheme can include indexing a first subset of the set of columns, the second secondary indexing scheme can include indexing a second subset of the set of columns, and a set difference between the first subset and the second subset can be non-null.
In various embodiments, the plurality of secondary indexing options includes a set of secondary indexing types that includes at least one of: a bloom filter, a projection index a data-backed index, a filtering index a composite index a zone map. a bit map, or a B-tree. The first secondary indexing scheme can include generating the first plurality of indexes in accordance with a first one of the set of secondary indexing types, and the secondary indexing scheme includes generating the second plurality of indexes in accordance with a second one of the set of secondary indexing types.
In various embodiments, the plurality of secondary indexing options includes a set of secondary indexing types. A first one of the secondary indexing types can include a first set of configurable parameters. Selecting the first secondary indexing scheme can include selecting the first one of the set of secondary indexing types and/or can include further selecting first parameter selections for each of the first set of configurable parameters for the first one of the set of secondary indexing types. Selecting the second secondary indexing scheme can include selecting the first one of the set of secondary indexing types and/or can include further selecting second parameter selections for each of the first set of configurable parameters for the first one of the set of secondary indexing types. The second parameter selection can be different from the first parameter selections.
In various embodiments, the first plurality of secondary indexes is different from a plurality of primary indexes of the first segment. The second plurality of secondary indexes can be different from a plurality of primary indexes of the second segment.
In various embodiments, the first segment is generated in a first temporal period, and the second segment is generated it a second temporal period that is after the first temporal period. After the first temporal period and prior to the second temporal period, the method can include updating the plurality of secondary indexing options to include a new secondary indexing option. The second secondary indexing scheme can be different from the first secondary indexing scheme based on the secondary indexing scheme being selected as the new secondary indexing option.
In various embodiments, selecting the first secondary indexing scheme for the first segment from the plurality of secondary indexing options can be based on first local distribution information corresponding to the first segment, user-provided hint data, and/or system-provided lint data. Selecting the second secondary indexing scheme for the second segment front the plurality of secondary indexing options can be based on second local distribution information corresponding to the second segment. user-provided hint data, and/or system-provided hint data.
In various embodiments, a non-transitory computer readable storage medium includes at least one memory section that Stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: generate a first segment that includes a first subset of a plurality of rows of a dataset: select a first secondary indexing scheme for the first segment from a plurality of secondary indexing options: generate a first plurality of secondary indexes for the first segment in accordance with the first secondary indexing scheme: store the first segment and the secondary indexes for the first segment in memory; generate a second segment that includes a second subset of the plurality of rows of the dataset; select a second secondary indexing scheme for the second segment from the plurality of secondary indexing options, where the second secondary indexing scheme is different from the first secondary indexing scheme: generate a second plurality of secondary indexes. For the second segment in accordance with the second secondary indexing scheme; store the second segment and the secondary indexes for the second segment in memory; and/or facilitate execution of a query against the dataset by utilizing the first plurality of secondary indexes to read at least one row from the first segment and utilizing the second plurality of secondary indexes to read at least one row from the second segment.
As discussed in conjunction with
Unlike the local distribution data 2542 which is determined for each segment individually, the user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630 can apply to the dataset 2502 as a whole. where same user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing him data 2630 is utilized by the secondary indexing scheme selection module 2530 to generate secondary indexing scheme selection data 2532 for many different segments with segment row data 2505 from the dataset: 2502.
In some cases, only user-generated secondary indexing hint data 2620 is determined and utilized by the secondary indexing scheme selection module 2530, where system-generated secondary indexing hint data 2630 is not utilized. In some cases. only system-generated secondary indexing hint data 2630 is determined and utilized by the secondary indexing scheme selection module 2530, where user-generated secondary indexing hint data 2620 is not utilized.
The user-generated secondary indexing hint data 2620 can be configured via user input, can be received from a client device or other computing device, and/or can be otherwise determined. As illustrated in
The client device 2601 can generate the user-generated secondary indexing hint data 2620 based on user input to an interactive interface 2650. The interactive interface can display one or more prompts for a user to enter the user-generated secondary indexing him data 2620 for the dataset 2502. For example, the interactive interface is displayed and/or the user-generated secondary indexing him data 2620 is generated by the client device 2601 in conjunction with execution of application data associated with the database system 10 that is received by the client device 2601 and/or stored in memory of the client device 2601 for execution by the client device 2601. As another example, the interactive interface is displayed in conjunction with a browser application associated with the database system 10 and accessed by the client device 2601 via a network.
The user-generated secondary indexing hint data 2620 can indicate information provided by the user regarding: known and/or predicted trends of the data in dataset 2502: known and/or predicted trends of the queries that will be performed upon the dataset: 2502; and/or other information that can be useful in selecting secondary indexing schemes for segments storing data of the dataset that will render efficient row reads during query executions In particular, user-generated secondary indexing hint data 2620 can indicate: “add-column-like” information and/or other information indicating an ordered or unordered list of columns that are known and/or expected to be commonly queried together, a known and/or expected probability value and/or relative likelihood for some or all columns to appear in a query predicate: a known and/or estimated probability value and/or relative likelihood for some or all columns to appear in one or more particular types of query predicates, such as equality-based predicates and/or range-based predicates: a known and/or estimated column cardinality of one or more columns; a known and/or estimated column distribution of one or more columns; a known and/or estimated numerical range of one or more columns; a known and/or estimated date or time-like behavior of one or more columns: and/or other information regarding the dataset 2502 and/or queries to be performed against the dataset 2502.
These user insights regarding the dataset 2502 and/or queries that will be performed against the dataset 2502 indicated in user-generated secondary indexing hint data 2620 can improve the performance of secondary indexing scheme selection module 2530 in generating secondary indexing scheme selection data 2532 that will render efficient row reads during query executions. These insights can be particular useful if the entirety of the dataset 2502 has not been received, for example, where the dataset 2502 is a stream of records that is received over a lengthy period of time, and thus distribution information for the dataset 2502 is unknown. This improves database systems by enabling intelligent selection of secondary indexing schemes based on user-provided distribution characteristics of the dataset when this information would otherwise be unknown.
These insights can also be useful in identifying which types of queries will be commonly performed and/or most important to end users, which further improves database systems by ensuring the selection of secondary indexing schemes for indexing of segments is relevant to the types of queries that will be performed. For example, this can help ensure that secondary indexing schemes that leverage these types of queries are selected for use to best improve IO efficiency based on the user-generated secondary indexing hint data 2620 indicating types of queries will be performed frequently. This helps ensure that other secondary indexing schemes that would rarely be useful in improving IO efficiency are thus not selected due to the user-generated secondary indexing hint data 2620 indicating types of query predicates that enable use of these secondary indexing schemes not being expected to be included in queries.
In some cases, the user-generated secondary indexing hint data 2620 does not include any selection of secondary indexing schemes to be utilized on some or all segments of the dataset 2502. In particular, the user-generated secondary indexing hint data 2620 can be implemented to serve as suggestions and/or added insight that can optionally be ignored by the secondary indexing scheme selection module 2530 in generating secondary indexing scheme selection data 2532. In particular, rather than enabling users to simply dictate which secondary indexing scheme will be used for a particular dataset based on their own insights. the user's insights are used as a tool to aid the secondary indexing scheme selection module 2530 in making intelligent selections.
Rather than relying solely on the secondary indexing scheme selection module 2530, the user-generated secondary indexing hint data 2620 can be configured to weigh the user-generated secondary indexing hint data 2620 in conjunction with other information, such as the local distribution information and/or the system-generated secondary indexing hint data 2630. For example a heuristic function and/or optimization is performed as a function of the user-generated secondary indexing hint data 2620, the local distribution information, and/or the system-generated secondary indexing him data 2630. This improves database systems by ensuring that inaccurate and/or misleading insights of user-generated secondary indexing haut data 2620 are not automatically applied in selecting secondary indexing schemes that would render sub-optimal IO efficiency. Furthermore, enabling users to simply dictate which secondary indexing scheme should be applied for a given dataset would render all segments of a given dataset having a same, user-specified index, and the added efficiency of per-segment indexing discussed previously would be lost.
Furthermore, in some cases, user-generated secondary indexing hint data 2620 can be ignored and/or can be de-weighted over time based on contradicting with local distribution data 2542 and/or system-generated secondary indexing him data 2630. In some cases, user-generated secondary indexing hint data 2620 can be removed entirely from consideration. In such embodiments. the user can be prompted via the interactive interface to enter new user-generated secondary indexing hint data 2620 and/or can be alerted that their user-generated secondary indexing hint data 2620 is inconsistent with local distribution data 2542 and/or system-generated secondary indexing hint data 2630.
The system-generated secondary indexing hint data 2630 can be generated automatically by an indexing hint generator system 2551, which can be implemented by the segment indexing module 2510, by one or more computing devices 18, and/or by other processing resources and/or memory resources of the database system: 10. Unlike the user-generated secondary indexing tint data 2620, the system-generated secondary indexing hint data 2630 can be generated without human intervention and/or the system-generated secondary indexing lint data 2630 is not based on user-supplied information. Instead, the system-generated secondary indexing hint data 2630 can be generated based on: current dataset information, such as distribution information for the portion of dataset 2502 that has been received and/or stored in segments 2424, historical query data, such as a log of queries that have been performed. queries that are performed frequently. queries flagged as having poor IO efficiency, and/or other information regarding previously performed queries: current and/or historical system health, memory, and/or performance information such as memory utilization of segments with various secondary indexing schemes and/or IO efficiency of segments with various indexing schemes: and/or other information generated by and/or tracked by database system 10.
As a particular example, the system-generated secondary indexing hint data 2630 can indicate current column cardinality. range, and or distribution of one or more columns. As another a particular example, the system-generated secondary indexing hint data 2630 can indicate “add-column-like” information and/or other information indicating an ordered or unordered list of columns that are commonly queued together, derived from some or all previous queries such as historically slow queries and/or column queries.
Different datasets 2502 can have different user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630. The same dataset 2502 can have different user-generated secondary indexing hint data 2620 configured by different users. The same dataset 2502 can have different secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630 generated over time, for example, where the user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630 optionally updated over time, and where segments are indexed by utilizing the most recent user-generated secondary indexing hint data 2620 and/or most recent system-generated secondary indexing hint data 2630.
In such cases, newer segments generated from more recently received data of the dataset 2502 can have secondary indexing scheme selection data 2532 generated based on applying more recently updated user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630, while older segments generated from older received data of the dataset 2502 can have secondary indexing scheme selection data 2532 generated based on prior versions of user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630. In some cases, one or more older segments can optionally be identified for re-indexing by applying the more recently updated user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630 to generate new secondary indexing scheme selection data 2532 for these older segments, for example, based on application of these user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630 rendering secondary indexing scheme selection data 2532 with more efficient row reads to these one or more older segments. Such embodiments are discussed in further detail in conjunction with
In some cases, newly generated and/or newly received user-generated secondary indexing hint data 2620 and/or system-generated secondary indexing hint data 2630 can be “tested” prior to being automatically utilized by the secondary indexing scheme selection module 2530 to determine whether they would render secondary indexing selections that induce favorable 10 efficiency and/or improved IO efficiency for currently stored segments. For example a user can elect to perform this test for their proposed user-generated secondary indexing hint data 2620 and/or the database system 10 can automatically perform this test prior to any reliance upon user-generated secondary indexing hint data 2620 in generating secondary indexes for new segments.
This testing can be performed by re-evaluating the secondary indexing schemes for one or more currently stored segments based on applying the proposed user-generated secondary indexing him data 2620 as input to the secondary indexing scheme selection module 2530 for an existing segment, determining if this would render a different secondary indexing scheme selection for the existing segment, testing the different secondary indexing scheme selection for the existing segment via que or more test queries to determine whether or not the IO efficiency for the segment would improve and/or be sufficiently efficient when this different secondary indexing scheme selection is applied; selecting to adopt the proposed user-generated secondary indexing hint data 2620 when at least a threshold number and/or percentage of existing segments have improved IO efficiency and/or have sufficient 10 efficiency with different secondary indexing scheme selections generated by applying the adopt the proposed user-generated secondary indexing hint data, and/or selecting to not adopt the proposed user-generated secondary indexing tint data 2620 when at least a threshold number and/or percentage of existing segments do have not improved JO efficiency and/or do not have sufficient IO efficiency with different secondary indexing scheme selections generated by applying the adopt the proposed user-generated secondary indexing hint data. Some or all of this process can optionally be performed by implementing the segment indexing evaluation system of
In various embodiments a segment indexing module includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, cause the segment indexing module to receive a user-generated secondary indexing hint data for a dataset front a client device. The client device generated the user-generated hint data based on user input in response to at least one prompt displayed by an interactive interface displayed via a display device of the client device. A plurality of segments each include distinct subsets of a plurality of rows of a database table for each of the plurality of segments a secondary indexing scheme is automatically selected from a plurality of secondary indexing options based on the user-provided secondary indexing hint data. A plurality of secondary indexes is generated for each of the plurality of segments in accordance with the corresponding secondary indexing scheme. The plurality of segments and the plurality of secondary indexes are stored in memory.
Step 2682 includes receiving a user-generated secondary indexing hint data for a dataset from a client device. Step 2684 includes generating a plurality of segments that each include distinct subsets of a plurality of rows of a dataset. Step 2686 includes automatically selecting, for each of the plurality of segments a secondary indexing scheme from a plurality of secondary indexing options based on the user-provided secondary indexing hint data. Step 2688 includes generating a plurality of secondary indexes for each of the plurality of segments in accordance with the corresponding secondary indexing scheme. Step 2690 includes storing the plurality of segments and the plurality of secondary indexes in memory.
In various embodiments, the user-generated secondary indexing hint data indicates query predicate trend data for future queries to be performed by at least one user against the dataset. In various embodiments, the query predicate trend data indicates an ordered list of columns commonly queried together and/or a relative likelihood for a column to appear in a predicate. In various embodiments, the user-generated secondary indexing hint data indicates estimated distribution data for a future plurality of rows of the dataset to be received by the database system for storage. In various embodiments, the estimated distribution data indicates an estimated column cardinality of the future plurality of rows of the dataset and/or an estimated column distribution of the future plurality of rows of the dataset.
In various embodiments, the method includes automatically generating, system-generated secondary indexing hint data for the dataset Automatically selecting the secondary indexing scheme is based on applying a heuristic function to the user-provided secondary indexing hint data and the system-generated secondary indexing hint data. In various embodiments, the system-generated secondary indexing hint data is generated based on accessing a log of previous queries performed upon the dataset, and/or generating statistical data for current column values of one or more columns of currently-stored rows of the dataset. In various embodiments the system-generated secondary indexing hint data indicates a current column cardinality, a current distribution of the data: a current column distribution; a current column range: and/or sets of columns commonly queried together for example, in historically slow queries, common queries, and/or across all queries.
In various embodiments, a heuristic function is further applied to local distribution data generated for each segment. In various embodiments, the method includes generating and/or determining the local distribution data for each segment.
In various embodiments, the method includes ignoring and/or removing at least some of the user-provided secondary indexing hint data based on the system-generated secondary indexing hint data contradicting the user-provided secondary indexing hint data. In various embodiments, the user-provided secondary indexing hint data does not include selection of a secondary indexing scheme to be applied to the plurality of segments. For example, different secondary indexing schemes are applied to different segments despite being selected based on the same user-provided secondary indexing hint data.
In various embodiments the method includes receiving updated user-provided secondary indexing hint data from the client device, for example after receiving the user-provided secondary indexing hint data. The secondary indexing scheme utilized for a more recently generated one of the plurality of segments is different from the secondary indexing scheme utilized for a less recently generated one of the plurality of segments based receiving the updated user-provided secondary indexing tint data after generating the first one of the plurality of segments and before generating the second of the plurality of segments.
In various embodiments, a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to receive a user-generated secondary indexing hint data for a dataset from a client device, where the client device generated the user-generated hint data based on user input in response to at least one prompt displayed by an interactive interface displayed via a display device of the client device: generate a plurality of segments that each include distinct subsets of a plurality of rows of a dataset: automatically select, for each of the plurality of segments, a secondary indexing scheme from a plurality of secondary indexing options based on the user-provided secondary indexing hint data, generate a plurality of secondary indexes for each of the plurality of segments in accordance with the corresponding secondary indexing scheme: and/or store the plurality of segments and the plurality of secondary indexes in memory.
Existing segments can be reindexed, for example, in order to take advantage of new hints, new index types, bug fixes. or updated heurisnes Reindexing can happen over time on a live system since segments for a dataset 2502 are heterogeneous. During reindexing, the secondary indexing scheme is evaluated for each segment to determine whether re-indexing would produce a different layout. For each segment group to be re-indexed, all existing segments in the group are read and new segments are created using the updated index layout. Once the new segments are written segment metadata is updated for future queries and the old segment group can be removed.
The segment indexing evaluation system 2710 can be implemented to evaluate index efficiency for particular segments to determine whether and/or how their secondary index structure should be changed. This can include identifying existing segments for re-indexing and identifying a new secondary indexing scheme for these existing segments that are determined and/or expected to be more efficient for IO efficiency of segments than their current secondary indexing scheme. The segment indexing evaluation system 2710 can be implemented to automatically re-index existing segments under a newly selected secondary indexing scheme determined for the existing segments. This improves the technology of database systems to enable the indexing schemes of particular segments to be altered to improve the 10 efficiency of these segments, which improves the efficiency of query executions.
This further improves the technology of database systems by enabling the per-segment indexing discussed previously to be adaptive to various changes over time. In particular, segments can be identified for reindexing and/or can be re-indexed via a new secondary indexing scheme based on: identifying segments with poor IO efficiency in one or more recently executed queries: changes in types of queries being performed against the dataset 2502, new types of secondary indexes that are supported as options in the secondary indexing scheme option data 2531: new heuristic functions and/or optimizations utilized by the secondary indexing scheme selection module 2530; receiving updated user-generated secondary indexing hint data 2620; automatically generating updated system-generated secondary hint data 2630; and/or other changes.
In this example a set of segments 1-R can be evaluated for re-indexing. For example, this evaluation is initiated based on a determination to evaluate the set of segments 1-R. This determination can be based on: a predetermined schedule and/or time period to re-evaluate indexing of the set of segments; identifying segments 1-R as having poor IO efficiency in one or more recently executed queries; changes in types of queries being performed against the dataset 2502: introducing new types of secondary indexes that are supported as options in the secondary indexing scheme option data 2531: introducing new heutistic functions and/or optimizations utilized by the secondary indexing scheme selection module 2530; receiving updated user-generated secondary indexing hint data 2620: automatically generating updated system-generated secondary hint data 2630; receiving a request and/or instruction to re-evaluate indexing of the set of segments; receiving a request from client device 2601 to evaluate how indexing of the set of segments would change in light of a newly supplied user-generated secondary indexing him data 2620, detected degradation in query efficiency, and/or another determination.
The set of segments 1-R can correspond to all segments in the database system and/or can correspond to all segments storing records of dataset 2502. The set of segments 1-R can alternatively correspond to a proper subset of segments in the database system and/or a proper subset of segments storing records of dataset 2502. This proper subset can be selected based on identifying segments as having poor IO efficiency in one or more recently executed queries. This proper subset can be selected based on identifying segments whose secondary indexing scheme was selected and generated before a predefined time and/or date. This proper subset can be selected based on identifying segments with segment layout indicating their secondary indexing scheme was selected in via a revision of the secondary indexing scheme selection module 2530 that is older than a current revision of the secondary indexing scheme selection module 2530 and/or a predetermined threshold revision of the secondary indexing scheme selection module 2530. This proper subset can be selected based on identifying segments whose secondary indexing scheme was selected based on: an version of the heuristic functions and/or optimizations utilized by the secondary indexing scheme selection module 2530 that is older than a current version of the heuristic functions and/or optimizations utilized by the secondary indexing scheme selection module 2530: a version of the user-generated secondary indexing hint data 2620 that is older than the current version of user-generated secondary indexing hint data 2620 utilized by the secondary indexing scheme selection module 2530; a version of the system-generated secondary indexing bitt data 2630 that is older than the current version of the user-generated secondary indexing hint data 2620 utilized by the secondary indexing scheme selection module 2530; an older version of the secondary indexing scheme option data 2531 that does not include at least one new secondary indexing type that is included in the current version of the secondary indexing scheme option data 2531 utilized by the secondary indexing scheme selection module.
The current secondary indexing scheme data 2731 of each of the set of segments 1-R can be determined based on accessing the segments 1-R in memory, based on accessing metadata of the segments 1-R, based on tracked information regarding the previous selection of their respective secondary indexing schemes, and/or another determination. The current secondary indexing scheme data 2731 of a given segment can indicate the secondary indexing scheme selection data 2532 that was utilized to generate the secondary index data 2545 of the segment when the segment was generated and/or in a most recent re-indexing of the segment; the secondary index data 2545 itself; information regarding the layout of the segment and/or format of the segment row data 2505 induced by the currently utilized secondary indexing scheme; and/or other information regarding the current secondary indexing schemes for the segment.
Secondary indexing efficiency metrics 2715-1-2715-R can be generated for the identified set of segments 2424-1-2424-R via an index efficiency metrics generator module 2722 based on their respective current secondary indexing scheme data 2731-1-2731-R. The index efficiency metrics generator module 2722 can perform one or more queries, such as a set of test queries. upon the dataset 2502 and/or upon individual ones of the set of segments to generate the secondary indexing efficiency metrics 2715-1-2715-R. The set of test queries can be predetermined, can be configured via user input, can be based on a log of common and/or recent queries, and/or can be based on previously performed queries with poor efficiency.
In some cases, secondary indexing efficiency metrics 2715 are automatically generated for segments as they are accessed in various query executions, and the index efficiency metric generator module 2722 can optionally utilize these tracked secondary indexing efficiency metrics 2715 by accessing a memory that in memory that stores the tracked secondary indexing efficiency metrics 2715 instead of or in addition to generating new secondary indexing efficiency metrics 2715-1-2715-R via execution of new queries.
In some embodiments, rather than running the set of test queries on the actual segments a set of virtual columns can be generated for the segments 2424-1-2424-R based on their current secondary indexing scheme data 2731-1-2731-R and the set of fest queries can be performed utilizing the virtual columns. This mechanism be ideal when the index efficiency metrics generator module 2722 is utilized to generate secondary indexing efficiency metrics 2715 for proposed secondary indexing schemes of these segments rather than their current secondary indexing schemes, as discussed in further detail in conjunction with
The secondary indexing efficiency metrics 2715 of a given segment can be based on raw metrics indicating individual values and/or blocks that are read, processed, and/or emitted. These raw metrics can be tracked in performance of the set of test queries to generate the secondary indexing efficiency metrics 2715.
A block that is read, processed, and/or emitted can include values of multiple records included in a given segment, where a given segment includes many blocks. For example, these blocks are implemented as the coding blocks within a segment discussed previously and/or are implemented as 4 Kilo-byte data blocks. These blocks can optionally be a fixed size, or can have variable sizes.
One of these raw metrics that can be tracked in performance of the set of test queries for a given segment can correspond to a “values read” metric. The “values read” metric can be tracked as a collection of value-identifiers for blocks and/or individual values included in the segment that were read from disk. In some cases, this metric has block-level granularity.
Another one of these raw metrics that can be tracked in performance of the set of test queries for a given segment cat correspond to a “values processed” metric. The “values processed” metric can be tracked as a collection of value identifiers for blocks and/or individual records included in the segment that were processed by the IO operator. This collection of value identifiers corresponding to values processed by the IO operator is always a subset of the collection of value identifiers that were read, and may be smaller when indexing allows decompression of specific rows in a block. In bytes, this metric may be larger than bytes read due to decompression. This metric can also have metric also have block-level granularity in cases where certain compression schemes that do not allow random access are utilized.
Another one of these raw metrics that can be tracked in performance of the set of test queries for a given segment can correspond to a “values emitted” metric. The “values emitted” metric can be tracked as a map of a collection of value-identifiers which satisfy all predicates and are emitted upstream. For example, this can include the number of blocks outputted as output data blocks of the IO operator and/or of one or more IO level nodes. The predicates can correspond to all query predicates that are pushed-down to one or more IO operators of the query that are executed in accordance with an IO pipeline as discussed in further detail in conjunction with
The raw metrics tracked for each given segment can be utilized to calculate one or more efficiency values of the secondary indexing efficiency metrics 2715. The secondary indexing efficiency metrics 2715 can include an IO efficiency value for the given segment. The IO efficiency value is computed with block granularity, and can be calculated as a proportion of blocks read that have an emitted value. For example, the IO efficiency value can be calculated by dividing the number of unique blocks with at least one emitted value indicated in the “values emitted” metric by the number of unique blocks read indicated in the “values read” metric. A perfect value of I means that every block that was read was needed to satisfy the plan IO efficiency values indicating higher proportions of values that are read also being emitted constitute better IO efficiency, and thus more favorable secondary indexing efficiency metrics 2715, than lower proportions of values that are read also being emitted.
The secondary indexing efficiency metrics 2715 can include an IO efficiency value for the given segment. The JO efficiency value can have a block granularity, and can be calculated as a proportion of blocks read that have an emitted value. For example, the IO efficiency value can be calculated by dividing the number of unique blocks with at least one emitted value indicated in the “values emitted” metric by the number of unique blocks read indicated in the “values read” metric. A perfect value of I means that every block that was read was needed to satisfy the plan. IO efficiency values indicating higher proportions of values that are read also being emitted constitute better IO efficiency, and thus more favorable secondary indexing efficiency metrics 2715, than IO efficiency values indicating lower proportions of values that are read also being emitted.
The secondary indexing efficiency metrics 2715 can include a processing efficiency value for the given segment. The processing efficiency value can have a byte granularity, and can be calculated as a proportion of bytes processed that are emitted as values. For example, the processing efficiency value can be calculated by dividing the sum of bytes emitted as indicated in the “values emitted” metric by the sum of bytes processed as indicated in the “values processed” metric. A perfect value of 1 means that every byte processed by the IO operator was needed to satisfy the plan Processing efficiency values indicating higher proportions of bytes that are processed also being emitted constitute better processing efficiency, and thus more favorable secondary indexing efficiency metrics 2715, than processing efficiency values indicating lower proportions of bytes that are processed also being emitted.
The inefficient segment identification module 2724 can identify a subset of the segments 1-R as inefficient segments. illustrated in
The secondary indexing scheme selection module 2530 can generate secondary indexing scheme selection data 2532 for each of the set of inefficient segments I-S. The secondary indexing scheme selection data 2532 for some or all of the inefficient segments 1-S can indicate a different secondary indexing scheme from their current different secondary indexing scheme.
The secondary indexing scheme selection module 2530 can be implemented in a same or similar fashion as discussed in conjunction with
As a particular example a segment layout description for each segment can be extracted for correlation with efficiency metrics. This layout description can indicate the index types and parameters chosen for each column, along with the revision of the secondary indexing scheme selection module 2530 used to determine that layout.
In some embodiments, the segment indexing evaluation system 2710 can facilitate display of the current secondary indexing scheme dam 2731 of inefficient segments I-S to a user, for example, via a display device of client device 2601. This can include displaying the current indexing strategy and/or other layout information for the inefficient segments. This can include displaying their secondary indexing efficiency metrics 2715 and/or some of all of the taw metrics tracked in performing the test queries.
In some cases, the secondary indexing scheme selection module 2530 can generate the indexing scheme selection data 2532 based on user interaction with an interactive interface, such as interactive interface 2650 of client device 2601 and/or another client device utilized by an administrator, developer, or different user, in response to reviewing some or all of this displayed information. This can include prompting the user to select whether to adopt the new secondary indexing schemes selected for these segments or to maintain their current secondary indexing schemes. In some embodiments, the user can be prompted to enter and/or select proposed user-generated secondary indexing hint data 2620 for these poor-performing segments based on the current indexing strategy and/or other layout information. In some cases, proposed hint data can be automatically determined and displayed. This proposed hint data can be generated based on automatically generating system-generated secondary indexing hint data 2630, for example, based on the current secondary indexing scheme data 2731 and/or their poor efficiency. This proposed hint data can be automatically populated with recent user-generated secondary indexing hint data 2620 and/ot system-generated secondary indexing hint data 2630 used to index newer segments, where these proposed hints that may be relevant to older segments as well.
In some embodiments, the secondary indexing scheme selection data 2532 for some or all of the inefficient segments 1-S is automatically utilized to generate respective secondary index data 2545 for inefficient segments 1-S via secondary index generator module 2540. This can include reformatting segment row data 2505 and/or otherwise changing the layout of the segment 2424 to accommodate the new secondary indexing scheme.
In other cases, the secondary indexing scheme selection data 2532 generated for some or all of the inefficient segments 1-S is considered a proposed secondary indexing scheme that undergoes evaluation prior to being adopted. The process discussed in conjunction with
The secondary indexing scheme selection data 2532 generated for some or all of the inefficient segments 1-S are processed via index efficiency metric generator module 2722 to generate secondary indexing efficiency metrics 2715 for the inefficient segments 1-S, indicating the level of efficiency that would be induced if the proposed secondary indexing scheme indicated in the secondary indexing scheme selection data 2532 were to be adopted. For example, virtual columns are determined for each segment 1-S in accordance with the proposed secondary indexing scheme, and these virtual columns are utilized to perform the set of test queries and generate the secondary indexing efficiency metrics 2715 indicating efficiency of the proposed secondary indexing scheme for each segment.
The inefficient segment identification module 2724 can be utilized to determine whether these proposed secondary indexing scheme are efficient or inefficient. This can include identifying a set of efficient segment based on these segments having favorable secondary indexing efficiency metrics 2715 for their proposed secondary indexing schemes. This can include identifying a set of inefficient segment based on these segments having unfavorable secondary indexing efficiency metrics 2715 for their proposed secondary indexing schemes, for example, based on comparison of the IO efficiency value and/or processing efficiency value to corresponding threshold values as discussed previously.
In some cases, determining whether a segment's secondary indexing efficiency metrics 2715 for their proposed secondary indexing schemes are favorable optionally includes comparing the secondary indexing efficiency metrics 2715 for the proposed secondary indexing scheme of the segment to the secondary indexing efficiency metrics 2715 for the current secondary indexing scheme. For example, a proposed secondary indexing schemes is only adopted for a corresponding segment if it has more favorable secondary indexing efficiency metrics 2715 than the secondary indexing efficiency metrics 2715 of the current secondary indexing scheme.
As proposed new indexing strategies render acceptable secondary indexing efficiency metrics for their corresponding segments, these segments can be re-indexed using their corresponding new indexing strategy. If the proposed new indexing strategies do not render acceptable secondary indexing efficiency metrics for their corresponding segments, the re-indexing attempt can be abandoned where their current indexing scheme is maintained, and/or additional iterations of this process can continue to evaluate additional proposed secondary indexing schemes for potential adoption in this fashion.
This is illustrated in
In some embodiments, with each iteration, a new, hypothetical segment layout description for an existing segment corresponding to the proposed secondary indexing scheme for the existing segment can be presented to the presented to the user via interactive interface 2650. The interactive interface 2650 can optionally prompt the user to add or remove user-generated secondary indexing hint data 2620 in order to see the results of potential changes on the segment layout, where the process can be re-performed with user-supplied changes to the user-generated secondary indexing hint data 2620. This functionality can be ideal in enabling end-users, developers, and/or administrators to evaluate the effectiveness of user-generated secondary indexing hint data 2620.
In some embodiments, this process is performed to identify poor or outdated user-generated secondary indexing hint data 2620 supplied by users that rendered selection of secondary indexing schemes that caused respective segments to have poor efficiency metrics In some cases, these poor hints are automatically removed from consideration in generating new segments and/or users are alerted that these hints are not effective via interactive interface 2650. In some cases, the heuristic functions and/or optimizations utilized by the secondary indexing scheme selection module 2530 are automatically updated over time to de-weight and/or adjust to the importance of user-provided hints relative to system-provided hints based on how effectively prior and/or current user-generated secondary indexing him data 2620 improved efficiency relative to system-generated secondary indexing hint data 2630.
In some cases, the index efficiency metric generator module 2722 and inefficient segment identification module 2724 are utilized to evaluate proposed secondary indexing scheme selections for all newly generated segments. For example, the process implemented by the segment indexing evaluation system 2710 of in
In some embodiments, space efficiency of index structures is alternatively or additionally evaluated. For example, a current index structure may induce efficient metrics for a given segment, but other index strategies with much cheaper storage requirements can be tested and determined to render favorable efficiency metrics. This can trigger re-indexing of segments to improve space efficiency without sacrificing IO efficiency or processing efficiency.
In such embodiments, instead of or in addition to identifying inefficient segments 1-S for re-indexing, the segment indexing evaluation system 2710 can optionally identify segments with unnecessarily complicated secondary indexing schemes and/or with secondary indexing schemes that require larger amounts of memory. In some cases, these segments can have their indexing schemes re-evaluated in a similar fashion to determine whether a less complicated and/or less memory intensive secondary indexing scheme could be utilized for the segment that would still yield favorable index efficiency metrics. The segment indexing evaluation system 2710 can identify such secondary indexing schemes for these and generate corresponding secondary index data 2545 for these segments accordingly.
In various embodiments a segment indexing evaluation system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, cause the segment indexing evaluation system to generate secondary index efficiency metrics for a set of secondary indexing schemes corresponding to a set of segments stored in the database system based upon performing at least one query that accesses row data included in the set of segments. A first segment of the set of segments is selected for reindexing based on the secondary index efficiency metrics for a first one of the set of secondary indexing schemes corresponding to the first segment A new set of secondary indexes are generated for the first segment based on applying a new secondary indexing scheme that is different from one of the set of secondary indexing schemes that corresponds to the first segment based on selecting the first segment for reindexing. The new set of secondary indexes are stored in conjunction with storage of the first segment. Execution of a query can be facilitated by utilizing the new set of secondary indexes to read at least one row from the first segment.
Step 2782 includes generating secondary index efficiency metrics for a set of secondary indexing schemes corresponding 10 a set of segments stored in the database system based upon performing at least one query that accesses row data included in the set of segments Step 2784 includes selecting a first segment of the set of segments for reindexing based on the secondary index efficiency metrics for a first one of the set of secondary indexing schemes corresponding to the first segment. Step 2786 includes generating a new set of secondary indexes for the first segment based on applying a new secondary indexing scheme that is different from one of the set of secondary indexing schemes that corresponds to the first segment based on selecting the first segment for reindexing. Step 2788 includes storing the new set of secondary indexes in conjunction with storage of the first segment. Step 2790 includes facilitating execution of a query by utilizing the new set of secondary indexes to read at least one row from the first segment.
In various embodiments, at least one of the set of secondary indexing schemes is currently utilized in query executions for access to rows of the corresponding one of a set of segments. In various embodiments, at least one of the set of secondary indexing schemes is a proposed indexing scheme for the corresponding one of a set of segments.
In various embodiments, the method includes selecting the new secondary indexing scheme as a proposed indexing scheme for the first segment based on selecting the first segment for reindexing, and/or generating secondary index efficiency metrics for the new secondary indexing scheme based on selecting the new secondary indexing scheme as the proposed indexing scheme for the first segment. Generating the new set of secondary indexes for the first segment is based on the secondary index efficiency metrics for the new secondary indexing scheme being more favorable than the secondary index efficiency metrics for the first one of the set of secondary indexing schemes.
In various embodiments, the method includes selecting a second segment of the set of segments for reindexing based on the secondary index efficiency metrics for a second one of the set of secondary indexing schemes corresponding to the second segment. The method can include selecting a second new secondary indexing scheme as a proposed indexing scheme for the second segment based on selecting the second segment for reindexing. The method can include generating secondary index efficiency metrics for the second new secondary indexing scheme based on selecting the second new secondary indexing scheme as the proposed indexing scheme for the second segment. The method can include selecting a third new secondary indexing scheme as another proposed indexing scheme for the second segment based on the secondary index efficiency metrics for the second new secondary indexing scheme comparing unfavorably to a secondary index efficiency threshold. The method can include generating secondary index efficiency metrics for the third new secondary indexing scheme based on selecting the third new secondary indexing scheme as the another proposed indexing scheme for the second segment. The method can include generating a new set of secondary indexes for the second segment by applying the third new secondary indexing scheme based on the secondary index efficiency metres for the third new secondary indexing scheme being more favorable than the secondary index efficiency metrics. For the second new secondary indexing scheme.
In various embodiments, the method includes selecting a subset of the set of segments for reindexing that includes the first segment based on identifying a corresponding subset the set of secondary indexing schemes with secondary index efficiency metrics that compare unfavorably to a secondary index efficiency threshold.
In various embodiments, the method includes selecting the at least one query based on receiving select query predicates generated via user input and/or based on identifying common query predicates in a log of historically performed queries and/or a recent query predicates in a log of historically performed queries.
In various embodiments, the index efficiency metrics include: an IO efficiency metric, calculated for each segment as a proportion of blocks read from the each segment that have an emitted value in execution of the at least one query; and/or a processing efficiency metric calculated for each segment as a proportion of bytes read from the each segment that are emitted as values in execution of the at least one query.
In various embodiments, the method includes facilitating display, via an interactive interface, of a prompt to enter user-generated secondary indexing hint data for secondary indexing of the first segment based on selecting the first segment for reindexing. User-generated secondary indexing lint data is received based on user input to the prompt. The new secondary indexing scheme for the first segment is selected based on the user-generated secondary indexing hint data.
In various embodiments, the method includes determining to generate the secondary index efficiency metrics for a set of secondary indexing schemes corresponding to a set of segments. This determination can be based on: detecting degradation in query efficiency; introduction of a new secondary index type that can be implemented in reindexed segments, where the new secondary indexing scheme is selected as the a new secondary index type: introduction of a new heuristic and/or optimization function for implementation in selecting new indexing strategies to re-index segments, where the new secondary indexing scheme is selected based on utilizing heuristic and/or optimization function; receiving new user-provided secondary indexing hint data and/or new user-provided secondary indexing hint data system-provided hint data, where the secondary index efficiency metrics are generated to evaluate whether applying this new hint data would improve efficiency of existing segments: and/or determining other information. The secondary index efficiency metrics can be generated based on determining to generate the secondary index efficiency metrics.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: generate secondary index efficiency metrics for a set of secondary indexing schemes corresponding to a set of segments stored in the database system based upon performing at least one query that accesses row data included in the set of segments: select a first segment of the set of segments for reindexing based on the secondary index efficiency metrics for a first one of the set of secondary indexing schemes corresponding to the first segment: generate a new set of secondary indexes for the first segment based on applying a new secondary indexing scheme that is different from one of the set of secondary indexing schemes that corresponds to the first segment based on selecting the first segment for reindexing; store the new set of secondary indexes in conjunction with storage of the first segment: and/or facilitate execution of a query by utilizing the new set of secondary indexes to read at least one row from the first segment.
The operator execution flow generator module 2803 can be implemented via one or more computing devices and/or via other processing resources and/or memory resources of the database system 10. The operator execution flow generator module 2803 can generate an operator execution flow 2817, indicating a flow of operators 2830 of the query to be performed by the query execution module 2504 to execute the query in accordance with a serial and/or parallelized ordering. Different portions of the operator execution flow 2817 can optionally be performed by nodes at different corresponding levels of the query execution plan 2405.
At the bottom of the operator execution flow 2817, one or more IO operators 2821 are included. These operators are performed first to read records required for execution of the query from corresponding segments. For example, the query execution module 2504 performs a query against dataset 2502 by accessing records of dataset 2502 in respective segments. As a particular example, nodes 37 at 10 level 2416 each perform the one or more IO operators 2821 to read records from their respective segments.
Rather than generating an operator execution flow 2817 that with IO operators 2821 that are executed in an identical fashion across all segments, for example, by applying index probing or other use of indexes to filter rows uniformly across all IO operators for all segments, the execution of IO operators must be adapted to account for different secondary indexing schemes that are utilized for different segments. To guarantee query correctness, all IO operators must be guaranteed to filter the correct set of records when performing record reads in the same fashion.
This can be accomplished by pushing all of the query predicates 2822 of the given query down to the IO operators. Executing the IO operators via query execution module 2504 includes applying the query predicates 2822 to filter records from segments accordingly 2424. In particular, performing the IO operators to perform rows reads for different segment requires that the IO operators are performed differently. For example index probing operations or other filtering via IO operators may be possible for automatically applying query predicates 2822 in performing row reads for segment indexed via a first secondary indexing scheme. However, this same IO process may not be possible for a second segment indexed via a different secondary indexing scheme. In this case, an identical filtering step would be required after reading the rows from the second segment.
The construction of an efficient IO pipeline for a given query and segment can be challenging. While a trivial scan-and-filter pipeline can satisfy many queries, most efficiency gains from building an IO pipeline that uses a combination of indexes. dependent sources, and filters to minimize unneeded IO. As a result, different elements must be used depending on the predicates involved, the indexes present in that segment, the presence or absence of variable-length skip lists, and the version of the cluster key index.
The query execution module 2504 can include an index scheme determination module 2832 that determines the secondary indexing scheme data 2833-1-2833-R indicating the secondary indexing scheme utilized for each of a set of segments 1-R to be accessed in execution of a given query. For example, the secondary indexing scheme data 2833-1-2833-R is mapped to the respective segments in memory accessible by the query execution module 2504, is received by the query execution module 2504, and/or is otherwise determined by the query execution module 2504. This can include extracting segment layout description data stored for each segment 1-R.
An IO pipeline generator module 2834 can select a set of IO pipelines 2835-1-2835-R for performance upon each segment 1-R to implement the IO operators of the operator execution flow 2817. In particular, each IO pipeline 2835 can be determined based on: the pushed to the IO operators in the operator execution flow 2817, and/or the secondary indexing scheme data 2833 for the corresponding segment. Different IO pipelines can be selected for different segments based on the different segments having different secondary indexing schemes.
An IO operator execution module 2840 can apply each IO pipeline 2835-1-2835-R to perform the IO operators of the operator execution flow 2817 for each corresponding segment 2424-1-2424-R. Performing a given IO pipeline can include accessing the corresponding segment in segment storage system 250810 read rows, utilizing the segment's secondary indexing scheme as appropriate and/or as indicated by the IO pipeline. Performing a given IO pipeline can optionally include performing additional filtering operators in accordance with a serial and/or parallelized ordering, for example, based on the corresponding segment not having a secondary indexing scheme that corresponds to corresponding predicates. Performing a given IO pipeline can include ultimately generating a filtered record set emitted by the given IO pipeline 2835 as output. The output of one or more IO operators 2821 as a whole, when applied to all segments 1-R, corresponds to the union of the filtered record sets generated by applying each IO pipeline 2835-1-2835-R to their respective segment. This output can be input to one or more other operators 2830 of the operator execution flow 2817, such as one or more aggregations and/or join operators applied the read and filtered records.
In some embodiments a given node 37 implements its own index scheme determination module 2832, its own IO pipeline generator module 2834, and/or its own IO operator execution module 2840 to perform IO operations upon its own set of segments 1-R for example, each of a plurality of nodes 37 participating at the IO level 2416 of a corresponding query execution plan 2405 generates and executes IO pipelines 2835 for its own subset of a plurality of segments required for execution of the query, such as the ones of the plurality of segments stored in its memory drives 2425.
In some embodiments, the IO pipeline for a given segment is selected and/or optimized based on one or more criteria. For example, the serialized ordering of a plurality of columns to be sources via a plurality of corresponding IO operators is based on distribution information for the column, such as probability distribution function (PDF) data for the columns, for example. based on selecting columns expected to filter the greatest number of columns to be read and filtered via IO operators either in the serialized ordering than IO operators for other columns. As another example, the serialized ordering of a plurality of columns to be sources via a plurality of corresponding IO operators is based on the types of secondary indexes applied to each column, where columns with more efficient secondary indexes and/or secondary indexing schemes that are more applicable to the set of query predicates 2822 are selected to be read and filtered via IO operators earlier in the serialized ordering than IO operators for other columns. As another example, index efficiency metrics and/or query efficiency metrics can be measured and tracked overtime for various query executions, where IO pipelines with favorable past efficiency and/or performance for a given segment and/or for types of secondary indexes are selected over other IO pipelines with less favorable past efficiency and/or performance.
As illustrated in
Each of the plurality of IO operators can be executed to access values of records 2422 in accordance with the query, and thus sourcing values of the segment as required for the query. Each of these IO operators 2821 can be denoted with a source, identifying which column of the dataset 2502 is to be accessed via this IO operator. In some cases a column group of multiple columns is optionally identified as the source for some IO operators, for example, when compound indexes are applied to this column group for the corresponding segment.
Each of these index source IO operators 2821, when executed for the given segment, can output a set of row numbers and/or corresponding values read from the corresponding segment. In particular. IO operators 2821 can utilize a set of row numbers to consider as input, which can be produced as output of one or more prier IO operators. The values produced by an IO operator can be decompressed in order to be evaluated as part of one or more predicates.
Depending on the type of index employed and/or the placement in the IO pipeline 2835, some IO operators 2821 may emit only row numbers, some IO operators 2821 may emit only data values, and/or some IO operators 282) may emit both row and data values. Depending on the type of index employed, a source element can be followed by a filter that filters rows from a larger list emitted by the source element based on query predicates.
Some or all of the plurality of IO operators 2821 of the IO pipeline 2835 of a given segment can correspond to index sources that utilize primary indexes, cluster key indexes and/or secondary indexes of the corresponding segment to filter ones of the tow numbers and/or corresponding values in their respective output when reading from the corresponding segment. These index source IO operators 2821 can further be denoted with an index type, identifying which type of indexing scheme is utilized for access to this source based on the type of indexing scheme was selected and applied to the corresponding column of the corresponding segment, and/or a predicate, which can be a portion of query predicates 2822 applicable to the corresponding source column to be applied when performing the IO upon the segment by utilizing the indexes.
These IO operators 2821 can utilize the denoted predicate as input for internal optimization. This filter predicate can be pushed down into each corresponding index, allowing them to implement optimizations. For example, bitmap indexes only need to examine the columns for a specific range or values.
These index source IO operators 2821 output only a subset of set of row numbers and/or corresponding value identified to meet the criteria of corresponding predicates based on utilizing the corresponding index type of the corresponding source for the corresponding segment. In this example, the IO operators 2821 sourcing colA, colB, and cold are each index source IO operators 2821.
Some or all of the plurality of IO operators 2821 of the IO pipeline 2835 of a given segment can correspond to table data sources. These table data source IO operators 2821 can be applied to columns without an appropriate index and/or can be applied to columns that are not mentioned in query predicates 2822. In this example, the IO operators 2821 sourcing colD is a table data source, based on colD not being mentioned in query predicates 2822. These table data source IO operators can perform a table scan to produce values for a given column. When upstream in the IO pipeline, these table data source IO operators 2821 can skip rows not included in their input list of rows received as output of a prior IO operator when performing the table scan. Some or all these IO operators 2821 can produce values for the cluster key for certain rows, for example, when only secondary indexes are utilized.
Some or all of the plurality of IO operators 2821 of the IO pipeline 2835 of a given segment can correspond to default value sources. These default source IO operators 2821 can always output a default value for a given source column when this column is not present in the corresponding segment.
The various index source table data source, and default IO operators 2821 included in a given IO pipeline can correspond to various type of pipeline elements that can be included as elements of the IO pipeline. These types can include:
The IO pipeline 2835 can further include filtering operators 2823 that filter values outputted by sources serially before these filters based on portions of the query predicates 2822. The filtering operators 2823 can serve as a type of pipeline element that evaluates a predicate expression on each incoming row. filtering rows that do not pass. In some embodiments, every column in the provided predicate must be sourced by other pipeline elements downstream of this pipeline element. In particular, these filtering operators 2823 can be required for some segments that do not have secondary indexes for one or more columns indicated at the query predicates 2822, where the column values of all rows of such columns are first read via a table data source IO operator 2821, and where one or more corresponding filtering operators 2823 are applied to filter the rows accordingly In some embodiments, the IO pipeline 2835 can further include logical operators such as AND and/or OR operators as necessary for the corresponding query predicates 2822.
In some embodiments, all possible secondary indexing schemes of the secondary indexing scheme option data 2531 that can be implemented in segments for use in query execution are required to receive a list of predicates to evaluate as input, and return a list of rows that pass those predicates as output, where execution of an index source IO operator includes utilizing the corresponding predicates of the of index source IO operator to evaluate return a list of rows that pass those predicates as output. These tow lists can be filtered and/or merged together in the IO pipeline as different indexes are used for the same query via different IO operators. Once the final row list is calculated, columns that are required for the query, but do not yet have values generated by the pipeline, can be read off disk.
In some embodiments. variable length columns are stored as variable-length quantity (VLQ) prefixed regions in row order. For example. VLQs and tow data can span across 4 Kilo-byte blocks. Seeking to a given row number can include starting at the first row and cursing through all of the data. Information on a per-LCK basis that enables seeking to the first byte in a variable length column for that key can be stored and utilized. However, in segments with high clustering this can be a large portion of the column span. In order to enable efficient row value lookups by row number for variable length columns, a row offset lookup structure for variable length columns can be included. These can be similar to the fixed length lookup structures used in decompression, but with extra variable-length specific information.
For example, a skip list can be built for every column. For variable length columns, the skip list can encode an extra byte offset of first row, and can be in accordance with a different structure than that of fixed length columns. new skip list structure can be required. Performing IO can include loading skip lists for variable length columns in the query into memory. Given a row number, the first entry that has a larger first row number can be identified. The previous entry in the skip list can be accessed, and one or more blocks that contain the value can be read. In some cases, the subsequent block must always be read based on the end location of the row being unknown. In some cases, every variable length column read can include reads to two 4 Kilo-byte blocks. In some cases, each 4 Kilo-byte data block of segment row data 2505 can be generated to include block delta encoded row offsets and/or a byte offset of first row.
In some embodiments, for queries that use secondary indexes and require cluster key column emission but don't actually require to search the cluster key index, look up of cluster key values by row number can be implemented via the addition of row numbers in the primary cluster key index. This can include adding row ranges to index partition information in index headers and/or Adding row off′set in the index. When IO is performed, the index partition a row falls into can be determined, a binary search for a cluster key that contains can be performed, and/or the cluster key can be emitted.
In this example, this example IO pipeline 2835 for this set of example query predicates 2822 can be generated for a first given segment based on colC having a cluster key (CK) index for the first given segment: based on colA having a bitmap index for the first given segment; and/or based on colB having a data-backed index for the first given segment. For example, these index types for cola and colB are secondary index types that were selected via the secondary indexing scheme selection module 2530 when the segment was generated and/or evaluated for re-indexing as discussed previously. The respective secondary index data 2545 for col A and colB of this first given segment was generated by the secondary index generator module accordingly to include a bitmap index for colA and a data-backed index for colB. When this IO pipeline 2835 for the first segment is executed, the secondary index data 2545 the bitmap index for colA and a data-backed index for colB of the secondary index data 2545 is accessed to perform their respective IO operators 2821.
While not illustrated, consider a second segment upon which this same query is performed. A different IO pipeline 2835 for this set of example query predicates 2822 can be generated for the second given segment based on the second given segment having different secondary indexing schemes for colA and colB. For example, colA has a bloom fiber index and colB has not indexing. The IO operator 2821 sourcing colA in the IO pipeline 2835 for this second segment can thus be generated with an index type of a bloom filter, and/or can similarly the (colA <=3 OR colA>5) predicates. IO operator 2821 sourcing colA in the IO pipeline 2835 for this second segment can be a table data source IO operator based on colB having no secondary indexes in the second segment. A separate filtering operator 2823 can be applied serially after the table data source IO operator sourcing colB to apply the respective (colB<=10) predicate In particular, this separate filtering operator 2823 can filter the outputted values received from the table data source IO operator for colB by selecting only the values that are less than or equal to 10.
IO operators 2821 and/or filtering operators 2823 further along the pipeline that are serially after prior IO operators 2821 and/or filtering operators 2823 in a serialized ordering of the IO pipeline can utilize output of prior IO operators 2821 and/or filtering operators 2823 as input. In particular. IO operators that receive row numbers from prior ones IO operators in the serial ordering can perform their reads by only accessing rows with the corresponding row numbers outputted by a prior IO operator.
Each pipeline element (e g. IO operators, filter operators, and/or logical operators) of an IO pipeline can either to union of intersect its incoming row lists from prior pipeline elements in the IO pipeline. In some embodiments, an efficient semi-sparse row list representation can be utilized for fast sparse operations. In some embodiments, pipeline can be optimized to cache derived values (such as filtered row lists) to avoid re-computing them in subsequent pulls.
In this example, the IO operator 2821 sourcing colC outputs a first subset of row numbers of a plurality of row numbers of the segment based on identifying only rows with colC′ values greater than or equal to 1, based on utilizing the cluster key index. For colC. The IO operator 2821 sourcing colA receives this first subset of the plurality of rows numbers outputted by the IO operator 2821 sourcing colC, and only access rows with row numbers in the first subset. The first subset is further filtered into a second subset of the first subset by identifying rows with row numbers in the first subset with colA values that are either less than or equal to 3 of are greater than 5, based on utilizing the bitmap index for colA.
Similarly, the IO operator 2821 sourcing colB receives the first subset of the plurality of row numbers outputted by the IO operator 2821 sourcing colC, and also only access rows with row numbers in the first subset. The first subset is filtered into a third subset of the first subset by identifying rows with tow numbers in the first subset with colB values that are either less than or equal to 10, based on utilizing the data-backed index for colB. The IO operator 2821 sourcing colB can be performed in parallel with the IO operator 2821 sourcing colA because neither IO operators is dependent on the other's output.
The union of the second subset and third subset are further filtered based on the filtering operators 2823 and logical operators to satisfy the required conditions of the query predicates 2822, where a final set of row numbers utilized as input to the final IO operator sourcing colD includes only the row numbers with values in colA, colB, and colC′ that satisfy the query predicates 2822. This final set of row numbers is thus utilized by the final IO operator sourcing colD to produce the values emitted for the corresponding segment, where this IO operator reads values of colD for only the row numbers indicated in its input set of row numbers.
The query processing system: 2802 of
The execution of queries by the query execution module cannot practically be performed by the human mind, particularly when the database system 10 is implemented to store and perform queries against records at a massive scale as discussed previously. In particular, the human mind is not equipped to perform IO pipeline generation and/or processing for millions billions. and/or trillions of records stored as many Terabytes. Petabytes, and/or Exabytes of data. Furthermore, the human mind is not equipped to distribute and perform IO pipeline generation and/or processing as multiple independent processes, such as hundreds. thousands, and/or millions of independent processes, in parallel and/or within overlapping time spans.
In various embodiments, a query processing system includes at least one processor; and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor: cause the query processing system to identify a plurality of predicates of a query for execution. A query operator flow for is generated a query by including the plurality of predicates in a plurality of IO operators of the query operator flow. Execution of the query is facilitated by, for each given segment of a set of segments stored in memory: generating an IO pipeline for each given segment based on a secondary indexing scheme of a set of secondary indexes of the each segment and based on plurality of predicates, and performing the plurality of IO operators open each given segment by applying the IO pipeline to the each segment.
Step 2882 includes identifying a plurality of predicates of a query for execution. Step 2884 includes generating a query operator flow for a query by including the plurality of predicates in a plurality of IO operators of the query operator flow. Step 2886 includes facilitating execution of the query to read a set of rows from a set of segments stored in memory.
Performing step 2886 can include performing steps 2888 and/or 2890 for each given segment of the set of segments. Step 2888 includes generating an IO pipeline for each given segment based on a secondary indexing scheme of a set of secondary indexes of the given segment, and based on the plurality of predicates. Step 2890 includes performing the plurality of IO operators upon the given segment by applying the IO pipeline to the given segment.
In various embodiments, the set of segments are stored in conjunction with different ones of a plurality of corresponding secondary indexing schemes. In various embodiments, a first IO pipeline is generated for a first segment of the set of segments. and a second IO pipeline is generated for a second segment of the set of segments. The first IO pipeline is different from the second IO pipeline based on the set of secondary indexes of the first segment being in accordance with a different secondary indexing scheme than the set of secondary indexes of the second segment.
In various embodiments, performing the plurality of IO operators upon at least one segment of the set of segments includes utilizing the set of secondary indexes of the at least one segment in accordance with the IO pipeline to read at least one row from the at least one segment. In various embodiments, performing the plurality of IO operators upon at least one segment of the set of segments includes filtering at least one row from inclusion in output of the plurality of IO operators based on the plurality of predicates. The set of rows is a proper subset of a plurality of rows stored in the plurality of segments based on the filtering of the at least one row. In various embodiments, the IO pipeline of at least one segment of the set of segments includes at least one source element and further includes at least one filter element. The at least one filter element can be based on at least some of the plurality of predicates.
In various embodiments, generating the IO pipeline for each segment includes selecting the IO pipeline from a plurality of valid IO pipeline options for each segment. In various embodiments selecting the IO pipeline from a plurality of valid IO pipeline options for each segment is based on index efficiency metrics generated for previously utilized IO pipelines of previous queries.
In various embodiments, the IO pipeline is generated for each given segment by one of the plurality of nodes that stores the given segment. Each of the plurality of IO operators are performed upon each segment by the one of the plurality of nodes that stores the given segment. A first node storing a first segment of the set of segments generates the IO pipeline for the first segment and performs the plurality of IO operators upon the first segment, and a second node storing a second segment of the set of segments generates the IO pipeline for the second segment and performs the plurality of IO operators upon the second segment.
In various embodiments, the query operator flow includes a plurality of additional operators, such as aggregation operators and/or join operators, for performance upon the set of rows read from the set of segments via performance of the plurality of IO operators. In various embodiments, the plurality of IO operators are performed by nodes at an IO level of a query execution plan, and these nodes send their output to other nodes at an inner level of the query execution plan, where these additional operators are performed by nodes at an inner level and/or root level of a query execution plan.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: identify a plurality of predicates of a query for execution; generate a query operator flow for a query by including the plurality of predicates in a plurality of IO operators of the query operator flow: and/or facilitate execution of the query by, for each given segment of a set of segments stored in memory, generating an IO pipeline for each given segment based on a secondary indexing scheme of a set of secondary indexes of the each segment and based on plurality of predicates, and/or performing the plurality of IO operators upon each given segment by applying the IO pipeline to the each segment.
As discussed in conjunction with
Each IO operator 2821 can maintain their own source queue 2855 based on the received flow of row numbers from prior sources. For example, as cow numbers are received as output from a first IO operator for a first corresponding source, corresponding IO requests indicating these row numbers are appended to the source queue 2855 for a subsequent, second IO operator that is after the first IO operator in the serialized ordering. IO requests with lower row numbers are prioritized in the second IO operator's source queue 2855 are executed before IO requests higher row numbers, and/or IO requests are otherwise ordered by row number in source queues 2855 accordance with a common ordering scheme across all IO operators. In particular, to prevent pipeline stall. the source queues 2855 of all different IO operators can all be ordered in accordance with a shared ordering scheme, for example. where lowest row numbers in source queues 2855 can therefore be read first in source queries for all sources.
As each: IO operator reads blocks from disk via a plurality of IO requests, they can each maintain an ordered list of completed and pending requests in their own source queue. The IO operators can serve each row lists and column views (when applicable) from that data.
The shared ordering scheme can be in accordance with an ordering of a shared IO request priority queue 2850. For example, the shared IO request priority queue 2850 is prioritized by row number, where lower row numbers are ordered before higher tow numbers. This shared IO request priority queue 2850 can include all IO requests for the IO pipeline across all source queues 2855, prioritized by row number.
For example, the final IO operator 2821 sourcing colD can make requests and read values before the first IO operator 2821 sourcing colC has finished completing all requests to output row numbers of the segment based on the value of colC based on all IO operators making requests in accordance with the shared IO request priority queue 2850.
As a particular example, IO requests across the IO pipeline as a whole are made to the corresponding segment one at a time. At a given time, a lowest tow number pending an IO request by one of the plurality of IO operators is read before any other pending IO requests with higher corresponding row numbers based on being most favorably ordered in the shared IO request priority queue 2850. This enables progress to be made for lower row numbers through the IO pipeline, for example, to conserve memory resources. In some embodiments. vectorized reads can be built from the priority queue when enough requests present and/or when IO is forced, for example, for final reads via a final IO operator in the serialized ordering of the pipeline.
The source queue 2855 of a given IO operator can include a plurality of pending 10 and completed IO by the corresponding IO operator. For example, completed IO can persist in the corresponding IO operator's queue until the corresponding output, such as a row number or value is processed by a subsequent IO operator to generate its own output.
In general, each disk block needs to be read only once. Multiple row lists and column views can be served from a single block. The IO pipeline can support read-ahead within a pipeline and also into the next pipeline in order to maintain deep IO queries.
The priority queue ordering can be also utilized in cases of pipeline deadlock to enable progress on a current row need when more memory is needed: necessary memory blocks can be allocated by identifying the lowest priority completed IO in the priority queue. When more memory is available. IO operators can read-along to maintain a number of in-flight requests. During an out of memory (OOM) event, completed IO can be dropped and turned back into pending IO, which can be placed back in the request queue. In particular, in an OOM condition, read-ahead blocks may need to be discarded and re-read on the subsequent pull when resources are available. Higher row numbers can be discarded first in these cases, for example, from the tail of source queues 2855, to maintain forward progress. In some embodiments, because rows are pulled in order, column leveling is not an issue In some embodiments, if the current completed IO for a source is dropped, the pipeline will stall until it can be re-read.
In various embodiments a query processing system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, cause the query processing system to determine an IO pipeline that includes a serialized ordering of a plurality of IO operators for execution open a segment in accordance with a set of query predicates. An IO request priority queue ordered by row number for a plurality of row-based IO for performance by the plurality of IO operators is maintained. Output for each of the plurality of IO operators is generated based on each of the plurality of row-based IO performing respective ones of the plurality of row-based IO in accordance with the IO request priority queue. A set of values of a proper subset of rows filtered from a plurality of rows stored in the segment are outputted, in accordance with the set of query predicates, based on the output of a last-ordered one of the plurality of IO operators in the serialized ordering.
Step 2982 includes determining au IO pipeline that includes a serialized ordering of a plurality of IO operators for execution upon a segment in accordance with a set of query predicates. Step 2984 includes maintaining an IO request priority queue ordered by row number for a plurality of row-based IO for performance by the plurality of IO operators. Step 2986 includes generating output for each of the plurality of IO operators based on each of the plurality of row-based IO performing respective ones of the plurality of row-based IO in accordance with the IO request priority queue. Step 2988 includes outputting a set of values of a subset of rows filtered from a plurality of rows stored in the segment, in accordance with the set of query predicates. based on the output of a last-ordered one of the plurality of IO operators in the serialized ordering.
In various embodiments, the subset of rows is a proper subset of the plurality of rows based on at least one row of the plurality of rows being filtered out by one of the plurality of IO operators due to not meeting the filtering requirements of the set of query predicates. Alternatively, the subset of rows includes all of the plurality of rows based on ne rows in the plurality of rows being filtered out by any of the plurality of IO operators due to all rows in the plurality of rows meeting the filtering requirements of the set of query predicates. As another example, the subset of rows includes none of the plurality of rows based on all rows in the plurality of rows being filtered out by the plurality of IO operators due to no rows in the plurality of rows meeting the filtering requirements of the set of query predicates.
In various embodiments, subsequent ones of the plurality of IO operators in the serialized ordering generate their output by utilizing output of prior ones of the ones of the plurality of IO operators in the serialized ordering. In various embodiments output of each of the plurality of IO operators includes a flow of data ordered by row number based on performing respective ones of the plurality of mow-based IO in accordance with the IO request priority queue. In various embodiments, the flow of data outputted by each of the plurality of IO operators includes a flow of row numbers ordered by row number and/or a flow of values of at least one column of rows in the plurality of rows, ordered by row number.
In various embodiments, the segment includes a plurality of secondary indexes generated in accordance with a secondary indexing scheme. The proper subset of rows are filtered from a plurality of rows stored in the segment based on at least one of the plurality of IO operators generating its output as a filtered subset of rows read in its respective ones of the plurality of row-based IO by utilizing the plurality of secondary indexes.
In various embodiments, the plurality of secondary indexes includes a first set of indexes for a first column of the plurality of rows stored in the segment in accordance with a first type of secondary index, and the plurality of secondary indexes includes a second set of indexes for a second column of the plurality of rows stored in the segment in accordance with a second type of secondary index. A first one of the plurality of IO operators generates its output in accordance with a first predicate of the set of predicates corresponding to the first column by utilizing the first set of indexes, and a second one of the plurality of 10 operators generates its output in accordance with a second predicate of the set of predicates corresponding to the second column by utilizing the second set of indexes.
In various embodiments, the IO pipeline further includes at least one filtering operator, and the proper subset of rows of the plurality of rows stored is further filtered in by the at least one filtering operator. In various embodiments, the at least one filtering operator is in accordance with one of the set of predicates corresponding to one column of the plurality of rows based on the segment not including any secondary indexes corresponding to the one column.
In various embodiments, generating output for each of the plurality of operator includes, via a first one of the plurality of IO operators, generating first output that includes a first set of row numbers as a proper subset of a plurality of row numbers of the segment via by performing a first set of row-based IO of the plurality of row-based IO in accordance with the IO request priority queue. Generating output for each of the plurality of operators can further include, via a second one of the plurality of IO operators that is serially ordered after the first one of the plurality of IO operators in the serialized ordering. generating second output that includes a second set of row numbers as a proper subset of the first set of row numbers by performing a second set of row-based IO of the plurality of row-based IO for only row numbers included in the first set of row numbers, in accordance with: the IO request priority queue.
In various embodiments, the first set of row-based IO includes reads to a first column of the plurality of rows, and the second set of row-based IO includes reads to a second column of the plurality of rows. The first set of row numbers are fibered from the plurality of row numbers by the first one of the plurality of IO operators based on applying a first one of the set of predicates to values of the first column. The second set of row numbers are filtered from the first set of rows numbers by the second one of the plurality of IO operators based on applying a second one of the set of predicates to values of the second column.
In various embodiments, the serialized ordering of the plurality of IO operators includes a parallelized set of IO operators that is serially after the first one of the plurality of IO operators. The parallelized set of IO operators includes the second one of the plurality of IO operators and further includes a third IO operator of the plurality of IO operators. Generating output for each of the plurality of operators can further include, via the third one of the plurality of IO operators, generating third output that includes a third set of row numbers as a second proper subset of the first set of row number of the segment by performing a second set of row-based IO of the plurality of row-based IO for only row numbers included in the first set of row numbers, in accordance with the IO request priority queue.
In various embodiments, the method further includes generating fourth output via a fourth one of the plurality of IO operators that is serially after the parallelized set of IO operators that corresponds to a proper subset of rows included in a union of outputs of the parallelized set of IO operators.
In various embodiments, respective ones of the plurality of row-based IO are maintained in a queue by the each of the plurality of IO operators in accordance the ordering of the IO request priority queue. In various embodiments, the queue maintained by the each given IO operator of the plurality of IO operators includes a set of IO competed by the given IO operator and further includes a set of IO pending completion by the given IO operator.
In various embodiments, the method includes detecting an out-of-memory condition has been met, and for removing a subset of the plurality of row-based IO from the queries maintained by the each of the plurality of IO operators by selecting ones of the plurality of row-based IO that are least favorably ordered in the IO request priority queue. In various embodiments, at least one of the plurality of row-based IO removed from a queue maintained by one of the plurality of IO operators was already completed by the one of the one of the plurality of IO operators. The at least one of the plurality of row-based IO is added to the queue maintained by one of the plurality of IO operators as pending completion based on being removed from the queue in response to detecting that memory is again available. The one of the plurality of IO operators re-performs the at least one of the plurality of row-based IO based on being indicated in the queue as pending completion.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: determine an IO pipeline that includes a serialized ordering of a plurality of IO operators for execution upon a segment in accordance with a set of query predicates, maintain an IO request priority queue ordered by row number for a plurality of row-based IO for performance by the plurality of IO operators: generate output for each of the plurality of IO operators based on each of the plurality of row-based IO performing respective ones of the plurality of row-based IO in accordance with the JO request priority queue; and/or output a set of values of a proper subset of rows filtered from a plurality of rows stored in the segment, in accordance with the set of query predicates, based on the output of a last-ordered one of the plurality of IO operators in the serialized ordering.
As the superset of rows is guaranteed to include all rows satisfying the query predicate, only data values for rows included in the superset indicated by the indexing scheme need be accessed. For some probabilistic indexing schemes, this superset of rows may be a small subset of the full set of rows that would otherwise need be accessed if the indexing scheme were not utilized, which improves IO efficiency over the case where no index were utilized as a smaller proportion of data values need be read. For example. a superset of 110 rows is returned based on accessing a probabilistic index structure stored to index a given column of a dataset: that includes 1 million tows, and the true set of rows corresponds to 100 rows of this superset of 110 rows. Rather than the data values for all 1 million rows in the data set, only the identified 110 data values for the column are read from memory, enabling the 10 false positive rows to be identified and filtered out.
This can be particularly desirable when the data values correspond to large values, text data, unstructured data, and/or variable length values that are expensive to read from memory and/or to temporarily store for comparison to filtering parameters and/or for movement between nodes implementing a query execution plan. While probabilistic indexes often support fixed-length columns, this construct can be implemented to apply a probabilistic index to variable-length columns, such as varchar data types. string data types, and/or text values. For example, the variable-length data of a variable-length column can be indexed via a probabilistic index based on hashing the variable-length values of this variable-length column, which is probabilistic in nature due to hash collisions where multiple data values hash to the same values, and utilizing the index for queries for equality with a particular value may include other values due to these hash collisions.
While a perfect indexing scheme that guarantees exactly the true set of rows be read could further improve IO efficiency. the corresponding index structure can be costly to store in memory and/or may be unreasonable for certain data types, such as variable-length column data. In particular, a probabilistic index structure indexing a given column may be far more memory efficient than a perfect indexing scheme particularly when the column values of the column are variable-length and/or have high cardinality A probabilistic indexing structure, while requiring false-positive rows be read and filtered, can thus be preferred over a perfect indexing structure for some or all columns as it can handle variable-length data and/or requires fewer memory resources for storage.
Thus, the utilization of probabilistic indexing schemes in query execution as discussed in conjunction with one or more embodiments described in conjunction with
Furthermore, the utilization of probabilistic indexing schemes in query execution as discussed in conjunction with one or more embodiments described in conjunction with
Embodiments of probabilistic indexing schemes described in conjunction with
As discussed in further detail herein, an IO pipeline, such as an IO pipeline 2835 as discussed in conjunction with
As illustrated in
An IO operator execution module 2840 can execute the IO pipeline 2835 to render a filtered row set from a full set of rows of a corresponding dataset against which the given query is executed. This can include performing row reads based on accessing index data and/or raw data values for rows stored in one or more segments of a segment storage system 2508, for example, as discussed in conjunction with
The probabilistic index-based IO construct 3010 can include a probabilistic index element 3012, a source element 3014 downstream from the probabilistic index element 3012 and applied to output of the probabilistic index element 3012, and/or a filter element 3016 that is downstream from the source element 3014 and applied to output of the probabilistic index element 3012. The probabilistic index element 3012, source element 3014, and/or filter element 3016 of the probabilistic index-based IO construct 3010 can collectively function as an IO operator 2821 of
The IO operator execution module 2840 can execute the probabilistic index-based IO construct 3010 against a dataset via one or more processing resources, such as a plurality of nodes 37 independently performing row reads at an IO level 2416 of a query exception plan 2405 as discussed in conjunction with
The probabilistic index-based IO construct 3010 can be utilized to implement a given query predicate 2822 based on the probabilistic index element 3012 being applied to access index data for a given column identified via a column identifier 3041 indicated in the query predicate. Index probe parameter data 3042 indicating which rows be identified can be determined based on the filter parameters 3048. For example. filter parameters indicating equality with, being less than, and/or being greater than a given literal value can be applied to determine corresponding index probe values utilized to identify corresponding row identifiers. such as a set of row numbers, indicated by the corresponding index data for the column.
The set of row identifiers returned based on given index probe parameter data 3042 denoting given filter parameters 3048 of predicates 2822 can be guaranteed to include all row identifiers for all rows that satisfy the filter parameters 3048 of the predicate 2822 for the given column. However, the set of row identifiers returned based on given index probe parameter data 3042 may include additional row identifiers for rows that do not satisfy the filter parameters 30-18 of the predicate 2822, which correspond to false-positive rows that need be filtered out to ensure query correctness.
The probabilistic index-based IO construct 3010 can be utilized to implement a given query predicate 2822 based on the source element 3014 being applied to access data values for the given column identified via the column identifier 3041 from memory. The source element 3014 can be applied such that only rows identified by the probabilistic index element 3012 be accessed.
The probabilistic index-based 10 construct 3010 can be utilized to implement a given query predicate 2822 based on the filter element 3016 being applied to filter rows from the set of row identifiers returned by the probabilistic index element. In particular, the false-positives can be identified and removed to render only the true set of rows satisfying the given filter parameters 3048 based on utilizing the data values of the given column read for the rows in the set of rows row identifiers returned by the probabilistic index element. Ones of this of rows row identifiers with data values of the given column meeting and/or otherwise comparing favorably to the filter parameters are maintained as true-positives included in the true set of rows, while other ones of this of rows row identifiers with data values of the given column not meeting or otherwise comparing unfavorably to the filter parameters are removed.
The row identifier set 3044 can include the true predicate-satisfying row set 3034 that includes all rows of the dataset satisfying one or more corresponding predicates 2822, for example, that were utilized to determine the index probe parameter data 3042 of the probabilistic index element 3012 The row identifier set 3044 can further include a false-positive row set 3035 that includes additional rows of the dataset that do not satisfy the one or more corresponding predicates 2822. For example, these rows are indexed via same index values as rows included in the true predicate-satisfying row set 3034.
The row identifier set 3044 can be a proper subset of an initial row set 3032. The initial row set 3032 can correspond to all rows of a corresponding dataset and/or all rows of a corresponding segment to which the corresponding probabilistic index-based IO constriction 3010 of the IO pipeline is applied. In some cases the initial row set 3032 is a proper subset of all rows of the corresponding dataset and/or all rows of the corresponding segment based on prior utilization of other indexes and/or filters previously applied upstream in the IO pipeline, where the probabilistic index-based IO construct 3010 is applied to only rows in the pre-filtered set of rows implemented as the initial row set 3032.
In some cases, the false-positive row set 3035 is non-null, bur is indistinguishable from the true predicate-satisfying row set 3034 due to the nature of the probabilistic indexing scheme until the respective data values are read and evaluated against the corresponding filtering parameters of the predicate 2822. In some cases, the false-positive tow set 3035 is null, but it is not known whether the false-positive row set 3035 is null due to the nature of the probabilistic indexing scheme until the respective data values are read and evaluated against the corresponding filtering parameters of the predicate 2822. The true predicate-satisfying row set 3034 can also be null or non-null. In cases where the true predicate-satisfying row set 3034 is null but the false-positive row set 3035 is non-null, the resulting output of the probabilistic index-based IO construct 3010 will be null once filtering element 3016 is applied.
The row identifier set 3044 can be utilized by 3014 by a source element 3014 to read data values for corresponding rows in row storage 3022 to render a data value set 3046. This row storage 3022 can be implemented via memory of the database system 10, such as via memory resources such as memory drives 2425 of one or more nodes 37, for example, such as segment row data 2505 of one or more segments 2424 generated and stored by the database system 10 as discussed in conjunction with
The data value set 3046 can be utilized by filter element 3016 to identify and remove the false-positive row set 3035. For example, each given data value of the data value set 3046 is processed via comparison to filtering parameters 3048 of the query predicate to determine whether the given data value satisfies the query predicate, where only the rows with data values satisfying the query predicate are identified in the outputted row set. This guarantees that the outputted row set corresponds to exactly the 1to predicate-satisfying row set 3034 based on guaranteeing that all out all rows in the false-positive row set 3035 are filtered out based on having data values comparing unfavorably to the corresponding predicate 2822.
The true predicate-satisfying row set 3034 outputted by a given probabilistic index-based IO constrict 3010 can be included in and/or utilized to generate a query resultant. The true predicate-satisfying row set 3034 outputted by a given probabilistic index-based IO construct 3010 can be further processed in further operators of the IO pipeline 2835, and/or can be further processed via further operators of the query operator execution flow 2817, for example, via inner and/or toot nodes of the query execution plan 2405.
The true predicate-satisfying row set 3034 can indicate only row identifiers, such as row numbers, for the rows of the true predicate-satisfying row set 3034, where this true predicate-satisfying row set 3034 is optionally further filtered and/or combined with other sets via further filtering operators and/or set operations via upstream operators of the IO pipeline 2835 and/or the query operator execution flow 2817. Corresponding data values of the data value set 3046 can optionally be outputted alternatively or in addition to the row identifiers, for example, based on the query resultant including the data values for the corresponding column, based on further processing of the data values upstream in the IO pipeline, and/or based on further processing of the data values via other operators of the IO pipeline 2835 and/or of the query operator execution flow 2817.
As illustrated, probabilistic index structure 3020. 1 is one of a set of probabilistic index structures 3020 for some or all of a set of columns 3023.1-3023. Y. In this case, the probabilistic index structure 3020. 1 is accessed based on the corresponding predicate 2822 involving column 3023.1. Note that some columns 3023 of the initial row set 3032 may be indexed via non-probabilistic indexing schemes and/or may not be indexed at all.
Different probabilistic index structures 3020 or different columns, such as two different given probabilistic index structures 3020. A and 3020.B of two columns 3023 A and 3023.B of the set of columns, can be stored via shared and/or distinct memory resources. Different probabilistic index structures for different columns, such as probabilistic index structures 3020.A and 3020. B, can be implemented as a combined index structure, or as distinct index structures based on different columns being indexed separately being indexed via different indexing schemes, and/or being indexed with different parameters. A given segment can store multiple different probabilistic index structures for data values of multiple ones of the columns for its set of rows. A given probabilistic index structure 3020 of a given column of a given dataset can include multiple individual probabilistic index structures stored in each of a set of different segments, indexing different corresponding subsets of rows in the given dataset for the given column via the same or different probabilistic indexing scheme and/or via the same or different parameters.
In this case, row identifier set 3044.2 outputted by probabilistic index element 3012 indicates row a, row b, and row d. but not row e, based on the index value 3043.2 in the probabilistic index structure 3020 I mapping to and/or otherwise indicating rows a, b, and d. The source element 3014 reads the data values 3024.1a, 3024. 1.b, and 3024. 1.d accordingly. Filter element 3016 applies filter parameters indicating some function, such as a logical condition or predicate, of data values 3024. J of column 3023.1. where row a and row d are identified in row identifier subset 3045 outputted by filtering element 3016 based on data value 3024.1.a and 3024.d satisfying filter parameters 3048, and based on data value 3024.1.c not satisfying filter parameters 3048. The row identifier subset 3045 is guaranteed to be equivalent to the true predicate-satisfying row set 3034 of row identifier set 3044.2, and is guaranteed to not include any rows of the false-positive row set 3035 of row identifier set 3044.2.
The query processing system 2802 of
The utilization of probabilistic indexes by the IO operator execution module 2840 to execute probabilistic index-based IO constructs 3010 of IO pipelines 2835 cannot practically be performed by the human mind, particularly when the database system 10 is implemented to store and perform queries against records at a massive scale as discussed previously. In particular. the human mind is not equipped to generate a row identifier set 3044, read corresponding data values, and filter the corresponding data values for millions, billions, and/or trillions of records stored as many Terabytes. Petabytes, and/or Exabytes of data. Furthermore, the human mind is not equipped to distribute and perform these steps of an IO pipeline as multiple independent processes, such as hundreds, thousands, and/or millions of independent processes, in parallel and/or within overlapping time spans.
In various embodiments a query processing system includes at least one processer and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, can cause the query processing system to: determine an IO pipeline that includes a probabilistic index-based IO construct for access of a first column of a plurality of rows based on a query including a query predicate indicating the first column; and/or apply the probabilistic index-based IO construct in conjunction with execution of the query via the IO pipeline. Applying the probabilistic index-based IO construct in conjunction with execution of the query via the IO pipeline can include applying an index element of the probabilistic index-based 10 construct to identify a first subset of rows as a proper subset of the plurality of rows based on index data of a probabilistic indexing scheme for the first column of the plurality of rows, and/or applying a filter element of the probabilistic index-based IO construct to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of a first subset of a plurality of column values corresponding to the first subset of rows that compare favorably to the query predicate.
Some or all of the steps of
Step 3082 includes storing a plurality of column values for a first column of a plurality of rows. Step 3084 includes indexing the first column via a probabilistic indexing scheme. Step 3086 includes determining an IO pipeline that includes a probabilistic index-based IO construct for access of the first column based on a query including a query predicate indicating the first column. Step 3088 includes applying the probabilistic index-based IO construct in conjunction with execution of the query via the IO pipeline.
Performing step 3088 can optionally include performing step 3090 and/or step 3092. Step 3090 includes applying at index element of the probabilistic index-based IO construct to identify a first subset of rows as a proper subset of the plurality of rows based on index data of the probabilistic indexing scheme for the first column. Step 3092 includes applying a filter element of the probabilistic index-based IO construct to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of a first subset of the plurality of column values corresponding to the first subset of rows that compare favorably to the query predicate. In various embodiments, the second subset of rows is a proper subset of the first subset of rows.
In various embodiments, applying the probabilistic index-based 10 construct in conjunction with execution of the query via the IO pipeline further includes applying a source element of the probabilistic index-based 10 construct to read the first subset of the plurality of column values corresponding to the first subset of rows. In various embodiments, the source element is applied after the index element in the IO pipeline, and/or the filter element is applied after the source element in the IO pipeline.
In various embodiments, the probabilistic indexing scheme is an inverted indexing scheme. In various embodiments, the first subset of rows are identified based on inverted index data of the inverted indexing scheme.
In various embodiments, the index data of the probabilistic indexing scheme includes a plurality of hash values computed by performing a hash function on corresponding ones of the plurality of column values. In various embodiments, the first subset of rows are identified based on a hash value computed for a first value indicated in the query predicate. In various embodiments. the plurality of column values for the first column are variable-length values, and/or the plurality of hash values are fixed-length values.
In various embodiments, the query predicate indicates an equality condition requiring equality with the first value. The first subset of rows can be identified based on having hash values for the first column equal to the hash value computed for the first value. A set difference between the first subset of rows and the second subset of rows can correspond to bash collisions for the bash value. The second subset of rows can be identified based on having column values for the first column equal to the first value.
In various embodiments, the second subset of rows includes every row of the plurality of rows with a corresponding column value of the first column comparing favorably to the query predicate. A set difference between the first subset of rows and the second subset of rows can include every row in the first subset of rows with a corresponding column value of the first column comparing unfavorably to the query predicate.
In various embodiments, the IO pipeline for the query includes a plurality of probabilistic index-based IO constructs based on a plurality of query predicates of the query that includes the query predicate. In various embodiments, the method further includes storing a second plurality of column values for a second column of the plurality of rows in conjunction with the probabilistic indexing scheme. The probabilistic index-based IO construct can be a first one of the plurality of probabilistic index-based IO constructs, and/or a second one of the plurality of probabilistic index-based IO constructs can correspond to access to the second column based on another query predicate of the plurality of query predicates indicating the second column.
In various embodiments, the plurality of rows are stored via a set of segments. The IO pipeline can be generated for a first segment of the set of segments, and/or a second IO pipeline can be generated for a second segment of the set of segments. The IO pipeline can be different from the second IO pipeline based on the first segment utilizing the probabilistic indexing scheme for the first column and based on the second segment utilizing a different indexing scheme for the first column.
In various embodiments, the method further includes determining a selected false-positive tuning parameter of a plurality of false-positive tuning parameter options. The probabilistic indexing scheme for the first column can be in accordance with the selected false-positive tuning parameter, and/or a size of a set difference between the first subset of rows and the second subset of rows can be based on the selected false-positive tuning parameter.
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments, a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: store a plurality of column values for a first column of a plurality of rows; index the first column via a probabilistic indexing scheme: determine an IO pipeline that includes a probabilistic index-based IO construct for access of the first column based on a query including a query predicate indicating the first column; and/or apply the probabilistic index-based IO construct in conjunction with execution of the query via the IO pipeline. Apply the probabilistic index-based IO construct in conjunction with execution of the query via the IO pipeline can include: applying an index element of the probabilistic index-based IO construct to identify a first subset of rows as a proper subset of the plurality of rows based on index data of the probabilistic indexing scheme for the first column; and/or applying a filter element of the probabilistic index-based IO construct to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of a first subset of the plurality of column values corresponding to the first subset of rows that compare favorably to the query predicate.
The conjunction 3112 can indicate a set of operands 3114, which can include at least two operands 3114. Each operand 3114 can involve at least one corresponding column 3023 of the dataset identified via a corresponding one or more column identifiers. In this example, two operands 3114. A and 3114. B are included, where operand 3114. A indicates a first column 3023.A identified by column identifier 3041.A, and operand 3114 B indicates a second column 3023 B identified by column identifier 3041.B. While not illustrated, conjunctions 3112 can optionally indicate more than two operands in other embodiments.
Corresponding operand parameters 3148 can indicate requirements for the data values in the corresponding columns of the operand 3114. For example, only rows with column values meeting the operand parameters of all of the operands 3114 of the conjunction operator will be outputted in executing the conjunction of the operator execution flow. In this example, the operand parameters 3148. A can indicate a logical constraint that evaluates to either true or false based on the data value of column A for the corresponding row. Similarly, the operand 3114.B can indicate a logical construct that evaluates to either true or false based on the data value of column B for the corresponding tow. For example, the conjunction evaluates to true when the value of column A is equal to a first literal value and when the value of column A is equal to a second literal value. Any other type of operands not based on equality, such as conditions based on being less than a literal value. greater than a literal value, including a consecutive text pattern, and/or other conditional statements evaluating to either true or false can be implemented as operand parameters 3148.
The IO pipeline generator module 2834 can generate a corresponding IO pipeline 2835 based on pushing the conjunction 3112 to the IO level as discussed previously. This can include adapting the probabilistic index-based 10 construct 3010 of
The probabilistic index-based conjunction construct 3110 can alternatively or additionally be considered a type of probabilistic index-based IO construct 3010 specific to implementing predicates 2822 that include conjunction constructs. The probabilistic index-based conjunction construct 3110 can be implemented upstream and/or downstream of other IO constructs of the IO pipeline, such as other IO probabilistic index-based IO constructs 3010 or other source utilize different non-probabilistic indexing schemes, and/or other constructs of the IO pipeline as discussed herein.
In particular, a set of index elements 3012 can be included as elements of parallel probabilistic index-based IO constructs 3010 based on the corresponding set of operands 3114 of the conjunction 3112 being implemented. For example, different processing core resources 48 and/or nodes 37 can be assigned to process the different index elements 3012, and/or the set of index elements 3012 can otherwise be processed in parallel. In this example a set of two index elements 3012.A and 3012.B are implemented for columns 3023.A and 3023.B, respectively based on these columns being indicated in the operands of the conjunction 3112 Index probe parameter data 3042 of each index element 3012 can be based on the operand parameters 3148 of the corresponding operand 3114. For example, index probe parameter data 3042.A of index element 3012.A indicates an index value determined based on the literal value to which the operand parameters 3148. A indicates the corresponding column value must be equal to satisfy the operand 3114. A, and/or index probe parameter data 3042.B of index element 3012.B can indicate an index value determined based on the literal value to which the operand parameters 3148. B indicates the corresponding column value must be equal to satisfy the operand 3114.B.
A set of source elements 3014 can be included in parallel downstream of the respective index elements. In some embodiments, the set of source elements 3014 are only included in cases where the column values were not previously sourced upstream of the probabilistic index-based conjunction construct 3110 for another use in other constructs of the IO pipeline. Different processing core resources 48 and/or nodes 37 can be assigned to process the different source elements 3014, and/or the set of source elements 3014 can otherwise be processed in parallel.
Each parallel track can be considered an adapted probabilistic index-based IO construct 3010. However, rather than also including each of a set of parallel filer elements 3016 in parallel to implement a set of full probabilistic index-based IO constructs 3010 of
Execution of an example probabilistic index-based conjunction construct 3110 is illustrated in
This results in identification of a set of row identifier sets 3044 via each probabilistic index element 3012. As each operand 3114 can be treated as a given predicate 2822, each row identifier set 3044.A and 3044.B can be guaranteed to include the true predicate-satisfying row set 3034 satisfying the corresponding operand 3114.A and/or 3148. B, respectively, as discussed previously. Each row identifier set 3044. A and 3044. B may also have false positive rows of corresponding false-positive row sets 3035.A and 3035 B, respectively, that are not distinguishable from the corresponding true operand-satisfying row sets 3034 until the corresponding data values are read and processed as discussed previously.
Each source element 3014 can read rows of the corresponding row identifier set 3044 from row storage 3022, such as from one or more segments, to render a corresponding data value set 3046 as discussed previously. Filter element 3016 can be implemented to identify rows included in both row identifier sets 3044. A and 3044.B. However, because the row identifier sets may include false positives, the filter element 3016 must further evaluate column A data values of data value set 3046 A of these rows and evaluate column B data values of data value set 3046. B to determine whether they satisfy or otherwise compare favorably to the respective operands of the conjunction, thus further filtering out false-positive row sets 3035. A and 3035.B in addition to facilitating a set intersection. For example, a function F(data value 3024.A) is based on the operand 3114. A and, for a given data value 3024 of column A for a given row evaluates to either true or false, where the given row only satisfies the operand 3114.A when the function evaluates to true: and a function G(data value 3024.B) is based on the operand 3114.B and, for a given data value 3024 of column B for a given row evaluates to either true or false, where the given row only satisfies the operand 3114B when the function evaluates to true.
Only ones of the rows included in both row identifier sets 3044.A and 3044.B having data values in data value sets 3046.A and 3046.B that satisfy both operands 3114.A and 3148.B are included in a true conjunction satisfying row set 3134 outputted by the filter element 3016. This true conjunction satisfying row set 3134 can be guaranteed to be equivalent to a set intersection between the true operandA-satisfying row set 3034. A and the true operandB-satisfying row set 3034.B. Note that, due 10 the potential presence of false-positives in row identifier set 3044. A and/or 3044. B, the true conjunction satisfying row set 3134 may be a proper subset of the set intersection of row identifier sets 3044 A and 3044.B, and thus the filter element that evaluates data values of these rows is thus necessary to ensure that exactly the true conjunction satisfying row set 3134 is outputted by the probabilistic index-based conjunction construct 3110. A set difference between the set intersection of row identifier sets 3044.A and 3044.B, and the true conjunction satisfying row set 3134, can include: one or more rows included in false-positive row set 3035.A and in false-positive row set 3035.B: one or more rows included in false-positive row set 3035.A and in the operandB satisfying row set 3034.B, and/or one or more rows included in false-positive row set 3035 B and in true operandA satisfying row set 3034.A. In some cases, the true conjunction satisfying row set 3134 can be equivalent to the intersection of row identifier sets 3044.A and 3044.B when the intersection of row identifier sets 3044.A and 3044.B does not include any rows of false-positive row set 3035.A or 3035.B. The true conjunction satisfying row set 3134 can be guaranteed to be a subset of the intersection of row identifier sets 3044. A and 3044.B as either an equivalent set or a proper subset.
In various embodiments a query processing system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processer, can cause the query processing system to: determine a query operator execution flow that includes a logical conjunction indicating a first column of a plurality of rows in a first operand and indicating a second column of the plurality of rows in a second operand: and/or facilitate execution of the logical conjunction of the query operator execution flow against the plurality of rows. Facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows can include: utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a proper subset of the plurality of rows based on the first operand: and/or filtering the first subset of rows to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of the first subset of rows having first column values of the first column that compare favorably to the first operand, and having second column values of the second column that compare favorably to the second operand.
Some or all of the method of
Some or all of the steps of
Step 3182 includes determining a query operator execution flow that includes a logical conjunction indicating a first column of a plurality of rows in a first operand and indicating a second column of the plurality of rows in a second operand. Step 3184 includes facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows.
Performing step 3184 can include performing step 3186 and/or 3188. Step 3186 includes utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a proper subset of the plurality of rows based on the first operand. Step 3188 includes filtering the first subset of rows to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of the first subset of rows having first column values of the first column that compare favorably to the first operand, and having second column values of the second column that compare favorably to the second operand.
In various embodiments. facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows further includes reading a first set of column values from memory based on reading column values of the first column only for rows in the first subset of rows. Filtering the first subset of rows to identify the second subset of rows can include utilizing the first set of column values.
In various embodiments, facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows further includes utilizing second index data of a probabilistic indexing scheme for the second column of the plurality of rows to identify a third subset of rows as another proper subset of the plurality of rows based on the second operand. The second subset of rows can be further identified based on filtering the third subset of rows. The second subset of rows can be a subset of the third subset of rows. In various embodiments, facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows further includes reading a second set of column values from memory based on reading column values of the second column only for rows in the third subset of rows, where filtering the third subset of rows to identify the second subset of rows includes utilizing the second set of column values. In various embodiments, the first subset of rows and the third subset of rows are identified in parallel via a first set of processing resources and a second set of processing resources. respectively.
In various embodiments, the first index data of the probabilistic indexing scheme for the first column are a first plurality of hash values computed by performing a first hash function on corresponding first column values of the first column. The first subset of rows can be identified based on a first bash value computed for a first value indicated in the first operand In various embodiments, second index data of the probabilistic indexing scheme for the second column can be a second plurality of hash values computed by performing a second hash function on corresponding second column values of the second column. The third subset of rows can be identified based on a second hash value computed for a second value indicated in the second operand. In various embodiments, the first operand indicates a first equality condition requiring equality with the first value. The first subset of rows can be identified based on having hash values for the first column equal to the first hash value computed for the first value. The second operand can indicate a second equality condition requiring equality with the second value. The third subset of rows can be identified based on having hash values for the second column equal to the second hash value computed for the second value.
In various embodiments, the second subset of rows includes every row of the plurality of rows with a corresponding first column value of the first column and second column value of the second column comparing favorably to the logical conjunction. The second subset of rows can be a proper subset of a set intersection of the first subset of rows and the third subset of rows and/or can be a non-null subset of the set intersection of the first subset of rows and the third subset of rows.
In various embodiments, the probabilistic indexing scheme is an inverted indexing scheme. The first subset of rows can be identified based on utilizing index data of the inverted indexing scheme. In various embodiments, a plurality of column values for the first column are variable-length values. In various embodiments a plurality of hash values were generated from the plurality of column values for the first column based on the probabilistic filtering scheme. The plurality of bash values can be fixed-length values. Identifying the first subset of rows can be based on the plurality of hash values.
In various embodiments, at least one of the first subset of rows having a first column value for the first column that compares unfavorably to the first operand is included in the first subset of rows based on the probabilistic indexing scheme for the first column. In various embodiments, the at least one of the first subset of rows is not included in the second subset of rows based on the first column value for the first column comparing unfavorably to the first operand.
In various embodiments, facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows includes applying at least one probabilistic index-based IO construct of an IO pipeline generated for the query operator execution flow. For example, at least one probabilistic index-based IO construct of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments, a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: determine a query operator execution flow that includes a logical conjunction indicating a first column of a plurality of rows in a first operand and indicating a second column of the plurality of rows in a second operand: and/or facilitate execution of the logical conjunction of the query operator execution flow against the plurality of rows. Facilitating execution of the logical conjunction of the query operator execution flow against the plurality of rows can include: utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a proper subset of the plurality of rows based on the first operand: and/or filtering the first subset of rows to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of the first subset of rows having first column values of the first column that compare favorably to the first operand, and having second column values of the second column that compare favorably to the second operand.
The disjunction 3212 can indicate a set of operands 3114, which can include at least two operands 3114. Each operand 3114 can involve at least one corresponding column 3023 of the dataset identified via a corresponding one or more column identifiers. In this example, two operands 3114. A and 3114. B are included, where operand 3114. A indicates a first column 3023.A identified by column identifier 3041.A, and operand 3114.B indicates a second column 3023.B identified by column identifier 3041.8 White not illustrated, disjunctions 3212 can optionally indicate more than two operands in other embodiments. The operands 3114.A and 3114.B of
The IO pipeline generator module 2834 can generate a corresponding IO pipeline 2835 based on pushing the disjunction 3212 to the IO level as discussed previously. This can include adapting the probabilistic index-based IO construct 3010 of FIGS. 30A-30H to implement a probabilistic index-based disjunction construct 3210. For example, the probabilistic index-based disjunction construct 3210 can be considered an adapted combination of multiple probabilistic index-based IO constructs 3010 it parallel to source and filter corresponding operands of the disjunction to output a plurality of sets of filtered rows in parallel, and to then output a union of this plurality of sets of filtered rows via a set union element 3218.
The probabilistic index-based disjunction construct 3210 can alternatively or additionally be considered a type of probabilistic index-based IO constrict 3010 specific to implementing predicates 2822 that include disjunction constructs. The probabilistic index-based disjunction construct 3210 can be implemented upstream and/or downstream of other IO constructs of the IO pipeline, such as other IO probabilistic index-based IO constructs 3010 or other source utilize different non-probabilistic indexing schemes, and/or other constructs of the IO pipeline as discussed herein.
In particular, a set of index elements 3012 can be included as elements of parallel probabilistic index-based IO constructs 3010 based on the corresponding set of operands 3114 of the disjunction 3212 being implemented. For example, different processing core resources 48 and/or nodes 37 can be assigned to process the different index elements 3012, and/or the set of index elements 3012 can otherwise be processed in parallel. In this example, a set of two of index elements 3012 A and 3012. B are implemented for columns 3023. A and 3023. B, respectively based on these columns being indicated in the operands of the disjunction 3212. Index probe parameter data 3042 of each index element 3012 can be based on the operand parameters 3148 of the corresponding operand 3114. For example, index probe parameter data 3042. A of index element 3012.A indicates an index value determined based on the literal value to which the operand parameters 3148. A indicates the corresponding column value must be equal to satisfy the operand 3114. A, and/or index probe parameter data 3042.B of index element 3012 B can indicate an index value determined based on the literal value to which the operand parameters 3148.B indicates the corresponding column value must be equal to satisfy the operand 3114.B.
A set of source elements 3014 can be included in parallel downstream of the respective index elements. In some embodiments, the set of source elements 3014 are only included in cases where the column values were net previously sourced upstream of the probabilistic index-based disjunction construct 3210 for another one in other constructs of the IO pipeline. Different processing core resources 48 and/or nodes 37 can be assigned to process the different source elements 3014, and/or the set of source elements 3014 can otherwise be processed in parallel.
A set of filter elements 3016 can be included in parallel downstream of the respective source elements to filter the rows identified by respective index elements based on whether the corresponding data values for the corresponding column satisfy the corresponding operand. Each filter element filter rows based on whether the corresponding data values for the corresponding column satisfy the corresponding operand. The set of filtering elements thus filter out the false-positive rows for each respective column. A set union 3218 can be applied to the output of the fillet elements to tender the true disjunction output of the disjunction, as the input to the set union included no false-positive rows for any given parallel track.
As illustrated in
The data values of a given column for some rows included the union may not be available, and thus require sourcing after the union. For example, the data values of a given column for some rows included the union may not be available based these rows not satisfying the operand for the given column, and not being identified via the probabilistic index for the given column based on not being false-positive rows identified via the probabilistic index. These rows were therefore not read for the given column due to not being identified via the probabilistic index. However, these rows are included in the set union output based on these rows satisfying the operand for a different column, thus satisfying the disjunction. The column values for the given column are then read for these rows the first time via the downstream source element of the given column.
Alternatively or in addition, the data values of a given column for some rows included the union may not be available. and thus require sourcing after the union, based on these rows having had respective data values read for the given column via source elements 3014 due being false-positive rows identified by the respective probabilistic index utilized for the given column. However, after being sourced via the respective source element, the respective filtering element filters out these rows due to not satisfying the respective operand, which can render the respective data values unavailable downstream. However, these rows are included in the set union output based on these rows satisfying the operand for a different column, thus satisfying the disjunction. The column values for the given column are then re-read for these rows via the downstream source element of the given column.
Execution of an example probabilistic index-based disjunction construct 3210 is illustrated in
This results in identification of a set of row identifier sets 3044 via each probabilistic index element 3012. As each operand 3114 can be treated as a given predicate 2822, each row identifier set 3044.A and 3044.B can be guaranteed to include the true predicate-satisfying row set 3034 satisfying the corresponding operand 3114.A and/or 3148.B, respectively, as discussed previously. Each row identifier set 3044.A and 3044.B may also have false positive rows of corresponding false-positive row sets 3035.A and 3035.B, respectively, that are not distinguishable from the corresponding true operand-satisfying row sets 3034 until the corresponding data values are read and processed as discussed previously.
Each source element 3014 can read rows of the corresponding tow identifier set 3044 from tow storage 3022, such as from one or more segments, to render a corresponding data value set 3046 as discussed previously. Each filter element 3016 cat be implemented to identify rows satisfying the corresponding operand. For example, a first filter element 3016.A applies a first function F(data value 3024.A) for rows in row identifier set 3044. A based on data values in data values set 3046. A to identify true operandA-satisfying row set 3034. A. filtering out false-positive row set 3035.A. A second filter element 3016. B can apply a second function G(data value 3024.B) for rows in row identifier set 3044. B based on data values in data values set 3046. B to identify true operandB-satisfying row set 3034. B. filtering out false-positive row set 3035.B. F(data value 3024. A) can be based on the operand 3114.A and, for a given data value 3024 of column A for a given row evaluates to either true or false, where the given row only satisfies the operand 3114. A when the function evaluates to true, and function G(data value 3024.B) can be based on the operand 3114.B and, for a given data value 3024 of column B for a given row evaluates to either true or false, where the given row only satisfies the operand 3114.B when the function evaluates to true.
Only ones of the rows included in either row identifier set 3044. A or 3044 B having data values in data value sets 3046. A and 3046.B that satisfy either operand 3114. A or 3148.B are included in a true disjunction satisfying row set 3234 outputted by the filter element 3016. This true disjunction satisfying row set 3234 can be guaranteed to be equivalent to a set union between the true operandA-satisfying row set 3034.A and the true operandB-satisfying row set 3034.B. Note that, due to the potential presence of false-positives in row identifier set 3044. A and/or 3044.B, the true disjunction satisfying row set 3234 may be a proper subset of the set anion of row identifier sets 3044.A and 3044.B. A set difference between the set union of row identifier sets 3044.A and 3044.B, and the true disjunction satisfying row set 3234, can include one or more rows included in false-positive tow set 3035.A and in false-positive row set 3035 B: one or more rows included in false-positive row set 3035 A and not included in row identifier set 3044.B; and/or one or more rows included in false-positive row set 3035.B and not included in row identifier set 3044. A. In some cases, the true disjunction satisfying row set 3234 can be equivalent to the intersection of row identifier sets 3044.A and 3044.B when the union of row identifier sets 3044.A and 3044.B includes only rows in either true operandA-satisfying row set 3034.A or true operandB-satisfying row set 3034.B. The true disjunction satisfying row set 3234 can be guaranteed to be a subset of the union of row identifier sets 3044 A and 3044 B as either an equivalent set or a proper subset.
Data value set 3247.A can include at least data value not included in data value set 3046 A, for example, based on the corresponding row satisfying operandB but not operandA. A data value set 3247.A can include at least data value included in data value set 3046.A that is filtered out as a false positive, for example, based on the corresponding row being included in the false-positive row set 3035.A and being included in the true operandB-satisfying row set 3034.B. A data value set 3046.A can include at least data value not included in data value set 3247.A, for example, based on the corresponding row being included in the false-positive row set 3035.A, and not being included in the true operandB-satisfying row set 3034.B, thus causing the row to be not included in the set union. Similar differences between data value set 3247.B and data value set 3247.B can similarly exist for similar reasons.
In some cases, not all of the columns sourced for the disjunction are re-sourced, due to some or all columns not being required for further use downstream. For example, columns A and B are both sourced via source elements 3014 prior to the set union element 3218 as illustrated in
Rows a, b, d, e, and f are included in the row identifier set 3044. A, for example, based on having data values of column A bashing to a same value indexed in the probabilistic index structure 3020. A or otherwise being indexed together, despite not all being equal to “hello”. Their respective values are read from memory in row storage 3022 via source element 3014.A, and filter element 3016. A automatically removes the false-positive row set 3035. A based on filtering out: row b due to having a column A value not equal to “hello.” and row d due to having a column A value not equal to “hello”. This renders true operandA-satisfying row set 3034.A.
Rows a, b, d, and f are included in the row identifier set 3044 B, for example, based on having data values of column B bashing to a same value indexed in the probabilistic index structure 3020.B or otherwise being indexed together, despite not all being equal to “world”. Their respective values are read from memory in row storage 3022 via source element 3014.B, and filter element 3016. B automatically removes the false-positive row set 3035.B based on filtering out row d due to having a column A value not equal to “world.” This renders true operandB-satisfying row set 3034.B.
Set anion element 3218 performs a set union upon true operandA-satisfying row set 3034 A and true operandB-satisfying row set 3034 B to render true disjunction satisfying row set 3234.
Another source element for column A is performed to read data values of column A for rows in true disjunction satisfying row set 3234, and/or only for rows in true disjunction satisfying row set 3234 whose data values were not already read and/or not already included in output of the set union based on being previously read and not filtered out. For example, this additional source element is included based on column A values for true disjunction satisfying row set 3234 being required further downstream. The resulting data value set 3047. A includes values of column A. In this case, the resulting data value set 3047.A includes the column A data value for false-positive rows b, which was previously read via the prior source element for column A due to being identified in row identifier set 3044.A. For example, the data value 3024.A.b is re-read via this source element 3014 and included in data value set 3047. A due to row b being included in output of set union element 3218.
Another source element for column B is performed to read data values of column B for rows in true disjunction satisfying row set 3234, and/or only for rows in true disjunction satisfying row set 3234 whose data values were not already read and/or not already included in output of the set union based on being previously read and not filtered out. For example, this additional source element is included based on column B values for true disjunction satisfying row set 3234 being required further downstream. The resulting data value set 3047.B includes values of column B. In this case, the resulting data value set 3047. B includes the column B data value row e, which was not read via the prior source element for row B due to not being identified in row identifier set 3044.B. For example, the data value 3024.B.c is read for the first time via this source element 3014 and included in data value set 3047.B due to row e being included in output of set union element 3218.
In various embodiments, a query processing system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, can cause the query processing system to: determine a query operator execution flow that includes a logical disjunction indicating a first column of a plurality of rows in a first operand and indicating a second column of the plurality of rows in a second operand: and/or facilitate execution of the logical disjunction of the query operator execution flow against the plurality of rows Facilitating exception of the logical disjunction of the query operator execution flow against the plurality of rows can include utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a proper subset of the plurality of rows based on the first operand; filtering the first subset of rows to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of the first subset of rows having first column values of the first column that compare favorably to the first operand, identifying a third subset of rows as a proper subset of the plurality of rows based on identifying rows of the plurality of rows having second column values of the second column that compare favorably to the second operand: and/or identifying a final subset of rows as a union of the second subset of rows and the third subset of rows.
Some or all of the method of
Some or all of the steps of
Step 3282 includes determining a query operator execution flow that includes a logical disjunction indicating a first column of a plurality of rows in a first operand and indicating a second column of the plurality of rows in a second operand. Step 3284 includes facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows.
Performing step 3284 can include performing step 3286. 3288. 3290, and/ot 3292. Step 3286 utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a proper subset of the plurality of rows based on the first operand. Step 3288 includes filtering the first subset of rows to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of the first subset of rows having first column values of the first column that compare favorably to the first operand. Step 3290 includes identifying a third subset of rows as a proper subset of the plurality of rows based on identifying rows of the plurality of rows having second column values of the second column that compare favorably to the second operand. Step 3292 includes identifying a final subset of rows as a union of the second subset of rows and the third subset of rows.
In various embodiments. facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes reading a first set of column values from memory based on reading column values of the first column only for rows in the first subset of rows. Filtering the first subset of rows to identify the second subset of rows can include utilizing the first set of column values. In various embodiments, facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes reading another set of column values from memory based on reading column values of the first column for rows in the final subset of rows as output column values of the logical disjunction. A set difference between the another set of column values and the first set of column values can be non-bull.
In various embodiments, a set difference between the first subset of rows and the second subset of rows is non-null. In various embodiments, a set intersection between the set difference and the final subset of rows is non-null.
In various embodiments, facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes utilizing second index data of a probabilistic indexing scheme for the second column of the plurality of rows to identify a fourth subset of rows as another proper subset of the plurality of rows based on the second operand. The third subset of rows can be identified based on filtering the fourth subset of rows. The third subset of rows can be a subset of the fourth subset of rows. In various embodiments. facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes reading a first set of column values from memory based on reading column values of the first column only for rows in the first subset of rows, where filtering the first subset of rows to identify the second subset of rows includes utilizing the first set of column values. In various embodiments. facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes reading a second set of column values from memory based on reading column values of the second column only for rows in the fourth subset of rows, where filtering the fourth subset of rows to identify the third subset of rows includes utilizing the second set of column values.
In various embodiments, facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes reading a third set of column values from memory based on reading column values of the first column for rows in the final subset of rows as first output column values of the logical disjunction, where a set difference between the third set of column values and the first set of column values is non-pull. In various embodiments, facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows further includes reading a fourth set of column values from memory based on reading column values of the second column for rows in the final subset of rows as second output column values of the logical disjunction, where a set difference between the fourth set of column values and the second set of column values is non-null.
In various embodiments, the second subset of rows and the third subset of rows are identified in parallel via a first set of processing resources and a second set of processing resources, respectively. In various embodiments, the first index data of the probabilistic indexing scheme for the first column includes a first plurality of hash values computed by performing a first hash function on corresponding first column values of the first column. The first subset of rows can be identified based on a first hash value computed for a first value indicated in the first operand. In various embodiments, the second index data of the probabilistic indexing scheme for the second column includes a second plurality of bash values computed by performing a second hash function on corresponding second column values of the second column. The fourth subset of rows can be identified based on a second hash value computed for a second value indicated in the second operand.
In various embodiments, the first operand indicates a first equality condition requiring equality with the first value. The first subset of rows can be identified based on having hash values for the first column equal to the first hash value computed for the first value. In various embodiments, the second operand can indicate a second equality condition requiring equality with the second value. The fourth subset of rows can be identified based on having hash values for the second column equal to the second bash value computed for the second value.
In various embodiments, the final subset of rows includes every row of the plurality of rows with a corresponding first column value of the first column and second column value of the second column comparing favorably to the logical disjunction. In various embodiments, the final subset of rows is a proper subset of a set union of the first subset of rows and the fourth subset of rows. In various embodiments, the probabilistic indexing scheme is an inverted indexing scheme. The first subset of rows can be identified based on index data of the inverted indexing scheme.
In various embodiments, a plurality of column values for the first column are variable-length values. In various embodiments, a plurality of hash values were generated from the plurality of column values for the first column based on the probabilistic indexing scheme for the first column, for example, as the first index data for the first column. The plurality of bash values can be fixed-length values. Identifying the first subset of rows can be based on the plurality of hash values.
In various embodiments, at least one of the first subset of rows having a first column value for the first column that compares unfavorably to the first operand is included in the first subset of rows based on the probabilistic indexing scheme for the first column, In various embodiments, the at least one of the first subset of rows is not included in the second subset of rows based on the first column value for the first column comparing unfavorably to the first operand. In various embodiments, the at least one of the first subset of rows is included in the final subset of rows based on being included in the third subset of rows.
In various embodiments, facilitating execution of the logical disjunction of the query operator execution flow against. The plurality of rows includes applying at least one probabilistic index-based 10 construct of an IO pipeline generated for the query operator execution flow. For example, at least one probabilistic index-based IO construct of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments, a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: determine a query operator execution flow that includes a logical disjunction indicating a first column of a plurality of rows in a first operand and indicating a second column of the plurality of rows in a second operand, and/or facilitate execution of the logical disjunction of the query operator execution flow against the plurality of rows. Facilitating execution of the logical disjunction of the query operator execution flow against the plurality of rows can include utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a proper subset of the plurality of rows based on the first operand; filtering the first subset of rows to identify a second subset of rows as a subset of the first subset of rows based on identifying ones of the first subset of rows having first column values of the first column that compare favorably to the first operand: identifying a third subset of rows as a proper subset of the plurality of rows based on identifying cows of the plurality of rows having second column values of the second column that compare favorably to the second operand: and/or identifying a final subset of rows as a union of the second subset of rows and the third subset of rows.
The logical connective 3312 can indicate a set of operands 3114, which can include at least two operands 3114. Each operand 3 | 14 can involve at least one corresponding column 3023 of the dataset identified via a corresponding one or more column identifiers. In this example, two operands 3114. A and 3114. B are included, where operand 3114. A indicates a first column 3023.A identified by column identifier 3041.A, and operand 3114.B indicates a second column 3023.B identified by column identifier 3041.8 While not illustrated, logical connective 3312 can optionally indicate more than two operands in other embodiments. The operands 3114. A and 3114 B of
The IO pipeline generator module 2834 can generate a corresponding IO pipeline 2835 based on pushing the negation of the logical connective to the IO level as discussed previously. This can include adapting the probabilistic index-based JO construct 3010 of
The probabilistic index-based logical connective negation construct 3310 can alternatively or additionally be considered a type of probabilistic index-based IO construct 3010 specific to implementing predicates 2822 that include negations of logical connectives. The probabilistic index-based logical connective negation construct 3310 can be implemented upstream and/or downstream of other IO constructs of the IO pipeline, such as other IO probabilistic index-based IO constructs 3010 or other source utilize different non-probabilistic indexing schemes, and/or other constructs of the IO pipeline as discussed herein.
Execution of an example probabilistic index-based conjunction negation construct 3311 is illustrated in
This results in identification of a set of row identifier sets 3044 via each probabilistic index element 3012. As each operand 3114 can be treated as a given predicate 2822, each row identifier set 3044. A and 3044.B can be guaranteed to include the true predicate-satisfying row set 3034 satisfying the corresponding operand 3114.A and/or 3148. B, respectively, as discussed previously. Each row identifier set 3044.A and 3044.B may also have false positive rows of corresponding false-positive row sets 3035.A and 3035 B, respectively, that are not distinguishable from the corresponding true operand-satisfying row sets 3034 until the corresponding data values are read and processed as discussed previously.
Each source element 3014 can read rows of the corresponding row identifier set 3044 from row storage 3022, such as from one or more segments, to render a corresponding data value set 3046 as discussed previously. A set intersect element 3319 can be applied these data values sets 3046. A and 3046.B to render an intersect set 3329, which can include identifiers of rows included in both the row identifier set 3044.A and 3044.B. Note that in this example, the set intersect element 3319 can simply implement an intersection based on row identifiers, without processing the sourced data values in this stage. The implementation of a set intersect element 3319 prior to filtering via read data values by filtering element 3016 as illustrated in
Filter element 3016 can be implemented to identify rows satisfying the logical connective based on data values of data value sets 3046.A and 3046.B with row values included in the intersect set 3329. Alternatively or in addition, the implicit implementation of a set intersection via the filtering element 3016 as discussed in conjunction with
For example a function F(data value 3024.A) is based on the operand 3114.A and, for a given data value 3024 of column A for a given row evaluates to either true or false, where the given row only satisfies the operand 3114.A when the function evaluates to true: and a function G(data value 3024.B) is based on the operand 3114.B and, for a given data value 3024 of column B for a given row evaluates to either true or false, where the given row only satisfies the operand 3114B when the function evaluates to true Only ones of the rows included in intersect set 3329 having data values in data value sets 3046. A and 3046. B that satisfy both operands 3114. A and 3148. B are included in a true conjunction satisfying row set 3134 outputted by the filter element 3016. This true conjunction satisfying row set 3134 can be guaranteed to be equivalent to a set intersection between the true operandA-satisfying row set 3034.A and the true operandB-satisfying row set 3034.B. This true conjunction satisfying row set 3134 can be a proper subset of the intersect set 3329 based on the intersect set 3329 including at least one false-positive row of false-positive row set 3035.A or false-positive row set 3035 B.
A set difference element 3308 can be applied to the initial row set 3032 and the true conjunction satisfying row set 3134 to identify the true negated row set 3334. As discussed previously, the initial row set 3032 can correspond to the row set inputted to the probabilistic index-based conjunction negation construct 3311. This initial row set 3032 can correspond to a full row set. such as a set of all rows in a corresponding data set against which a corresponding query is executed against. For example, the initial row set 3032 can be the full set of rows of the dataset when no prior upstream filtering of the full set of rows has been applied to prior operators of the IO pipeline. Alternatively, the initial row set 3032 can be a subset of the full set of rows of the dataset when prior upstream filtering of the full set of rows has already been applied in prior operators of the IO pipeline, and/or when the set difference is against this subset rather than the full set of rows in the operator execution flow 2817.
As illustrated in
Data value set 3347.A can include at least data value included in data value set 3046.A, for example, based on the corresponding row satisfying operandA but not operandB, and thus not being included in the true conjunction satisfying row set 3134, which is mutually exclusive from the true negated row set 3334, thus tendering the corresponding row being included in the true negated row set 3334. In this case, the corresponding data value can be re-read via the subsequent source element for column A based on having been filtered out due to not satisfying operandB and/or can be retrieved from local memory based on having already been read via the prior source element 3014 for column A based on being identified in row identifier set 3044.A.
Data value set 3347.A can include at least data value included in data value set 30-16.A, for example, based on the corresponding row being a false-positive row of false-positive row set 3035.A, and thus not being included in the true conjunction satisfying row set 3134, which is mutually exclusive from the true negated row set 3334, thus rendering the corresponding row being included in the true negated row set 3334. In this case, the corresponding data value can be re-read via the subsequent source element for column A based on having been filtered out due to being a false-positive row for column A and/or can be retrieved from local memory based on having already been read via the prior source element 3014 for column A based on being identified in row identifier set 3044.A.
Data value set 3347. A can include at least data value not included in data value set 3046.A, for example, based on the corresponding row not being identified in row identifier set 3044. A due to not satisfying the operandA or being a false-positive. and thus not being included in the true conjunction satisfying row set 3134, which is mutually exclusive from the true negated row set 3334, thus rendering the corresponding row being included in the true negated row set 3334. In this case, the corresponding data value can be read via the subsequent source element for column A for the first time based on never having been read via the prior source element 3014 for column A.
Execution of an example probabilistic index-based disjunction negation construct 3313 is illustrated in
Each source element 3014 can read rows of the corresponding tow identifier set 3044 from tow storage 3022, such as from one or more segments, to render a corresponding data value set 3046 as discussed previously. A set union element 3218 can be applied these data values sets 3046. A and 3046. B to render a union set 3339, which can include identifiers of rows included in either the row identifier set 3044. A and 3044.B. Note that in this example, the set union element 3218 can simply implement an intersection based on row identifiers prior to filtering out false-positives. The implementation of set union element 3218 prior to filtering via read data values by filtering element 3016 as illustrated in
Filter element 3016 can be implemented to identify rows satisfying the logical connective based on data values of data value sets 3046.A and 3046.B with row values included in the union set 3339. Alternatively or in addition the implementation of filtering elements for each data value set 3046 prior to applying the set union element 3218 as discussed in conjunction with
For example a function F(data value 3024. A) is based on the operand 3114. A and, fora given data value 3024 of column A for a given row evaluates to either true or false, where the given row only satisfies the operand 3114.A when the function evaluates to true; and a function G(data value 3024. B) is based on the operand 3114. B and, for a given data value 3024 of column B fora given row evaluates to either true or false, where the given row only satisfies the operand 3114B when the function evaluates to true. Only ones of the rows included in intersect set 3329 having data values in data value sets 3046. A and 3046. B that satisfy either operands 3114. A or 3148.B are included in a true disjunction satisfying row set 3234 outputted by the filter element 3016. This true disjunction satisfying row set 3234 can be guaranteed to be equivalent to a set union between the true operand A-satisfying row set 3034 A and the true operandB-satisfying row set 3034 B. This true conjunction satisfying row set 3134 can be a proper subset of the union set 3339 based on the union set 3339 including at least one false-positive row of false-positive row set 3035.A or false-positive row set 3035.B.
A set difference element 3308 can be applied to the initial row set 3032 and the true disjunction satisfying row set 3234 to identify the true negated row set 3334. As discussed previously, the initial row set 3032 can correspond to the row set inputted 10 the probabilistic index-based conjunction negation construct 3311. This initial row set 3032 can correspond to a full row set, such as a set of all rows in a corresponding data set against which a corresponding query is executed against. For example, the initial row set 3032 can be the full set of rows of the dataset when no prior upstream filtering of the full set of rows has been applied in prior operators of the IO pipeline. Alternatively, the initial row set 3032 can be a subset of the full set of rows of the dataset when prior upstream filtering of the full set of rows has already been applied in prior operators of the IO pipeline, and/or when the set difference is against this subset rather than the full set of rows in the operator execution flow 2817.
As illustrated in
Data value set 3347.A can include at least data value included in data value set 3046.A, for example, based on the corresponding row being a false-positive row of false-positive row set 3035.A and also not satisfying operandB, and thus not being included in the true disjunction satisfying row set 3234, which is mutually exclusive from the true negated row set 3334, thus rendering the corresponding row being included in the true negated row set 3334. In this case, the corresponding data value can be re-read via the subsequent source element for column A based on having been filtered out due to being a false-positive row for column A and due to the row also not satisfying operandB, and/or can be retrieved from local memory based on having already been read via the prior source element 3014 for column A based on being identified in row identifier set 3044.A.
Data value set 3347.A can include at least data value not included in data value set 3046. A, for example, based on the corresponding row not being identified in row identifier set 3044. A due to not satisfying the operandA or being a false-positive. and based on operandB for the row also not being satisfied and the row thus not being included in the true disjunction satisfying row set 3234, which is mutually exclusive from the true negated row set 3334, thus rendering the corresponding row being included in the true negated row set 3334. In this case, the corresponding data value can be read via the subsequent source element for column A for the first time based on never having been read via the prior source element 3014 for column A.
In various embodiments a query processing system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processer, can cause the query processing system to: determine a query operator execution flow that includes a negation of a logical connective indicating a first column of a plurality of rows in a first operand of the logical connective and indicating a second column of the plurality of rows in a second operand of the logical connective: and/or facilitate execution of the negation of the logical connective of the query operator execution flow against the plurality of rows. Facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows can include: utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a first proper subset of a set of rows of the plurality of rows based on the first operand: utilizing second index data of a probabilistic indexing scheme for the second column of the plurality of rows to identify a second subset of rows as a second proper subset of the set of rows based on the second operand: applying a set operation upon the first subset of rows and the second subset of rows based on a logical operator of the logical connective to identify a third subset of rows from the set of rows: filtering the third subset of rows to identify a fourth subset of rows based on comparing first column values and second column values of the third subset of rows to the first operand and the second operand, and/or identifying a final subset of rows as a set difference of the fourth subset of rows and the set of rows based on the negation.
Some or all of the method of
Some or all of the steps of
Step 3382 includes determining a query operator execution flow that includes a negation of a logical connective indicating a first column of a plurality of rows in a first operand of the logical connective and indicating a second column of the plurality of rows in a second operand of the logical connective. Step 3384 includes facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows.
Performing step 3384 can include performing step 3386. 3388. 3390. 3392, and/or 3394. Step 3386 includes utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a first proper subset of a set of rows of the plurality of rows based on the first operand. Step 3388 includes utilizing second index data of a probabilistic indexing scheme for the second column of the plurality of rows to identify a second subset of rows as a second proper subset of the set of rows based on the second operand. Step 3390 includes applying a set operation upon the first subset of rows and the second subset of rows based on a logical operator of the logical connective to identify a third subset of rows from the set of rows. Step 3392 includes filtering the third subset of rows to identify a fourth subset of rows based on comparing first column values and second column values of the third subset of rows to the first operand and the second operand. Step 3394 includes identifying a final subset of rows as a set difference of the fourth subset of rows and the set of rows based on the negation.
In various embodiments, the set of rows is a proper subset of the plurality of rows identified based on at least one prior operator of the query operator execution flow. In various embodiments, the set of rows is the plurality of rows. Alternatively, the set of rows can be a proper subset of the plurality of rows.
In various embodiments facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows further includes reading a first set of column values from memory based on reading column values of the first column only for rows in the first subset of rows. Filtering the third subset of rows to identify the fourth subset of rows can include utilizing the ones of the first set of column values for rows in the third subset of rows. In various embodiments facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows further includes reading a second set of column values from memory based on reading column values of the second column only for rows in the second subset of rows. Filtering the third subset of rows to identify the fourth subset of rows can further include utilizing the ones of the second set of column values for rows in the third subset of rows.
In various embodiments. facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows further includes reading a third set of column values from memory based on reading column values of the first column for rows in the final subset of rows as first output column values of the negation of the logical connective. An intersection between the third set of column values and the first set of column values can be non-null. In various embodiments facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows further includes reading a fourth set of column values from memory based on reading column values of the second column for rows in the final subset of rows as second output column values of the negation of the logical connective. An intersection between the fourth set of column values and the second set of column values can be non-pull.
In various embodiments, the set operation is an intersection operation based on the logical connective including a logical conjunction Filtering the third subset of rows can include identifying ones of the third subset of rows with first column values comparing favorably to the first operand and second column values comparing favorably to the second operand.
In various embodiments, the set operation is a union operation based on the logical connective including a logical disjunction. Filtering the third subset of rows includes identifying ones of the third subset of rows with either first column values comparing favorably to the first operand or second column values comparing favorably to the second operand.
In various embodiments, a set difference between the third subset of rows and the fourth subset of rows includes at least one row based on: the at least one row having a first column values comparing unfavorably to the first operand and being identified in the first subset of rows based on the probabilistic indexing scheme for the first column, and/or the at least one row having a second column values comparing unfavorably to the second operand and being identified in the second subset of rows based on the probabilistic indexing scheme for the second column. In various embodiments, an intersection between the third subset of rows and the final subset of rows includes at least que row based on: the at least one row having a first column values comparing unfavorably to the first operand and being identified in the first subset of rows based on the probabilistic indexing scheme for the first column, and/or the at least one row having a second column values comparing unfavorably to the second operand and being identified in the second subset of rows based on the probabilistic indexing scheme for the second column.
In various embodiments, the fourth subset of rows includes every row of the plurality of rows with a corresponding first column value of the first column and second column value of the second column comparing favorably to the logical connective. The fourth subset of rows can be a proper subset of the third subset of rows. In various embodiments, the first subset of rows and the second subset of rows are identified in parallel via a first set of processing resources and a second set of processing resources. respectively.
In various embodiments, the first index data of the probabilistic indexing scheme for the first column includes a first plurality of hast values computed by performing a first bash function on corresponding first column values of the first column. The first subset of rows can be identified based on a first hash value computed for a first value indicated in the first operand. In various embodiments, the second index data of the probabilistic indexing scheme for the second column includes a second plurality of hash values computed by performing a second hash function on corresponding second column values of the second column. The second subset of rows can be identified based on a second hash value computed for a second value indicated in the second operand.
In various embodiments, the first operand indicates a first equality condition requiring equality with the first value. The first subset of rows can be identified based on having bash values for the first column equal to the first hash value computed for the first value. In various embodiments, the second operand indicates a second equality condition requiring equality with the second value. The second subset of rows can be identified based on having hash values for the second column equal to the second bash value computed for the second value.
In various embodiments, the probabilistic indexing scheme for the first column is an invented indexing scheme. The first subset of rows can be identified based on index data of the invented indexing scheme. In various embodiments a plurality of column values for the first column are variable-length values. In various embodiments a plurality of bash values were generated front the plurality of column values for the first column based on the probabilistic indexing scheme in various embodiments, the plurality of hash values are fixed-length values. Identifying the first subset of rows can be based on the plurality of hash values.
In various embodiments, at least one of the first subset of rows having a first column value for the first column that compares unfavorably to the first operand is included in the first subset of rows based on the probabilistic indexing scheme for the first column. In various embodiments, the at least one of the first subset of rows is not included in the fourth subset of rows based on the first column value for the first column comparing unfavorably to the first operand. In various embodiments, the at least one of the first subset of rows is included in the final subset of rows based on being included in the second subset of rows.
In various embodiments, facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows includes applying at least one probabilistic index-based IO construct of an IO pipeline generated for the query operator execution flow. For example, at least one probabilistic index-based IO construct of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that Stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: determine a query operator execution flow that includes a negation of a logical connective indicating a first column of a plurality of rows in a first operand of the logical connective and indicating a second column of the plurality of rows in a second operand of the logical connective; and/or facilitate execution of the negation of the logical connective of the query operator execution flow against the plurality of rows Facilitating execution of the negation of the logical connective of the query operator execution flow against the plurality of rows can include: utilizing first index data of a probabilistic indexing scheme for the first column of the plurality of rows to identify a first subset of rows as a first proper subset of a set of rows of the plurality of rows based on the first operand; utilizing second index data of a probabilistic indexing scheme for the second column of the plurality of rows to identify a second subset of rows as a second proper subset of the set of rows based on the second operand: applying a set operation upon the first subset of rows and the second subset of rows based on a logical operator of the logical connective to identify a third subset of rows from the set of tows, filtering the third subset of rows to identify a fourth subset of rows based on comparing first column values and second column values of the third subset of rows to the first operand and the second operand: and/or identifying a final subset of rows as a set difference of the fourth subset of rows and the set of rows based on the negation.
Thus a set of row identifiers identified for a given fixed-length value generated for the given value is guaranteed to include all rows with variable-length values matching or otherwise comparing favorably to the given value, with the possibility of also including false-positive rows. The variable-length data values of these identified rows can be read from memory, and can each be compared to the given value to identify ones of the rows with variable-length values comparing favorably to the given value filtering out the false positives. For example, each variable-length data values of the identified rows, once read from memory, are tested for equality with the given value to render a true output set of rows that is guaranteed to include all rows with variable-length values equal to the given value, and that is further guaranteed to include no rows with variable-length values not equal to the given value.
These steps can be implemented by utilizing some or all properties of the IO pipeline constructs of
As illustrated in
The equality condition 3422 can indicate a column identifier 3041 of a variable-length column 3023, such as a column storing text data or other data having variable-lengths and/or having unstructured data. The equality condition 3422 can further indicate a literal value 3448, such as particular text value or other variable-length value for comparison with values in the column. Thus, a true set of rows satisfying equality condition 3422 can correspond to all rows with data values in the column 3023 denoted by column identifier 3041 that are equivalent to literal value 3448.
An IO pipeline can be generated via IO pipeline generator module 2834, for example, as discussed in conjunction with
The IO pipeline generator module 2834 can determine a fixed-length value 3458 for utilization to probe a probabilistic index structure 3020 for the variable-length column based on performing a fixed-length conversion function 3450 upon the literal value 3448 of the equality condition 3422. For example, the fixed-length conversion function 3450 can be a hash function applied to the literal value 3448, where the fixed-length value 3458 is a hash value. The fixed-length conversion function 3450 can correspond to a function utilized to index the variable-length column via a corresponding probabilistic indexing scheme.
The corresponding IO pipeline can include a probabilistic index element 3012, where the index probe parameter data 3042 is implemented to indicate the column identifier for the variable-length column and the fixed-length value 3458 generated for the literal value via the fixed-length value 3458. A source element 3014 can be applied downstream from the probabilistic index element to source variable-length data values of the column denoted by the column identifier 3041 for only the rows indicated in output of the probabilistic index element. A fillet element 3016 can be applied downstream from the source element 301410 compare the read data values to the literal value 3448 to identify which ones of the rows with data values are equivalent to the literal value. filtering out other ones of the rows with data values that are not equivalent to the literal value as false-positive rows identified due to the probabilistic nature of the probabilistic indexing scheme.
These elements of the IO pipeline 2835 can be implemented as a probabilistic index-based IO construct 3010 of
A fixed-length conversion function 3450 can be performed upon data values 3024 of the given column to determine a corresponding index value 3043 for each data value, rendering a fixed-length value mapping 3462 indicating the index value 3043 for each data value 3024. This fixed-length value mapping 3462 can be utilized to generate a probabilistic index structure 3020 via a probabilistic index structure generator module 3470. The resulting probabilistic index structure 3020 can indicate, for each given index value, ones of the set of tows, such as row numbers, memory locations, or other row identifiers of these rows, having data values 3024 for the given column that map to this given fixed-length value. For example, this probabilistic index structure 3020 is implemented as an inverted index structure mapping the fixed-length index values, such as bash values, to respective rows.
In some embodiments, the resulting probabilistic index structure 3020 can be stored as index data, such as a secondary index 2546, of a corresponding segment having the set of rows for the given column. Other sets of rows of a given dataset that are included in different segments can similarly have their rows indexed via the same type of probabilistic index structure 3020 via the same or different fixed-length conversion function 3450 performed upon data values of its columns In some cases, different fixed-length conversion functions 3450 are selected for performance for sets of rows of different segments, for example, based on different cardinality, different access frequency, different query types, or other different properties of the column data for different segments. In some embodiments, a false-positive rate induced by the fixed-length conversion function 3450 is selected as a false-positive tuning parameter, where the false-positive tuning parameter is selected differently for different segments based on user input and/or automatic determination. Configuration of false-positive rate is discussed in further detail in conjunction with
In other embodiments, the resulting probabilistic index structure 3020 can be stored as index data, such as a secondary index 2546, for all rows of the given dataset in one or more locations. For example a common probabilistic index structure 3020 can be generated for all rows of a dataset, even if these rows are stored across different segments, different storage structures. and/or different memory locations.
In this example, the values “bello” and “bine” map to a same index value 3043.1, and the value “planet” maps to a different index value 3043.1. For example, the fixed-length conversion function 3450 is a hash function that, when performed upon “hello” renders a same hash value as when performed upon “blue”, which is different from the hash value outputted when performed upon “planet.” While this simple example is presented for illustrative purposes, much larger text data can be implemented as data values 3024 in other embodiments. The number Z of index values 3043 in the probabilistic index structure 3020 can be a large number, such as thousands of different index values.
The probabilistic index structure 3020 of
Index access 3452 is performed to utilize fixed-length value 3458.1 to identify a corresponding row identifier set 3044.i based on probabilistic index structure 3020. For example, the fixed-length value 3458, i is determined to be equal to index value 3043.1, and the row identifier set 3044.i is determined based on being mapped to index value 3043.1 via probabilistic index structure 3020. A as discussed in conjunction with
Data value access 3454 is performed to read rows identified in row identifier set 3044.i from row storage 3022, such as rows stored in a corresponding one or more segments. A dam value set 3046 that includes the corresponding data values 3024 for rows identified in row identifier set 3044 is identified accordingly. The data value access 3454 performed by query processing system 2802 can be implemented as source element 3014 of a corresponding IO pipeline 2834, and/or can otherwise be performed via other processing performed by a query processing system: 2802 executing a corresponding query against a dataset.
Equality-based filtering 3459 is performed by determining ones of the data value set 3046 equal to the given literal value “hello” to render a row identifier subset 3045, and/or optionally a corresponding subset of data values 3024 of data value set 3046. This can be based on comparing each data value 3024 in data value set 3046 to the given literal value, and including only ones of row identifiers in row identifier set 3044 with corresponding ones of the set of data values 3024 in data value set 3046 that are equivalent to the literal value. In this case rows a, e, and f are included based on having data values 3024 of “hello”, while rows b and d are filtered out based on being false-positive rows with values of “blue” that were indexed to the same index value. The equality-based filtering 3459 performed by query processing system 2802 can be implemented as filtering element 3016 of a corresponding IO pipeline 2834, and/or can otherwise be performed via other processing performed by a query processing system 2802 executing a corresponding query against a dataset.
Applying a probabilistic index, such as an inverted index, in this fashion to variable-length columns, such as varchar columns, can reduce the size of the index data being stored as fixed-length values are stored. In particular, a number of fixed length values Z are generated and stored, where Z is smaller than the number of columns X due to the hash collision or otherwise probabilistic nature of the index. Furthermore, the size of each fixed length value can be smaller than most and/or all corresponding variable length data, such as length text data, of the corresponding variable length column. Thus, the probabilistic index structure 3020 is relatively inexpensive to store, and can be comparable in size to the index structures of fixed-length data. Furthermore, the use of the probabilistic index structure 3020 for variable length data only induces only a minor increase in processing relative to identifying only the true rows via a true index, as only a small number of additional false positive rows may be expected to be read and/or filtered from memory, relative to the IO requirements that would be necessitated if all data values needed to be read in the case where no indexing scheme was utilized due to the column including variable-length values. The reduction in IO cost for variable length: data via storage of an index comparable to indexes of fixed-length columns improves the technology of database systems by efficiently utilizing memory resources to index variable length data to improve the efficiency of reading variable length data.
The size of the fixed-length index values outputted by the fixed-length conversion function 3450 to generate the probabilistic index structure can be tuned to increase and/or reduce the rate of false positives. As the rate of false positives increases. increasing the IO cost in performing query executions, the corresponding storage cost of the probabilistic index structure 3020 as a whole can decrease. In particular, in the case of a hash function, increasing the number of hash values and/or fixed-length of the bash values increases the storage cost of the probabilistic index structure 3020, while reducing the rate of bash collisions and thus reducing the IO cost as less false-positives need be read and filtered in query executions. Configuration of this trade-off between IO cost and index storage cost via selection of a false-positive fuming parameter, such as the fixed-length of the bash values, is discussed in further detail in conjunction with
In various embodiments a query processing system includes at least one processer and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, can cause the query processing system to: identify a filtered subset of a plurality of rows having variable-length data of a column equal to a given value. Identify the filtered subset of the plurality of rows having variable-length data of the column equal to the given value can be based on: identifying a first subset of rows as a proper subset of the plurality of rows based on a plurality of fixed-length index values of the column: and/or comparing the variable-length data of only rows in the first subset of rows to the given value to identify the filtered subset as a subset of the first subset of rows.
Some or all of the method of
Some or all of the method of
Some or all of the steps of
Step 3482 includes storing a plurality of variable-length data of a column of a plurality of rows. Step 3484 includes storing a plurality of fixed-length index values of a probabilistic indexing scheme for the column. Step 3486 includes identifying a filtered subset of the plurality of rows having variable-length data of the column equal to a given value.
Performing step 3486 can include performing step 3488 and/or 3490 Step 3488 includes identifying a first subset of rows as a proper subset of the plurality of rows based on the plurality of fixed-length index values. Step 3490 includes comparing the variable-length data of only rows in the first subset of rows to the given value to identify the filtered subset as a subset of the first subset of rows.
In various embodiments, identifying the filtered subset of the plurality of rows is further based on reading a set of variable-length data based on reading the variable-length data from only rows in the first subset of rows. Comparing the variable-length data of only the rows in the first subset of rows to the given value can be based on utilizing only variable-length data in the set of variable-length data.
In various embodiments, the variable-length data is implemented via a string datatype, a varchar datatype, a text datatype. or other variable-length datatype. In various embodiments a set difference between the filtered subset and the first subset of rows is non-null. In various embodiments, the probabilistic indexing scheme for the column is an inverted indexing scheme. The first subset of rows can be identified based on inverted index values of the inverted indexing scheme.
In various embodiments, the plurality of fixed-length index values of the probabilistic indexing scheme are a plurality of hash values computed by performing a hash function on corresponding variable-length data of the column. In various embodiments, identifying the filtered subset of the plurality of rows includes computing a first hash value for the given value and/or identifying ones of the plurality of rows having corresponding ones of the plurality of hash value equal to the first hash value. In various embodiments a set difference between the first subset of rows and the filtered subset includes ones of the plurality of rows with variable-length data of the column having hash collisions with the given value.
In various embodiments, the fixed-length is based on a false-positive tuning parameter of the hash function. A first number of rows included in the first subset of rows can be based on the false-positive tuning parameter of the bash function. A second number of rows included in a set difference between the first subset of rows and the filtered subset can be based on the tuning parameter of the hash function. In various embodiments, the method further includes determining the false-positive tuning parameter as a selected false-positive tuning parameter from a plurality of false-positive tuning parameter options.
In various embodiments, identifying the filtered subset of the plurality of rows includes applying at least one probabilistic index-based IO construct of an IO pipeline generated for a query indicating the given value in at least one query predicate. For example, at least one probabilistic index-based IO construct of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: store variable-length data of a column of a plurality of rows: store a plurality of fixed-length index values of a probabilistic indexing scheme for the column; and/or identify a filtered subset of the plurality of rows having variable-length data of the column equal to a given value. Identifying the filtered subset of the plurality of rows can be based on: identifying a first subset of rows as a proper subset of the plurality of rows based on the plurality of fixed-length index values: and/or comparing the variable-length data of only rows in the first subset of rows to the given value to identify the filtered subset as a subset of the first subset of rows.
While the set of rows identified for each n-gram can be guaranteed to be the true set of rows rather than being probabilistic in nature, possible false-positive rows may be inherently present in the resulting intersection based on ordering not being considered when applying the intersection. These false-positives can thus be filtered out via reading and filtering of the text data of the identified rows in the intersection to identify only rows with text data having the n-grams in the appropriate ordering as dictated by the given text pattern. Such searches for inclusion of a text pattern can thus be implemented by leveraging techniques of the probabilistic index-based constructs described previously, despite the index structure not necessarily indexing the n-grams of text data in a probabilistic fashion.
As illustrated in
The text inclusion condition 3522 can indicate a column identifier 3041 of a column 3023, such as the column variable-length column 3023 of
An IO pipeline can be generated via IO pipeline generator module 2834, for example, as discussed in conjunction with
The IO pipeline generator module 2834 can determine a substring set 3552 for utilization to probe an index structure for the column based on performing a substring generator function 3550 upon the consecutive text pattern 3548 of the text inclusion condition 3522. For example, the text inclusion condition 3522 can generate substrings 3554.1-3554.R as all substrings of the consecutive text pattern 3548 of a given fixed-length 3551, such as the value n of a corresponding set of n-grams implementing the substring set 3552. The fixed-length 3551 can be predetermined and can correspond to a fixed-length 3551 utilized to index the text data via a subset-based index structure as described in further detail in conjunction with
In cases where the consecutive text pattern 3548 includes wildcard characters or other indications of breaks between words and/or portions of the pattern, these wildcard characters can be skipped and/or ignored in generating the substrings of the substring set. For example a consecutive text pattern 3548 having one or more wildcard characters can render a substring set 3552 with no substrings 3554 that include wildcard characters.
The corresponding IO pipeline can include a plurality of R parallel index elements 3512 that each correspond to one of the R substrings 3554.1-3554.R of the substring set 3552. Each index element 3512 can be utilized to identify ones of the rows having text data in the column identified by the text column identifier that includes the substring based on a corresponding substring-based index structure. A set intersect element can be applied to the output of the R parallel index elements 3512 to identify rows having all of the substrings 3554.1-3554.R, in any order.
This plurality of R parallel index elements 3512 and set intersect element 3319 can be collectively considered a probabilistic index element 3012 of
These steps can be applied as source element 3014 and filter element 3016 accordingly, and the entire process can thus be considered an adapted implementation of the probabilistic index-based IO construct 3010 of
A substring generator function 3550 can be performed upon data values 3024 of the given column to determine a corresponding substring set 3552 for each data value, rendering a substring mapping 3562 indicating the substring set 3552 of one or more substrings for each data value 3024. Each substring can correspond to an index value 3043, where a given row is indexed via multiple index values based on its text value including multiple corresponding substrings. The fixed-length 3551 of the substring generator function 3550 utilized to build the corresponding substring-based index structure 3570 can dictate the fixed-length 3551 of the substring generator function 3550 performed by the IO pipeline generator module 2834 of
This substring mapping 3562 can be utilized to generate a substring-based index structure 3570 via an index structure generator module 3560. The resulting substring-based index structure 3570 can indicate, for each given substring, ones of the set of rows, such as row numbers, memory locations, or other row identifiers of these rows, having data values 3024 for the given column corresponding to text data that includes the given substring. For example, this substring-based index structure 3570 is implemented as an inverted index structure mapping the substrings as index values 3043 to respective rows.
In some embodiments, the resulting substring-based index structure 3570 can be stored as index data, such as a secondary index 2546, of a corresponding segment having the set of rows for the given column. Other sets of rows of a given dataset that are included in different segments can similarly have their rows indexed via the same type of substring-based index structure 3570 via the same or different fixed-length 3551 performed upon data values of its columns. In some cases, different substring generator functions 3550 are selected for performance for sets of rows of different segments, for example, based on different cardinality. different access frequency, different query types, or other different properties of the column data for different segments. In some embodiments, a false-positive rate induced by the fixed-length 3551 is selected as a false-positive tuning parameter, where the false-positive tuning parameter is optionally selected differently for different segments based on user input and/or automatic determination. Configuration of false-positive rate is discussed in further detail in conjunction with
In other embodiments, the resulting substring-based index structure 3570 can be stored as index data, such as a secondary index 2546, for all rows of the given dataset in one or more locations. For example a common substring-based index structure 3570 can be generated for all rows of a dataset, even if these rows are stored across different segments, different storage structures. and/or different memory locations.
The substring-based index structure 3570 can be considered a type of probabilistic index structure 3020 as a result of rows being identified for inclusion of subsets of a consecutive text pattern that may not include the consecutive text pattern. However, unlike the example probabilistic index structure of
The substring-based index structure 3570 of
In some embodiments a given column storing text data, such as a given column 3023.A, can be indexed via both the probabilistic index structure 3020 of
Selection of whether to index a given column of text data via the probabilistic index structure 3020 of
A set of corresponding index accesses 3542. 1, 3542.2, and 3542.3 are performed to utilize each corresponding substring 3554 to identify each of a corresponding set of row identifier sets 3044 based on substring-based index structure 3570. This can include probing the substring-based index structure 3570 for index values corresponding to the substrings in the substring set. For example, the row identifier set 3044.6 is determined via index access 3542.1 based on being mapped to the index value 3043 for “red”, the row identifier set 3044.2 is determined via index access 3542.2 based on being mapped to the index value 3043 for “bea”; and the row identifier set 3044.4 is determined via index access 3542.3 based on being mapped to the index value 3043 for “ear”. The index accesses can be optionally performed in parallel, for example, via parallel processing resources, such as a set of distinct nodes and/or processing core resources. Each index access 3452 performed by query processing system 2802 can be implemented as an index element 3512 of a corresponding IO pipeline 2834 as illustrated in
An intersect subset 3544 can be generated based on performing a set intersection open the outputted row identifier sets 3044 of the index accesses 3542 via a set intersect element 3319. The intersect subset 3544 in this example includes row a and row c, indicating that rows a and row c include all substrings “red”. “bea”, and “ear”. The intersect subset 3544 can be implemented as a row identifier set 3044 of embodiments of
Data value access 3454 is performed to read rows identified in intersect subset 3544 from row storage 3022, such as rows stored in a corresponding one or more segments. A data value set 3046 that includes the corresponding data values 3024 for rows identified in intersect subset 3544 is identified accordingly. The data value access 3454 performed by query processing system 2802 can be implemented as source element 3014 of a corresponding IO pipeline 2834, and/or can otherwise be performed via other processing performed by a query processing system 2802 executing a corresponding query against a dataset.
Inclusion-based filtering 3558 is performed by determining ones of the data value set 3046 that include the consecutive text pattern “red % bear” to render a row identifier subset 3045, and/or optionally a corresponding subset of data values 3024 of data value set 3046. This can be based on comparing each data vale 3024 in data value set 3046 to the given consecutive text pattern 3548, and including only ones of row identifiers in row identifier set 3044 with corresponding ones of the set of data values 3024 in data value set 3046 that include the consecutive text pattern 3548. In this case row a is included based on having a data value 3024 of “binge red bear” that includes the text pattern “red % bear” while row c is filtered out based on being false-positive rows with a value of “bear red” that does not match the text pattern due to including all substrings in a wrong ordering not matching the given text pattern. The inclusion-based filtering 3558 performed by query processing system 2802 can be implemented as filtering element 3016 of a corresponding IO pipeline 2834, and/or can otherwise be performed via other processing performed by a query processing system 2802 executing a corresponding query against a dataset.
Note that if the consecutive text pattern 3548 is a pattern, such as a string literal, with length less than or equal to the fixed-length 3551, the filtering element need not be applied. A plurality of index accesses 3452 may still be necessary to probe for all possible substrings that include the given pattern. However, a set union, rather than a set intersection can be applied to the output of row identifiers identified via this plurality of index accesses 3452.
In various embodiments a query processing system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, can cause the query processing system to identify a filtered subset of a plurality of rows having text data of a column of the plurality of rows that includes a consecutive text pattern. Identifying the filtered subset of the plurality of rows having text data of the column of the plurality of rows that includes. The consecutive text pattern can be based on: identifying a set of substrings included in the consecutive text pattern, identifying a set of subsets of rows by utilizing the index data of the column to identify, for each substring of the set of substrings, a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings; identifying a first subset of rows as an intersection of the set of subsets of rows: and/or comparing the text data of only rows in the first subset of rows to the consecutive text pattern to identify the filtered subset as a subset of the first subset of rows that includes rows having text data that includes the consecutive text pattern.
Some or all of the method of
Some or all of the method of
Some or all of the steps of
Step 3582 includes storing a plurality of text data as a column of a plurality of rows. Step 3584 includes storing index data corresponding to the column indicating, for each given substring of a plurality of substrings having a same fixed-length, ones of the plurality of rows with text data that include the given substring of the plurality of substrings. Step 3586 includes identifying a filtered subset of the plurality of rows having text data of the column that includes a consecutive text pattern.
Performing step 3586 can include performing step 3588, 3590, 3592, and/or 3594. Step 3588 includes identifying a set of substrings included in the consecutive text pattern. Each substring of the set of substrings can have the same fixed-length as substrings of the plurality of substrings. Step 3590 includes identifying a set of subsets of rows by utilizing the index data to identify, for each substring of the set of substrings, a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings. Step 3592 includes identifying a first subset of rows as an intersection of the set of subsets of rows. Step 3594 includes comparing the text data of only rows in the first subset of rows to the consecutive text pattern to identify the filtered subset as a subset of the first subset of rows that includes rows having text data that includes the consecutive text pattern.
In various embodiments, identifying the filtered subset of the plurality of rows is further based on reading a set of text data based on reading the text data from only rows in the first subset of rows Comparing the text data of only the rows in the first subset of rows to the consecutive text pattern can be based on utilizing only text data in the set of text data.
In various embodiments, the text data is implemented via a string datatype, a varchar datatype, a text datatype, a variable-length datatype, or another datatype operable to include and/or depict text data.
In various embodiments a set difference between the filtered subset and the first subset of rows is non-null. In various embodiments, the set difference includes at least one row having text data that includes every one of the set of substrings in a different arrangement than an arrangement dictated by the consecutive text pattern in various embodiments, the index data for the column is in accordance with an inverted indexing scheme In various embodiments, each subset of the set of subsets is identified in parallel with other subset of the set of subsets via a corresponding set of parallelized processing resources.
In various embodiments, the text data for at least one row in the filtered subset has a first length greater than a second length of the consecutive text pattern. In various embodiments, the consecutive text pattern includes at least one wildcard character.
Identifying the set of substrings can be based on skipping the at least one wildcard character In various embodiments, each of the set of substrings includes no wildcard characters.
In various embodiments, the method includes determining the same fixed-length for the plurality of substrings as a selected fixed-length parameter from a plurality of fixed-length options. For example, the selected fixed-length parameter is automatically selected or is selected based on user input. In various embodiments, each of the plurality of substrings include exactly three characters. In various embodiments, identifying the set of substrings included in the consecutive text pattern includes identifying every possible substring of the same-fixed length included in the consecutive text pattern.
In various embodiments, the index data corresponding to the column further indicates, for each row in the plurality of rows a corresponding set of substrings for the text data of the row. In various embodiments, the corresponding set of substrings for the text data of the each row includes every possible substring of the same-fixed length included in the text data.
In various embodiments, identifying the filtered subset includes applying at least one probabilistic index-based IO construct of an IO pipeline generated for a query indicating the consecutive text pattern in at least one query predicate. For example. at least one probabilistic index-based IO construct of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to: store a plurality of text data as a column of a plurality of rows; store index data corresponding to the column indicating, for each substring of a plurality of substrings having a same fixed-length, ones of the plurality of rows with text data that include the each substring of the plurality of substrings: and/or identify a filtered subset of a plurality of rows having text data of a column of the plurality of rows that includes a consecutive text pattern. Identifying the filtered subset of the plurality of rows having text data of the column of the plurality of rows that includes the consecutive text pattern can be based on: identifying a set of substrings included in the consecutive text pattern: identifying a set of subsets of rows by utilizing the index data of the column to identify, for each substring of the set of substrings, a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings; identifying a first subset of rows as an intersection of the set of subsets of rows: and/or comparing the text data of only rows in the first subset of rows to the consecutive text pattern to identify the filtered subset as a subset of the first subset of rows that includes rows having text data that includes the consecutive text pattern.
A given text pattern can be split into a plurality of substrings. Unlike the substrings generated for the text pattern as illustrated in
Each of these non-overlapping substrings can be utilized to identify corresponding rows with text data that includes the given non-overlapping substring, based on the suffix-based index. A set intersection can be applied to the set of outputs to identify rows with all of the non-overlapping substrings of the text pattern.
While the set of rows identified for each non-overlapping substring can be guaranteed to be the true set of rows rather than being probabilistic in nature, possible false-positive rows may be inherently present in the resulting intersection based on ordering not being considered when applying the intersection. These false-positives can thus be filtered out via reading and filtering of the text data of the identified rows in the intersection to identify only rows with text data having the non-overlapping substrings in the appropriate ordering as dictated by the given text pattern. Such searches for inclusion of a text pattern can thus be implemented by leveraging techniques of the probabilistic index-based constructs described previously, despite the index structure not necessarily indexing the text data via suffix-based indexing in a probabilistic fashion.
As illustrated in
An IO pipeline can be generated via IO pipeline generator module 2834, for example, as discussed in conjunction with
The IO pipeline generator module 2834 can determine a substring set 3652 for utilization to probe an index structure for the column based on performing a substring generator function 3650 upon the consecutive text pattern 3548 of the text inclusion condition 3522. For example, the text inclusion condition 3522 can generate substrings 3654.1-3654. R as a set of non-overlapping substrings of the consecutive text pattern 3548 split at a plurality of split points.
In cases where the consecutive text pattern 3548 includes wildcard characters or other indications of breaks between words and/or portions of the pattern, these wildcard characters can be skipped and/or ignored in generating the substrings of the substring set. For example a consecutive text pattern 3548 having one or more wildcard characters can render a substring set 3652 with no substrings 3654 that include wildcard characters.
The plurality of split points can optionally be dictated by a split parameter 3651 denoting where these split points be located. For example, the split parameter 3651 denotes that split points occur at wildcard characters of the consecutive text pattern 3548, and that these wildcard characters not be included in any of the non-overlapping substrings. As another example, the spot parameter 3651 denotes that split points be breaks between distinct words of the consecutive text pattern that includes a plurality of words. A particular ordered combination of the non-overlapping substrings can collectively include all of the consecutive text pattern 3548, and/or can include all of the consecutive text pattern 3548 except for characters, such as wildcard characters and/or breaks between words, utilized as the plurality of split points. The split parameter 3651 can correspond to a split parameter 3651 utilized to index the text data via a suffix-based index stricture as described in further detail in conjunction with
The corresponding IO pipeline can include a plurality of R parallel index elements 3512 that each correspond to one of the R substrings 3654. 1-3654 R of the substring set 3652 Each index element 3512 can be utilized to identify ones of the rows having text data in the column identified by the text column identifier that includes the substring based on a corresponding suffix-based index structure. A set intersect element can be applied to the output of the R parallel index elements 3512 to identify rows having all of the substrings 3654.1-3654.R, in any order.
This plurality of R parallel index elements 3512 and set intersect element 3319 can be collectively considered a probabilistic index element 3012 of
These steps can be applied as source element 3014 and filter element 3016 accordingly, and the entire process can thus be considered au adapted implementation of the probabilistic index-based IO construct 3010 of
Generating the suffix-based index structure 3670 can optionally include performing the substring generator function 3650 upon data values 3024 of the given column to determine a corresponding substring set 3652 of non-overlapping substrings. such as a plurality of distinct words, for each data value. This can optionally render a substring mapping indicating the substring set 3652 of one or more non-overlapping substrings, such as words, for each data value 3024.
It can be infeasible for each non-overlapping substrings, such as each word, to correspond to an index value 3043, for example, of an inverted index structure, as these non-overlapping substrings are not of a fixed-length like the substrings of the substring-based index structure of
In some embodiments, the resulting suffix-based index structure 3670 can be stored as index data, such as a secondary index 2546, of a corresponding segment having the set of rows for the given column. Other sets of rows of a given dataset that are included in different segments can similarly have their rows indexed via the same type of suffix-based index structure 3670 via the same or different fixed-length 3551 performed upon data values of its columns. In some cases, different substring generator functions 3650 are selected for performance for sets of rows of different segments, for example, based on different cardinality. different access frequency, different query types, or other different properties of the column data for different segments.
In other embodiments, the resulting suffix-based index structure 3670 can be stored as index data, such as a secondary index 2546, for all rows of the given dataset in one or more locations. For example, a common suffix-based index structure 3670 can be generated for all rows of a dataset, even if these rows are stored across different segments, different storage structures. and/or different memory locations.
The suffix-based index structure 3670 can be considered a type of probabilistic index structure 3020 as a result of rows being identified for inclusion of subsets of a consecutive text pattern that may not include the consecutive text pattern. However, unlike the example probabilistic index structure of
The substring-based index structure 3570 of
In some embodiments, a given column storing text data, such as a given column 3023.A, can be indexed via both the probabilistic index structure 3020 of
Selection of whether to index a given column of text data via the probabilistic index structure 3020 of
In some embodiments a given column storing text data, such as a given column 3023.A, can indexed via either the substring-based index structure 3570 of
A set of corresponding index accesses 3542. 1 and 3542.2 are performed to utilize each corresponding substring 365410 identify each of a corresponding set of row identifier sets 3044 based on suffix-based index structure 3670. This can include probing the suffix-based index structure 3670 to determine the set of rows with text data that includes the corresponding substring 3654. This can include traversing down a suffix-structure such as a suffix army and/or suffix tree, progressing one character at a time based on the given corresponding substring 3654, to reach a node of an array and/or tree structure corresponding to the full substring 3654, and/or identify the set of rows mapped to this node of the array and/or tree structure. For example, the row identifier set 3044. 1 is determined via index access 3542. 1 based on being mapped to suffix index data for “red”: and the row identifier set 3044.2 is determined via index access 3542.2 based on being mapped to the suffix index data, such as corresponding index values 3043, for “bear.” The index accesses can be optionally performed in parallel, for example, via parallel processing resources, such as a set of distinct nodes and/or processing core resources. Each index access 3452 performed by query processing system 2802 can be implemented as an index element 3512 of a corresponding IO pipeline 2834 as illustrated in
An intersect subset 3544 can be generated based on performing a set intersection upon the outputted row identifier sets 3044 of the index accesses 3542 via a set intersect element 3319. The intersect subset 35-44 in this example includes row a and row c, indicating that rows a and row c include all substrings “red” and “bear”. The intersect subset 3544 can be implemented as a row identifier set 3044 of embodiments of
Data value access 3454 is performed to read rows identified in intersect subset 3544 from row storage 3022, such as rows stored in a corresponding one or more segments. A data value set 3046 that includes the corresponding data values 3024 for rows identified in intersect subset 3544 is identified accordingly. The data value access 3454 performed by query processing system 2802 can be implemented as source element 3014 of a corresponding IO pipeline 2834, and/or can otherwise be performed via other processing performed by a query processing system: 2802 executing a corresponding query against a dataset.
Inclusion-based filtering 3558 is performed by determining ones of the data value set 3046 that include the consecutive text pattern “red%bear” to render a row identifier subset 3045, and/or optionally a corresponding subset of data values 3024 of data value set 3046. This can be based on comparing each data value 3024 in data value set 3046 to the given consecutive text pattern 3548, and including only ones of row identifiers in row identifier set 3044 with corresponding ones of the set of data values 3024 in data value set 3046 that include the consecutive text pattern 3548. In this case row a is included based on having a data value 3024 of “huge red bear” that includes the text pattern “red%bear” while row c is filtered out based on being false-positive rows with a value of “bear red” that does not match the text pattern due to including all substrings in a wrong ordering not matching the given text pattern. The inclusion-based filtering 3558 performed by query processing system 2802 can be implemented as filtering element 3016 of a corresponding IO pipeline 2834, and/or can otherwise be performed via other processing performed by a query processing system 2802 executing a corresponding query against a dataset Note that if the consecutive text pattern 3548 is a single word and/or is not split into more than one substring 3654 via the split parameter, the filtering element need not be applied, as no false-positives will be identified in this case.
In various embodiments, a query processing system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, can cause the query processing system to identifying a filtered subset of the plurality of rows having text data of the column that includes a consecutive text pattern. Identifying the filtered subset of the plurality of rows having text data of the column that includes the consecutive text pattern can be based on: identifying a non-overlapping set of substrings of the consecutive identifying a filtered subset of the plurality of rows having text data of the column that includes a consecutive text pattern: splitting the text pattern into the non-overlapping set of substrings at a corresponding set of split points; identifying a set of subsets of rows by utilizing suffix-based index data corresponding to the plurality of rows to identify, for each substring of the non-overlapping set of substrings, a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings: identifying a first subset of rows as an intersection of the set of subsets of rows, and/or comparing the text data of only rows in the first subset of rows to the consecutive text pattern to identify the filtered subset as a subset of the first subset of rows that includes rows having text data that includes the consecutive text pattern.
Some or all of the method of
Some or all of the method of
Some or all of the steps of
Step 3682 includes storing a plurality of text data as a column of a plurality of rows in conjunction with corresponding suffix-based index data for the plurality of text data. Step 3684 includes identifying a filtered subset of the plurality of rows having text data of the column that includes a consecutive text pattern.
Performing step 3684 can include performing step 3686, 3688, 3690, and/or 3692. Step 3686 includes identifying a non-overlapping set of substrings of the consecutive text pattern based on splitting the text pattern into the non-overlapping set of substrings at a corresponding set of split points. Step 3688 includes identifying a set of subsets of rows by utilizing the suffix-based index data to identify, for each substring of the non-overlapping set of substrings, a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings. Step 3690 includes identifying a first subset of rows as an intersection of the set of subsets of rows. Step 3692 includes comparing the text data of only rows in the first subset of rows to the consecutive text pattern to identify the filtered subset as a subset of the first subset of rows that includes rows having text data that includes the consecutive text pattern.
In various embodiments, identifying the filtered subset of the plurality of rows is further based on reading a set of text data based on reading the text data from only rows in the first subset of rows, having the text data of only the rows in the first subset of rows to the consecutive text pattern can be based on utilizing only text data in the set of text data.
In various embodiments, the text data is implemented via a string datatype a varchar datatype a text datatype a variable-length datatype, or another datatype operable to include and/or depict text data. In various embodiments, the suffix-based indexing data is implemented via a suffix array a suffix tree a string B-tree, or another type of indexing structure.
In various embodiments, a set difference between the filtered subset and the first subset of rows is non-null In various embodiments, the set difference includes at least one row having text data that includes every one of the set of substrings in a different arrangement than an arrangement dictated by the consecutive text pattern.
In various embodiments, the text data for at least one row in the filtered subset has a first length greater than a second length of the consecutive text pattern. In various embodiments, each of the set of split points correspond to separation between each of a plurality of different words of the consecutive text data. In various embodiments, the consecutive text pattern includes at least one wildcard character. Each of the set of split points can correspond to one wildcard character of the at least one wildcard character. In various embodiments, each of the non-overlapping set of substrings includes no wildcard characters.
In various embodiments, each subset of the set of subsets is identified in parallel with other subsets of the set of subsets via a corresponding set of parallelized processing resources.
In various embodiments, the corresponding suffix-based index data for the plurality of text data indicates, for at least one of the plurality of text data a set of suffix substrings of each of a plurality of non-overlapping substrings of the text data. The plurality of non-overlapping substrings of the text data can be split at a corresponding plurality of split points of the text data. Every row included in the first subset of rows can include each of the set of non-overlapping substrings in the plurality of non-overlapping substrings of its text data.
In various embodiments, identifying the corresponding subset of the set of subsets for the each substring of the set of substrings includes identifying ones of the plurality of rows indicated in the suffix-based index data as including the each substring as one of plurality of non-overlapping substrings of the text data based on the set of suffix substrings of the one of plurality of non-overlapping substrings being indexed in the suffix-based index data.
In various embodiments, identifying the filtered subset includes applying at least one probabilistic index-based IO constrict of an IO pipeline generated for a query indicating the consecutive text pattern in at least one query predicate. For example. at least one probabilistic index-based IO construct of
In various embodiments, a filtering element of the probabilistic index-based IO construct is included in the IO pipeline based on the non-overlapping set of substrings including a plurality of substrings. In various embodiments, the method further includes identifying a filtered subset of the plurality of rows having text data of the column that includes a second consecutive text pattern. Identifying the filtered subset of the plurality of rows having text data of the column that includes the second consecutive text pattern can be based on identifying a non-overlapping set of substrings of the second consecutive text pattern as a single substring: identifying a single subset of rows by utilizing the suffix-based index data to identify, for each substring of the non-overlapping set of substrings, a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings: and/or foregoing filtering of the single subset of rows based on identifying the non-overlapping set of substrings of the second consecutive text pattern as the single substring. In various embodiments, the non-overlapping set of substrings of the second consecutive text pattern is identified as a single substring based on the consecutive text pattern including a single word and/or the consecutive text pattern not including any wildcard characters.
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database systems, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to store a plurality of text data as a column of a plurality of rows in conjunction with corresponding suffix-based index data for the plurality of text data; and/or identify a filtered subset of the plurality of rows having text data of the column that includes a consecutive text pattern. Identifying the filtered subset of the plurality of rows having text data of the column that includes the consecutive text pattern can be based on: identifying a non-overlapping set of substrings of the consecutive text pattern based on splitting the text pattern into the non-overlapping set of substrings at a corresponding set of split points: identifying a set of subsets of rows by utilizing the suffix-based index data to identify, for each substring of the non-overlapping set of substrings. a corresponding subset of the set of subsets as a proper subset of the plurality of rows having text data of the first column that includes the each substring of the set of substrings; identifying a first subset of rows as an intersection of the set of subsets of rows: and/or comparing the text data of only rows in the first subset of rows to the consecutive text pattern to identify the filtered subset as a subset of the first subset of rows that includes rows having text data that includes the consecutive text pattern.
The false-positive tuning parameter utilized by a given probabilistic indexing scheme to index a given column of a given dataset can be selected automatically by processing resources of the database system 10 and/or based on user input, for example from a discrete and/or continuous set of possible false-positive tuning parameter options. For example, the false-positive tuning parameter can be intelligently selected for probabilistic indexing based on weighing the trade-off of size of index vs, rate of false positive rows to have values read and filtered out.
As illustrated in
As illustrated in
In such embodiments, the probabilistic index structure generator module 3470 can optionally generate separate probabilistic index structures 3020 for each different segment storing rows of the dataset via secondary index generator module 2540 of
The probabilistic index structure generator module 3470 can generate a corresponding probabilistic index structure 3020 based on applying a selected false-positive tuning parameter 3720. This false-positive tuning parameter 3720 can be selected from a discrete or continuous set of possible false-positive tuning parameters indicated in false-positive tuning parameter option data 3715.
In some cases, a first false-positive tuning parameter inducing a first false-positive rate rendering a greater rate of false positives than a second false-positive rate induced by a second false-positive tuning parameter can be selected based on being more favorable than the second false-positive tuning parameter due to the first false-positive tuning parameter inducing a more favorable IO efficiency in query execution than the second false-positive tuning parameter due to less false-positive rows needing to be read and filtered out. Alternatively, the second false-positive tuning parameter can be selected based on being more favorable than the first false-positive tuning parameter due to the second false-positive tuning parameter inducing a more favorable storage efficiency of the index data for the probabilistic indexing scheme than the second false-positive tuning parameter.
As discussed previously, a probabilistic indexing scheme can be implemented as an inverted index function that indexes column data based on a bash value computed for the column values via a bash function, for example, as discussed in conjunction with
As discussed previously a probabilistic indexing scheme can be implemented as a substring-based indexing scheme indexes text data based on its fixed-length substrings, for example, as discussed in conjunction with
Different columns of a given dataset can be indexed via a same or different type of probabilistic indexing scheme utilizing different respective false-positive tuning parameters of the set of possible false-positive tuning parameter options. Alternatively or in addition different segments can index a same column via a probabilistic indexing scheme utilizing different respective false-positive tuning parameters of the set of possible false-positive tuning parameter options.
In some embodiments, the false-positive tuning parameter selection module 3710 is selected from the options in the false-positive tuning parameter option data 3715 via user input to an interactive user interface displayed via a display device of a client device communicating with the database system 10. For example, an administrator can set the false-positive tuning parameter option data 3715 of probabilistic indexing structures 3020 for one or more columns of a dataset as a user configuration sent to and/or determined by the database system 10).
Alternatively, as illustrated in
As illustrated in
In some embodiments, the false-positive tuning parameter is automatically selected for one or more segments by the secondary indexing scheme selection module 2530 of the segment indexing module 2510 of
One or more false-positive tuning parameters can otherwise be automatically selected and/or optionally changed overtime for one or more corresponding columns that are indexed via a corresponding probabilistic indexing scheme via at least one processor of the database system 10, for example, based on automatic optimization of and/or evaluation of a trade-off between IO efficiency and storage efficiency. Alternatively or in addition, one or more false-positive tuning parameters can be selected and/or optionally changed overtime for one or more corresponding columns that are indexed via a corresponding probabilistic indexing scheme based on user configuration data received from a client device of a corresponding user, such as an administrator.
In various embodiments a database system includes at least one processor and a memory that stores operational instructions. The operational instructions, when executed by the at least one processor, can cause the database system to: determine a selected false-positive tuning parameter of a plurality of false-positive tuning parameter options, store index data for a plurality of column values for a first column of a plurality of rows in accordance with a probabilistic indexing scheme that utilizes the selected false-positive tuning parameter; and/or facilitating execution of a query including a query predicate indicating the first column. Facilitating execution of a query including a query predicate indicating the first column includes identifying a first subset of rows as a proper subset of the plurality of rows based on the index data of the probabilistic indexing scheme for the first column: and/or identifying a second subset of rows as a proper subset of the first subset of rows based on identifying ones of a first subset of the plurality of column values corresponding to the first subset of rows that compare favorably to the query predicate A number of rows included in a set difference between the first subset of rows and the second subset of rows can be based on the selected false-positive tuning parameter.
Some or all of the method of
Some or all of the steps of
Step 3782 includes determining a selected false-positive tuning parameter of a plurality of false-positive tuning parameter options. Step 3784 includes storing index data for a plurality of column values for a first column of a plurality of rows in accordance with a probabilistic indexing scheme that utilizes the selected false-positive tuning parameter. Step 3786 includes facilitating execution of a query including a query predicate indicating the first column.
Performing step 3786 can include performing step 3788 and/or 3790. Step 3788 includes identifying a first subset of rows as a proper subset of the plurality of rows based on the index data of the probabilistic indexing scheme for the first column. Step 3790 includes identifying a second subset of rows as a proper subset of the first subset of rows based on identifying ones of a first subset of the plurality of column values corresponding to the first subset of rows that compare favorably to the query predicate. A number of rows included in a set difference between the first subset of rows and the second subset of rows is based on the selected false-positive tuning parameter.
In various embodiments, determining the selected false-positive tuning parameter is based on user input selecting the selected false-positive tuning parameter from the plurality of false-positive tuning parameter options. In various embodiments, a Storage size of the index data is dictated by the selected false-positive tuning parameter. A false-positive rate of the probabilistic indexing scheme can be dictated by the selected false-positive tuning parameter. The false-positive rate can be a decreasing function of the storage size of the index data.
In various embodiments, determining the selected false-positive tuning parameter is based on automatically selecting the selected false-positive tuning parameter. In various embodiments, the selected false-positive tuning parameter is automatically selected based on at least one of: index data storage efficiency, or IO efficiency conditions. In various embodiments, the selected false-positive tuning parameter is automatically selected based on a cardinality of the column values of the first column.
In various embodiments, the method further includes generating index efficiency data based on execution of a plurality of queries that includes the query. In various embodiments, the method further includes determining to update the probabilistic indexing scheme for the first column based on the index efficiency data compares unfavorably to an index efficiency threshold. It various embodiments, the method further includes generating updated index data in accordance with an updated probabilistic indexing scheme for the first column that utilizes a newly selected false-positive tuning parameter that is different from the selected false-positive tuning parameter based on determining to update the probabilistic indexing scheme.
In various embodiments, the selected false-positive tuning parameter is selected for the first column. The method can further include determining a second selected false-positive tuning parameter of the plurality of false-positive tuning parameter options for a second column of the plurality of rows. The method can further include storing second index data for a second plurality of column values for the second column of the plurality of cows in accordance with a second probabilistic indexing scheme that utilizes the second selected false-positive tuning parameter.
In various embodiments, the probabilistic indexing scheme and the second probabilistic indexing scheme utilize a same indexing type. The second selected false-positive tuning parameter can be different from the first false-positive tuning parameter. In various embodiments, the second selected false-positive tuning parameter is different from the first false-positive tuning parameter based on: the first column having a different cardinality from the second column; the first column having a different data type from the second column; the first column having a different access mate from the second columns the first column appearing in different types of query predicates from the second column; column values of the first column having different storage requirements from column values of the second column: column values of the first column having different IO efficiency from column values of the second columns and/or other factors.
In various embodiments, the plurality of rows are stored via a set of segments. The selected false-positive tuning parameter can be selected for a first segment of the set of segments. The index data for a first subset of the plurality of column values can be in accordance with the probabilistic indexing scheme that utilizes the selected false-positive tuning parameter for ones of the plurality of rows in the first segment of the set of segments. In various embodiments the method further includes determining a second selected false-positive tuning parameter of the plurality of false-positive tuning parameter options for a second segment of the set of segments. various embodiments the method further includes storing second index data for a second subset of the plurality of column values for the first column in accordance with a second probabilistic indexing scheme that utilizes the second selected false-positive tuning parameter for other ones of the plurality of rows in the second segment of the set of Segments.
In various embodiments, the probabilistic indexing scheme and the second probabilistic indexing scheme utilize a same indexing type. The second selected false-positive tuning parameter can be different from the first false-positive tuning parameter. In various embodiments, the second selected false-positive tuning parameter is different from the first false-positive tuning parameter based on: column values for rows in first segment having a different cardinality from column values for rows in second segment: column values for rows in first segment having a different access rate from column values for rows in second segment: column values for rows in first segment appearing in different types of query predicates from column values for rows in second segment; and/or other factors.
In various embodiments, the index data of the probabilistic indexing scheme includes a plurality of hash values computed by performing a bash function on corresponding ones of the plurality of column values. The hash function can utilize the selected false-positive tuning parameter. In various embodiments, a rate of hash collisions of the hash function is dictated by the selected false-positive tuning parameter In various embodiments a same fixed-length of the plurality of hash values is dictated by the selected false-positive tuning parameter.
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps described above.
In various embodiments a non-transitory computer readable storage medium includes at least one memory section that stores operational instructions that, when executed by a processing module that includes a processor and a memory, causes the processing module to determine a selected false-positive tuning parameter of a plurality of false-positive tuning parameter options; store index data for a plurality of column values for a first column of a plurality of rows in accordance with a probabilistic indexing scheme that utilizes the selected false-positive tuning parameter: and/or facilitating execution of a query including a query predicate indicating the first column. Facilitating execution of a query including a query predicate indicating the first column includes identifying a first subset of rows as a proper subset of the plurality of rows based on the index data of the probabilistic indexing scheme for the first column; and/or identifying a second subset of rows as a proper subset of the first subset of rows based on identifying ones of a first subset of the plurality of column values corresponding to the first subset of rows that compare favorably to the query predicate. A number of rows included in a set difference between the first subset of rows and the second subset of rows can be based on the selected false-positive tuning parameter.
In some embodiments, query execution can require that filters be in conjunctive normal form (CNF), where filters are optionally exclusively handed in CNF. In such embodiments, the general-purpose normalization of filters to CNF can add new variables to the fillet expression and/or can take a long time to complete, sometimes longer than the length of the actual query execution.
Query expressions 2511 can be written in any format (e.g. CNF/disjunctive normal form (DNF) and/or neither, where sub-predicates may be expressed in CNF or DNF, but may be combined in such a way, for example, via disjunctions (e.g. OR operators, optionally expressed as ‘OR’ or ‘|’) or conjunctions (e.g. AND operators, optionally expressed as ‘AND’ or ‘&&’). In particular users that write query expressions 2511 as query requests to be executed via database system 10 can write filters in any format they choose. When the user requests a filter in disjunctive normal form (DNF), the filter is optionally re-normalized into a collection of CNF filter predicates that can be executed separately to equivalently satisfy the user's filtering constraints, for example, as illustrated in the example of
In some embodiments, a query's execution can begin based on materializing (e.g. reading) on-disk data. As this data is materialized, the query execution engine processes incoming data as requested by the user's command. This processing can include filtering. If possible, some or all filtering is performed before data is materialized, for example, via predicates executed at the IO level, for example, via a corresponding IO pipeline, which can leverage index data and/or source certain columns to generate a filtered set of rows (e.g. set of row numbers/row identifiers identifying which pre-filtered rows need be processed/have their column values materialized if necessary).
In some embodiments, the pre-materialization process (e.g., the filtering via IO pipelines) exclusively filters in accordance with applying a single CNF predicate. In such embodiments, when a query plan, for example, such as the query plan in
One solution to this issue can include re-normalizing the collection of CNFs into a single CNF, but this operation of convening into CNF can have major drawbacks in processing time as described above.
The operator execution flow 2817 for execution can implement a pre-data materialization portion 3901 that performs filtered row set generation 3903 to emit a filtered row set 3905 for further filtering and/or processing via a query resultant generation process 3904 implemented via a post-data materialization portion 3902. Some or all features and/or functionality of operator execution flow 2817 of
The pre-data materialization portion 3901 can be implemented via any parallelized and/or serialized flow of IO operators 2821, such as via source elements. filtering elements, and/or index elements of an IO pipeline 2835. The pre-data materialization portion 3901 can be implemented via any parallelized and/or serialized flow of any other operators 2520 alternatively or in addition to operators 2821.
The pre-data materialization portion 3901 can be executed via query execution module 2504 based on accessing row values and/or index data stored in database storage 2450, such as one or more segments stored in one or more memory drives and/or index data such as secondary index data 2545 for one or more columns, cluster key index data, and/or other index data that is implemented as probabilistic index data or non-probabilistic data. The index data can be implemented as an inverted index structure or any other type of index structure, such as any of the types of indexes described herein.
The pre-data materialization portion 3901 can be executed via one or more nodes 37, for example, in parallel. For example, pre-data materialization portion 3901 is executed via nodes participating at an IO level 2416 of a query execution plan 2405 for the given query.
The post-data materialization portion 3902 can be implemented via any parallelized and/or serialized Dow of IO operators 2830, which can perform further filtering. joining, aggregation, new column generation, manipulation of data, and/or any other processing in conjunction with executing the corresponding query. This can optionally include performing further IO operators 2821, such as via source elements. filtering elements, and/or index elements of an IO pipeline 2835.
The post-data materialization portion 3902 can be executed via one or more nodes 37, for example, in parallel. For example, post-data materialization portion 3902 is executed via nodes participating at one or more inner levels 2414 of a query execution plan 2405 for the given query, and/or can be executed via a root level node participating at a root level 2412.
The execution of operator execution flow 2817 can render execution of a corresponding query expression, for example. based on semantic equivalence with the query expression and/or based on being guaranteed to generate a query resultant that is the correct query resultant for the query expression when applied to the specified data set in database storage 2450.
The query expression 2511 can correspond to a query expression generated by a user, for example, in conjunction with a query request 2515 received from an external requesting entity 2508 and/or otherwise generated and/or received by a customer. an administrator, and/or any other user. The query expression 2511 can optionally correspond to a query expression automatically generated for execution by database system: 10 and/or another system, for example, without user intervention. The query expression 2511 can otherwise be received, accessed, generated, and/or determined for execution. The query expression 2511 can be expressed in SQL and/or any query language.
The query expression 2511 can include a filtering predicate 4815, for example, expressed in a WHERE clause of a SQL. query and/or otherwise indicating conditions by which rows be filtered for output and/or further processing. In particular, in the examples discussed in conjunction with
The filtering predicate 4815 can optionally correspond to the original written form of the filtering predicate in the query expression 2511, where the user that generated the query expression expressed some of all predicates in filtering predicate 4815 as a disjunction of CNF predicates and/or other collection of CNF predicates. For example, the collection of CNFs can be derived from a DNF expression written by the user as filtering predicate 4815 of query expression 2511.
The filtering predicate 4815 can optionally correspond to a processed form of the filtering predicate 4918 in the query expression 2511, for example, based on extracting predicates and/or based on rewriting the query expression to write predicates into CNF, combine predicates into CNF, perform a distributed property to extract a common predicates out of multiple predicates and/or apply a common predicate into multiple predicates, or other rewrites/extractions from the original expression to render some or all predicates in filtering predicate 4815 as a disjunction of CNF predicates and/or other collection of CNF predicates.
The filtered row set generation 3903 performed via pre-data materialization portion can be implemented to generate the filtered row set 3905 based on applying a disjunction to the plurality of CNF predicates 3910.1-3910.R. Various means of applying the disjunction to the plurality of CNF predicates 3910.1-3910. R are discussed in conjunction with
The filtered row set 3905 can include a set of rows satisfying the disjunction to the plurality of CNF predicates 3910.1-3910.R, and/or set of rows further filtered from a superset of rows satisfying the disjunction to the plurality of CNF predicates 3910.1-3910.R. The set of rows can be expressed as row numbers and/or row identifiers. Some or all column values can be included for ones of the rows that meet the disjunction to the plurality of CNF predicates 3910.1-3910. R, for example, based on these column values being needed for output and/or for further filtering and/or processing.
Some or all of the set of rows can be determined based on sourcing values for a corresponding column from storage (e.g. from a corresponding segment) and applying a filter to identify which rows meet a corresponding predicate 3910 denoting a condition applied to the corresponding column Some or all of the set of rows can be determined based on accessing index data for a corresponding column to identify which rows meet a corresponding predicate 3910 denoting a condition applied to the corresponding column, for example, where the column values themselves need not be accessed. Sourcing and filtering vs, applying index data can be dependent on whether index data was generated and/or is accessible for the given column.
The filtering predicate 4815 implemented in pre-data materialization portion is optionally not semantically equivalent to the original written form of the filtering predicate in the query expression 2511, and instead corresponds to a broader predicate that renders identification of a superset of rows as filtered row set 3905 that need be further filtered in the post-data materialization portion to render the identification of only filtered rows meeting the filtering predicates as required by the query expression 2511.
At the IO (pre-materialization) layer. filtering. I/O, and materialization of values can be handled in the IO Operator, such: as an IO operator 2821. For a given query predicate and/or a given segment, the operator can compile and executes an IO pipeline. The pipeline can represent a sequence of filters and operations that are performed sequentially to determine and materialize the set of row values that satisfy the predicate. The pipeline filters may use inverted indexes and/or cluster key indexes, and/or may also materialize and filter row values in order to apply filters that are not supported by indexes. The compiled pipeline may vary from segment to segment based on the presence of indexes and statistics about data distribution in the segment. The IO pipeline 2835 can be implemented via some or all features and/or functionality of IO pipeline discussed in conjunction with
In some embodiments, the filter implemented via the IO pipeline is in CNF form (e.g. a plurality of disjunction predicates to which a conjunction is applied (e.g, disjunction predicate 1 AND disjunction predicate 1 AND . . . ). For each disjunction predicate in the CNF, the cost of evaluating the filters in the disjunction can be evaluated, and disjunction predicates can be ordered such that the least expensive and most selective disjunctions were evaluated first. The structure of the pipeline can allow elements to only handle rows that have satisfied the filters of previous elements, so placing the cheapest disjunction predicates earlier in the pipeline can allows subsequent more expensive disjunctions to process less data. Placing more selective disjunctions first can also allow more rows to be eliminated earlier in the pipeline. A given disjunction predicate can optionally include a disjunction applied to one or more sub-predicates (e.g, disjunction predicate 1=predicate A OR predicate B OR . . . ), where predicates A. B, and Care simple predicates with no nested ANDs as required by CNF.
To support a predicate that is in Union of CNF form (e.g. a disjunction of CNF predicates), the pipeline can instead be composed of several parallel CNF pipelines, with a UNION at the end that combines the resulting rows. As illustrated in
One approach to generating the plurality of parallelized branches 3911 can include performing the compilation logic on each CNF in the union and UNION-ing the result of each pipeline. For example, each parallelized branch 3911 can be generated as its own IO pipeline 2935 as described herein, which can have its own set of serialized and/or patallehzed IO pipeline elements applied to implement the corresponding CNF predicate. In some embodiments. further optimizations can be applied to enable sharing of work across multiple CNFs in the same pipeline. Such embodiments are discussed in further detail in conjunction with
The IO pipeline can be generated to include three corresponding parallelized branches 3911.1-3911.3. In this example. the CNF predicate 3910.1 is implemented via an index element 3512 to an inverted index for the column colD to identify rows having values for colD equal to 1. In this example, the CNF predicate 3910.2 is implemented via an index element 3512 to an inverted index for the column col2 to identify rows having values for col2 equal to 2 and having values for col2 equal to 22. In this example, the CNF predicate 3910.3 is implemented via a source element 301410 read column values of col3, where filter element 3016 is applied to filter these rows based on ones of the rows with sourced column values equal to 3. For example, the sourcing and filtering is applied to implement CNF predicate 3910.3 based on col3 not being indexed.
In some embodiments, the example CNF predicate 3910.2 of
In some embodiments. CNF predicates 3910 having a conjunction between multiple simple predicates can implement conditions where it is possible for a row to satisfy the predicate (e.g. “col2>2 AND col2<22”, or “col2=2 AND col1=22”).
Some or all features and/or functionality of the filtering predicate 4815 and/or corresponding IO pipeline 2835 of
The IO pipeline can be generated to include two corresponding parallelized branches 3911. 1 and 3911.2 that implement CNF predicate 3910.1 and 3910.2, respectively. In this example, the CNF predicate 3910.1 is implemented via three serialized portions 3922.1. 3922.2, and 3922.3 included within the parallelized branch 3911. 1 in series. For example, the ordering was selected based on the cost to evaluate each conjuncted predicates 2923.1.1. 2923.1.2 and/or 2923.1.3 and/or the number of rows expected and/or known to be filtered out via each conjuncted predicates 2923.1.1. 2923.1.2 and/or 2923. 13 based on column distribution data for corresponding columns. As a particular example serialized portion 3022. 1 is applied first and serialized portion 3022.3 is applied last based on conjuncted predicates 2923. 1. 1 being least costly to evaluate: conjuncted predicates 2923. 1. 1 being expected to filter the most rows out, conjuncted predicates 2923.1.3 being most costly to evaluate: and/or conjuncted predicates 2923.1.3 being expected to filter the least rows out.
In this example, conjuncted predicate 2923. 1 is implemented via an index element 3512 to an inverted index for the column colD to identify rows having values for colD equal to 1.
In this example, conjuncted predicate 2923.2 includes two disjunct sub-predicates 3924. 1.2. 1 and 3924. 1.2.2 that are disjuncted via an OR operator, and are thus applied in parallel via two parallelized sub-branches 3921.1 and 3921.2 within the serialized portion 3922.2 of parallelized branch 3911.1 whose output is combined via a union element 3218 within serialized portion 3922.
The disjunct sub-predicate 3924. 1.2. 1 can be implemented via a source element 3014 to source the column values of column 2 for incoming rows emitted by serialized portion 3922. 1. followed serially by a fitter element to emit only rows with the sourced value of col2 being equal to zero. For example, this sourcing and filtering is applied based on col2 not being indexed. The disjunct sub-predicate 3924. 1.2.2 can be implemented via an index element 3512 to an inverted index for the column col3 identify rows having values for col3 equal to 3.
In this example, conjuncted predicate 2923.3 is implemented via a filter element 3016 to identify rows emitted by serialized portion 39222 having values for co21 equal to 0. For example, the conjuncted predicate 2923.3 is further implemented via the source element 3014 sourcing the rows for column 2 in implementing disjunct sub-predicate 3924. 1.2. 1, where the column values of these source values are again utilized to apply a filter element 3016 of serialized portion 3922.3.
Meanwhile, in parallelized branch 3911.2, the CNF predicate 3910.2 of this example is implemented via an index element 3512 to an inverted index for col2 to identify rows having values of colD equal to two.
In some embodiments, the example CNF predicate 3910.21 of
In some embodiments. CNF predicates 3910 having multiple disjuncted sub-predicates of a conjuncted predicate can implement conditions where it is required for both sub-predicate conditions be checked (e.g. “(col1=) AND (col 2>0 OR col3 =3) AND col2<5) OR col2=2)” or “(col1=1 AND (col 2=0) OR col3=3) AND col4=S) OR col2=2)”).
Some or all features and/or functionality of the filtering predicate 4815 and/or corresponding IO pipeline 2835 of
As illustrated in
In other embodiments, the common predicate is optionally applied serially prior to the disjunction of CNF predicates. where the input to each parallelized processes corresponds to output of the additional branches 3913, and thus corresponds to only rows meeting the common predicates 3912.
The one or more additional branches 3913 can implement the common predicate 3912 via an index element 3512 to a corresponding index structure for a corresponding column. The one or more additional branches 3913 can implement the common predicate 3912 via a source element 3014 to source column values for a corresponding column, and via a filtering element 3016 serially after this source element 3014 to filtering element.
This application of the one or more common branches 3913 to apply one or more common predicates 3912 can be implemented based on applying a type of optimization in generating the IO pipeline to share IO elements, such as a shared filtering-based optimization.
In some embodiments, the common predicate 3912 can correspond to a filter upon a time column of the dataset/corresponding segment. For example, as well as the CNFs to be unioned a filtering predicate 4815 may include filters on the time column of the segment/dataset. These filters can be intersected with any other filters in the predicate, and can thus be implemented as common predicate 3912. A Union of CNF predicate with time filters can be in the format looks like (time_filters) AND (CNF) OR CNF2 OR CNF3 . . . ) where time_filters may look like time_col>time_min AND time_col <time_max. Time columns optionally do not have indexes, so the row values of the time column can be materialized so that those values may be filtered on. If time filters are present in a Union of CNF pipeline, they can be placed at the end of the pipeline, after the UNION. For example, this placement serially after the UNION of CNF predicates can be based on the time column not being indexed and requiring reading of values via source element 3014, where this placement aids in optimizing processing due to needing to read the time column for fewer rows. Thus, in some embodiments, once the CNF filters have been applied, the row values of the time column are materialized for the row numbers that satisfy the CNF filters, and the time filter is applied to those values. In embodiments where the common predicates 3912 apply filters to a time column storing temporal values for the rows, this application of the one or more common branches 3913 to apply one or more common predicates 3912 can be implemented based on applying a time column-based optimization.
In the example of
In this example, the parallelized branches 3911.1-3911.3 are each implemented via an index element 3512 to a corresponding inverted index structure of a corresponding column to render identification of rows satisfying the corresponding predicate 3910 as discussed previously, where a union 3218 is applied to the outputs of the parallelized branches 3911.1-3911.3 as discussed previously. This output can be further processed via an additional branch 3913 that includes a source element 3014 sourcing the time column time_col and a filter element 3016 that filters the rows based on rows with sourced values of time_col that are greater than the minimum value min val and are less than the maximum value max_val.
Some or all features and/or functionality of the filtering predicate 4815 and/or corresponding IO pipeline 2835 of
The common predicate 3912 of
In some embodiments, pipeline elements that materialize row values are either used for row-level filtering inside the pipeline, or emit rows as the output of the pipeline operator. Source elements that are not necessary for any filters and which only materialize the final output rows of the pipeline can be placed after the UNION element, and materialize row values once for the entire pipeline. These elements only need to materialize those rows which passed the filters of the union of CNFs. In some embodiments, if a source element is necessary for a row-level filter within a CNF, rows only need to be materialized once for the filters in that CNF. However, in some embodiments, the rows emitted by the other CNFs must also be materialized after the UNION is applied.
As illustrated in
This use of a same source element 3014 for a given column to source column values for rows satisfying multiple CNF predicates 3910 (e.g. to source column values for rows outputted via R nested IO pipelines implementing parallelized branches 3911.1-3911.R) can be based on applying an optimization to share IO elements, such as a shared source element-based optimization. This sourcing of rows for multiple CNF predicates via shared source elements 3014 rather than sourcing for different CNF predicates separately via different source elements in parallel can improve processing efficiency, for example, due to the amount of sourcing can be reduced (e.g. a given row satisfying multiple CNFs has its column values sourced only once) and/or due to the IO pipeline being simplified.
In embodiments where additional common predicates are applied via one or more additional branches 3913 of
In cases where multiple columns need be sources, the source elements 3014 can be applied in parallel via a corresponding plurality of parallelized branches, for example, as illustrated in the example of
In some cases, all output columns 4816 are distinct from columns referenced in predicates 3910.1-3910.R, and are thus sourced for the first time via source elements 3014 applied to the disjuncted CNF output row set. In some cases, one or more output columns 4816 correspond to columns referenced in predicates 3910.1-3910.R, but are still sourced for the first time via source elements 3014 applied to the disjuncted CNF output row set based on having had index elements 3512 applied in parallelized branchers 3911 to identify the rows for disjuncted CNF output row set 3906.
In some cases, one or more output columns 4816 correspond to columns referenced in predicates 3910.1-3910.R and are sourced in corresponding parallelized branchers 3911 to identify the rows for disjuncted CNF output row set 3906 (e.g, based on these columns not being indexed), where the source elements 3014 are again re-applied for these columns to ensure columns are sourced for all rows satisfying any of the CNF predicates 3910.1-3910 R. Such an example is illustrated in
In the example of
Source elements 3014 for columns col4, col5, and col6 can be applied in parallel serially after the union 3218 to render sourcing of corresponding column values for only rows satisfying at least one of the CNF predicates 3910.1-3910.3. Note that in this example, the output columns 4816 are distinct from the columns upon which filtering predicate 4815 is applied/based upon.
The example filtering predicate 4815 of this example can correspond to the filtering predicate 4815 of
In other embodiments, the re-sourcing is optionally not required in cases where the sourcing via a source element 3014 is applied in a given parallelized branch 3911 implementing a given predicate 3910 serially before any other filtering via filter elements or index elements of this given parallelized branch 3911, and thus corresponds to values being sourced for all input rows. where values for columns satisfying all other predicates are thus guaranteed to have been already sourced.
As illustrated in
In some embodiments, the common nested predicate 3935 is nested within a proper subset of the CNF predicates 3910. wherein at least one CNF predicates 3910 does not include the common nested predicate 3935 and is thus not implemented via cross-branch shared IO pipeline elements 3432 in its parallelized branch 3911. In offer embodiments, the common nested predicate 3935 is nested within all of the CNF predicates 3911. In some embodiments, one or more CNF predicates 3910 exclusively include the common nested predicate 3935, where unshared IO elements 3933 are thus not implemented in a corresponding parallelized branch 3911.
This use of a same shared IO pipeline elements 3432 for implementing multiple CNF predicates 3910 can be based on applying an optimization to share IO elements, such as a shared filtering-based optimization.
In some embodiments, these more cross-branch shared IO pipeline elements 3432 correspond to one or more index elements 3512 to probe a corresponding inverted index for a corresponding column for values meeting the common nested predicate 3935. For example, the shared filtering-based optimization is implemented as a shared index element-based optimization based on sharing an index element to probe an index for implementing multiple CNF predicates.
For example, in some embodiments, an invented index filter is composed of one or more index probes, where each probe represents a value being searched for in the index. In some embodiments, same probe being placed in a dedicated single-probe index element and reused for the multiple filters that require the output of that index search, where this element could optionally n have an incoming row list from earlier disjunctions in the CNF. In other embodiments, when the predicate contains multiple CNFs, index probes may be reused across CNFs. In such cases, these index probes are required to be constructed at the beginning of the pipeline with no incoming filters, so that their emitted row lists are not limited by the filters of any respective CNFs.
In other embodiments, these more cross-branch shared IO pipeline elements 3432 correspond to a source element 3014 and corresponding filtering element 3016 to source and fillet a column, for example, for such nested predicates open columns that are not indexed, where the a shared filtering-based optimization is applied to facilitate use of the source element 3014 and corresponding filtering element 3016 as cross-branch shared IO pipeline elements 3432 implementing multiple CNF predicates 3910.
In the example of
The CNF predicate 3910.1 (col1=1 OR col1=0) AND (col3-3)) can be implemented via a union 3218 applied to output of filtering elements 3512. 1 and 3512.2 to identify rows having colD equal to one or zero, and via another index element serially after this union to identify ones of the rows having colD equal to one or zero that also have col3-3.
The filtering element 3512.2 can be a cross-branch shared IO pipeline element 3932 based on being utilized to implement both CNF predicate 3910.1 and CNF predicate 3910.2. The filtering element 3512.1, union operator 3218, and filtering element 3512.4 can be unshared elements 3933.1.a, 3933. 1.b, and 3933.1.c, respectively, based on only being utilized in implementing the CNF predicate 3910.1 and not CNF predicate 3910.2. While not strictly parallel with the implementing of CNF predicate 3910.2 due to 3512.2 being a cross-branch shared IO pipeline element 3932, this progression of IO elements can be considered parallelized branch 3911.1 as described herein.
The CNF predicate 3910.2 (col1=4 AND col1=2 AND col1=0) can be implemented via an intersect 3219 applied to output of filtering elements 3512.2 and 3512.3 to identify rows having colD equal to one or having colD equal to one and col 2 equal to four. As discussed above, filtering element 3512.2 can be a cross-branch shared IO pipeline element 3932 based on being utilized to implement both CNF predicate 3910. 1 and CNF predicate 3910.2. The filtering element 3512.3 and intersect element 3219 can be unshared elements 3933.2.a and 3933.2 b, respectively, based on only being utilized in implementing the CNF predicate 3910.2 and not CNF predicate 3910.1. While not strictly parallel with the implementing of CNF predicate 3910.1 due to 3512.2 being a cross-branch shared IO pipeline element 3932, this progression of IO elements can be considered parallelized branch 3911.2 as described herein.
In some embodiments, the example filtering predicate 3910.2 automatically evaluates to FALSE based on a given column value for colI not having the value of 2 and the value of 4 and the value of 0. In such cases, the IO pipeline optionally does not include a parallelized branch for such predicates that will evaluate to FALSE for all rows based on all rows being guaranteed to fail the corresponding condition, where the resulting filtered row set will be null and not provide any new rows to the output row set 3909 generated via UNION 3218.
In some embodiments. CNF predicates 3910 having a common nested predicate 3935 can implement conditions where it is possible for a row to satisfy the predicate 3910.2 (e.g. “(col1=1 OR colD>0) AND (col3=3) OR (col1<IO AND colD>0)” or “(col1=1 OR col1=0) AND (col3=3) OR (col2=4 AND col3=2 AND col1=0)”.
In particular, the pre-data materialization portion 3901. 1 updated from pre-data materialization portion 3901.1 can be generated via flow optimizer module 4914 to push filtering into the IO level, for example, where a disjunction of CNF predicates 3910.1-3910. R as pushed into the pre-data materialization portion 3901.1 as a result of applying flow optimizer module 4914. where the disjunction of CNF predicates 3910. 1-3910.R is not applied in an initial operator execution flow 2817.0.
The operator execution flow 2817.0 can be the same or different for different segments. The operator execution flow 2817.1 can be the same or different for different segments. For example, different segments have different operator execution flows 2817.0, and/or different segments have different operator execution flows 2817. 1 generated from a same operator execution flow 2817.0, based on different segments having different IO pipelines generated based on having different data distribution for their columns rendering different optimal arrangements of IO elements, having different indexes generated for different columns and/or having different subsets of columns indexed: and/or other reasons, for example, as discussed previously herein.
In some embodiments, all plans 2817.0 that undergo optimization will be checked to see if they have any collections of CNFs that are candidates to be executed pre-materialization. In some embodiments plan 2817.0 satisfies these requirements if. and/or only if. (a) the plan sends identical data blocks to several streams of execution at one point: (b) these identical data blocks are being sent to CNF filtering operations without its original data values being thrown away or altered. (As described in the motivation and problem section, this means that these CNFs are the collection of CNFs generated upon encountering a user-submitted DNF filter expression.): and/or (c) the identical data blocks being sent to the CNF filtering operations come unaltered from I/O materialization.
In some embodiments, when a plan satisfies all of these requirements a transformation/rewrite can be applied to the plan (e.g., via rewrite process 3945) that will allow this filter to occur pre-materialization. In some embodiments, this transformation will apply if it has not been applied before, and/or the amount of individual filtering expressions is not much larger than it was before the transformation. The requirement of all three of one or both of these additional requirements can further implement pre-materialization rewrite requirement 4940.
In some embodiments, it is guaranteed guarantee that data blocks go directly off of disk into a collection of CNF expressions (e.g. requirements (a)—(c) are met and/or pre-materialization rewrite requirement 4940 is otherwise met) a union of those CNF expressions can be incorporated into the pre-materialization filters. This can be incorporated safely (e.g. while guaranteeing query correctness) because we know that any data blocks that fail all CNF expressions will not be required anywhere else upstream and can be discarded, which can save memory and processing time. Such implementing of the union of those CNF expressions applied via rewritten pre-data materialization portion 3901.1 can be implemented via some or all features and/or functionality of applying a disjunction of CNF predicates described in conjunction with
In some embodiments, the original collection of CNF filters expressions can be required to be executed, for example. via operators of in post-data materialization portion 3902, even when the union of these filters is executed pre-materialization: this ensures that the rewritten plan behaves equivalently to its previous plan. For example, some or all of the post-data materialization portion 3902 of operator execution flow 2817.1 is unchanged from the post-data materialization portion 3902 of operator execution flow 2817.0. In some embodiments, this requirement is based on the collection of CNF predicates 3910.1-3910.R being applied/combined defiantly beyond just a disjunction being applied, where a UNION of corresponding CNF expression output does not render the correct. fully filtered output for the given query expression 2511. In some embodiments, the original collection of CNF filters expressions of post-data materialization portion 3902.0 is optionally modified and/or removed in generating post-data materialization portion 3902.1, for example, in cases where equivalent filtering is guaranteed to have been applied in the rewritten pre-data materialization portion 3901.1.
In some embodiments, this disjunction of CNF predicates satisfies a DNF expression, such as a user-submitted DNF expression indicating as the filtering predicate 4815 of the query expression 2511, for example, based on requirement (b) being met, where the CNFs are the collection of CNFs generated upon encountering a user-submitted DNF filter expression.
The implementing of filtering in pre-data materialization portion 3901. 1 via a disjunction of CNF predicates can include performing one or more optimizations, for example, when generating a corresponding IO pipeline and/or otherwise updating the operator execution flow 2817.1 via flow optimizer module 4914.
In some embodiments, one or more optimization types 3950 implement an optimization discussed in conjunction with
In some embodiments, one or more optimization types 3950 implement a shared filtering-based optimization, for example, with au optimization condition 3951 denoting that the filtering predicate 2815 include a common predicate 3912 common to all CNF predicates and/or denoting that the filtering predicate 2815 include a nested common predicate 3935 common to some or all CNF predicates.
In some embodiments, one or more optimization types 3950 implement a time column-based optimization, for example. with an optimization condition 3951 denoting that the filtering predicate 2815 include a tune-based predicate 3912 for a time.
In some embodiments, one or more optimization types 3950 implement a shared source element-based optimization, for example, with an optimization condition 3951 denoting that the query expression include output columns that be emitted. In some embodiments, one or more optimization types 3950 implement a shared index element-based optimization, for example, with an optimization condition 3951 denoting that multiple CNF predicates include a same common nested predicate for a column having index data that can be probed.
In some embodiments, one or more optimization types 3950 implement a single CNF predicate generation optimization for example, to implement some or all functionality of
In some embodiments, one or more optimization types 3950 implement other types of optimizations to simplify and/or improve efficiency of pre-data materialization portion 3901 and/or the operator execution flow 2817 as a whole.
A pre-materialization filter may already exist at DO upon rewrite. In that case. Boolean algebra Distributive Law can be utilized in order to incorporate the series of upstream CNF expressions properly into the existing I/O CNF expression. For example. this distribution is applied to render a disjunction of a set of new CNF expressions that each include conjunction with the existing CNF predicates 3951 of the existing pre-materialization filter of pre-data materialization portion 3901.0. This can optionally be applied based on the filtering predicate 4815 denoting the disjunction of CNF predicates 3910.1-3910 R be conjuncted with this additional CNF predicates 3951.
Such an example of applying this Boolean algebra Distributive Law to render generation of a disjuncted set of new CNF predicates is illustrated in
The example of
In particular, the IO level can be written to apply CNF1| CNF2| CNF3 as discussed previously. In some embodiments. after this rewrite of the IO level is applied, the original collection of CNF filters expressions CNF1, CNF2, and CNF3 are still executed during post-data materialization portion 3902 as illustrated in
As discussed previously, database system 10 can be implemented to rewrite a plan evaluating a simple DNF by adding extra filters to DO to be evaluated pre-materialization. A DNF evaluated like in the pre-rewrite query can be evaluated as a Union of CNFs during VO. The example of
In this example a fiber CNF4 is already included at the IO level as a CNF predicate 3951. Applying the disjunction of CNF filters CNF1. CNF2, and CNF3 (e.g. CNF predicates 3910.1-3910.3, respectively), can include applying a conjunction of CNF predicate CNF4 with the disjunction of these predicates CNF1-CNF3, for example, based on the filtering expression 4815 indicating semantically equivalent filtering. This expression CNF4 AND (CNF1 OR CNF2 OR CNF3) can be rewritten as (CNF4 AND CNF1) OR (CNF4 AND CNF2) OR (CNF4 AND CNF3) based on applying the Boolean distributive law as discussed in conjunction with
In some embodiments, as discussed in conjunction with
In this example, CNFs 3910.1-3910.3 are expressed as disjunctions DISJ1, DISJ2, and DISJ3, respectively, based on being composed of only a single disjunction, being a simple predicate, and/or having no nested conjunctions. The disjunction of these CNFs 39101-39103 is thus a conjunction itself, expressed as CNFS Thus, the conjunction with this conjunction CNF5 with the existing predicate 3951 CNF4 is a new, single conjunction CNF6 rather than being a disjunction of multiple conjunctions. Such an optimization to write a simple pre-materialization filter in this special case of the CNF predicates 3910 can be useful in further simplifying query execution.
In some embodiments, the database system 10 is configured to optimize the processing speed of the Union of CNF's runtime execution. This can include utilizing a different set of potential optimizations than processing one single CNF. Such an optimization can be implemented via flow optimizer module 4914 and/or can otherwise be implemented when generating 10 pipeline 2835. Such an embodiment can be implemented in the cases of some or all examples presented in
In some embodiments, the database system 10 is configured to allow the Union of CNFs in post-materialization filter processing (e.g., in post-data materialization portion 3902). In other embodiments, after data materialization, only I CNF can be processed during a single operation within the query plan. Such an embodiment can be implemented in the cases of some or all examples presented in
In some embodiments, when a plan where DNF evaluation is immediately followed by a UNION ALL with no immediate logic, some or all of the post-IO/materialization operations up to and including the UNION ALL can be deleted safely for example. from the post-data materialization portion 3902 in implementing the rewrite to generate operator execution flow 2817.1 from operator execution flow 28170. Such an embodiment can be implemented in the cases of some of all examples presented in
In some embodiments, instead of rewriting DNFs as a series of CNF, this DNF can instead be implemented as a Union of CNF via a new operation. This runtime operation can perform Union of CNF evaluation mid-query in a single stream of data instead of the multiple streams. Such an embodiment can be implemented in the cases of some or all examples presented in
Some or all of the method of
Some or all of the steps of
Step 4782 includes determining a query for execution. Step 4784 includes generating an IO pipeline based on filtering requirements of the query to implement a disjunction of a plurality of conjunctive normal form (CNF) filtering predicates via a union operator applied to a plurality of parallelized IO pipeline branches that each include at least one serialized IO pipeline operator implementing a corresponding one of the plurality of CNF filtering predicates. Step 4786 includes executing the IO pipeline in accordance with execution of the query to generate a filtered subset of a plurality of rows that each meet at least one of the plurality of CNF filtering predicates based on applying the union operator to the plurality of parallelized IO pipeline branches.
In various examples, the method further includes generating a query operator execution flow for the query that includes a pre-data materialization portion serially before a post-data materialization portion, wherein the pre-data materialization portion is implemented via the IO pipeline: In various examples, executing the query operator execution flow in accordance with execution of the query based on executing the IO pipeline to generate the filtered subset of the plurality of rows, and further based on executing the post-data materialization portion of the query operator execution flow to generate the query resultant for the query based on processing the filtered subset of the plurality of rows via execution of a plurality of operators of the post-data materialization portion of the query operator execution flow.
In various examples, processing the filtered subset of the plurality of rows via execution of the plurality of operators of the post-data materialization portion of the query operator execution flow includes further filtering the filtered subset of the plurality of rows to identify a proper subset of the filtered subset of the plurality of rows.
In various examples, each IO pipeline operator of the at least one serialized IO pipeline operator of each of the parallelized IO pipeline branches is one of: an index element: a source element: or a filtering element.
In various examples, one of the parallelized IO pipeline branches includes a source element to identify a superset of a corresponding set of rows satisfying the corresponding one of the plurality of CNF filtering predicates. In various examples, the one of the parallelized IO pipeline branches further includes a filtering element serially after the source element to identify the corresponding set of rows based on filtering at least one row from the superset that does not satisfy the corresponding one of the plurality of CNF filtering predicates.
In various examples, one of the parallelized IO pipeline branches includes an index element to identify a corresponding set of rows satisfying the corresponding one of the plurality of CNF filtering predicates based on accessing index data for a corresponding, column. In various examples, the index element is applied to access the index data based on identifying rows that satisfy a conjunction of a plurality of disjuncts indicated by the corresponding one of the plurality of CNF filtering predicates.
In various examples, one of the plurality of parallelized IO pipeline branches implements a corresponding one of the plurality of CNF filtering predicates that indicates a conjunction of a plurality of predicates based on including a plurality of sets of IO elements applied in series In various examples, each set of IO elements of the plurality of sets of IO elements implements a corresponding one of the plurality of predicates indicated by the corresponding one of the plurality of CNF filtering predicates Ja various examples, application of the plurality of sets of IO elements in series renders successive filtering of rows to implement the conjunction of the plurality of predicates.
In various examples, one set of IO elements of the plurality of sets of IO elements is implemented via a plurality of parallelized IO sub-branches nested within the one of the parallelized IO pipeline branches based on the corresponding one of the plurality of predicates including a disjunction of a plurality of predicates.
In various examples, each of a set of multiple ones of the plurality of parallelized IO pipeline branches implement a corresponding conjunction of a corresponding plurality of predicates of their corresponding one of the plurality of CNF filtering predicates based on including a corresponding plurality of sets of IO elements applied in series. In various examples, at least two of the set of multiple ones of the plurality of parallelized IO pipeline branches include a plurality of parallelized IO sub-branches based on implementing disjunctions of predicates included in corresponding ones of the corresponding plurality of predicates.
In various examples, generating the IO pipeline is based on including at least one additional IO pipeline operator serially after the union operator is applied to further filter the filtered subset of the plurality of rows.
In various examples, the plurality of CNF filtering predicates each include a corresponding set of predicates to which a conjunction is applied. In various examples, at least one of the corresponding set of predicates is a shared predicate included in all corresponding sets of predicates for all of the plurality of CNF filtering predicates. In various examples, the at least one additional TO pipeline operator applied serially after the union operator implements the shared predicate, and wherein none of the plurality of parallelized IO pipeline branches implement the shared predicate in their respective at least one serialized operator.
In various examples, generating the IO pipeline is based on sharing at least one serialized IO pipeline operator across al least two of the plurality of parallelized IO pipeline branches based on implementing corresponding ones of the plurality of CNF filtering predicates that share a same predicate in respective sets of predicates to which a conjunction is applied.
In various examples, generating the IO pipeline is based on including a source IO pipeline operator serially after the union operator to source column values of a corresponding column for only rows in the filtered subset of the plurality of rows. In various examples, one of the plurality of parallelized IO pipeline branches includes a second source IO pipeline operator to source the column values of the corresponding column for implementing the corresponding one of the plurality of CNF filtering predicates.
In various examples, each of the plurality of parallelized IO pipeline branches implements the corresponding one of the plurality of CNF filtering predicates to apply filtering to a same plurality of input rows processed by all of the plurality of parallelized IO pipeline branches.
In various examples, any one of more of the various examples listed above are implemented in conjunction with performing some or all steps of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database systems, cause the one or more computing devices to perform any or all of the method steps of
In various embodiments a database system includes at least one processor and at least one memory that stores operational instructions. In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to perform some or all steps of
In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to: determine a query for execution; generate an IO pipeline based on filtering requirements of the query to implement a disjunction of a plurality of conjunctive normal form (CNF) filtering predicates via a union operator applied to a plurality of parallelized IO pipeline branches that each include at least one serialized IO pipeline operator implementing a corresponding one of the plurality of CNF filtering predicates, and/or execute the IO pipeline in accordance with execution of the query to generate a filtered subset of a plurality of rows that each meet at least one of the plurality of CNF filtering predicates based on applying the union operator to the plurality of parallelized IO pipeline branches.
Some or all of the method of
Some or all of the steps of
Step 4882 includes determining a query for execution. Step 4884 includes determining a disjunction of a plurality of conjunctive normal form (CNF) filtering predicates based on filtering predicates of the query. Step 4886 includes generating an IO pipeline based on the disjunction of the plurality of CNF filtering predicates that includes, based on applying at least one type of optimization, at least one shared IO element common to implementing of multiple ones of plurality of the CNF filtering predicates. Step 4888 includes executing the IO pipeline in accordance with execution of the query to generate a filtered subset of a plurality of rows that each meet at least one of the plurality of CNF filtering predicates.
In various examples, the disjunction of a plurality of CNF filleting predicates is determined based on extracting a corresponding portion of a query expression for the query written as the disjunction of the plurality of CNF filtering predicates in CNF form. In various examples, the disjunction of a plurality of CNF filtering predicates is determined based on converting a non-CNF portion of the query expression for the query, written in non-CNF form, into the disjunction of the plurality of CNF filtering predicates, for example, in a query request by a corresponding requesting entity. In various examples, the query expression is indicated by/extracted from a query request generated by and/or received from a corresponding requesting entity.
In various examples, the at least one type of optimization includes a time column-based optimization based on the filtering predicates of the query including a time column-based filtering predicate applied to a time column of the plurality of rows storing temporal values. In various examples, the at least one shared IO element includes a single set of IO elements for applying the time column-based filtering serially after a plurality of other IO elements implementing the disjunction of the plurality of CNF filtering predicates based on applying the time column-based optimization.
In various examples, the single set of IO elements for the time column is serially after a plurality of other IO elements based on the filtering predicates indicating a conjunction of the time column-based filtering predicate with the disjunction of the plurality of CNF filtering predicates.
In various examples, the single set of IO elements includes a source element to source the time column, and further includes a filter element serially after the source element to further filter the filtered subset of the plurality of rows based on identifying ones of the filtered subset of the plurality of rows satisfying the time column-based filtering predicate.
In various examples, the time column-based filtering predicate specifies a required time range as a conjunction of multiple range-based predicates.
In various examples, the at least one type of optimization includes a shared source element-based optimization based on the query denoting sourcing of column values for all rows of the plurality of rows satisfying any ones of the plurality of CNF filtering predicates. In various examples, the at least one shared IO element includes at least one source element serially after a plurality of other IO elements implementing the disjunction of the plurality of CNF filtering predicates based on applying the shared source element-based optimization.
In various examples, the plurality of CNF filtering predicates are based on a set of columns, and wherein the at least one source element sources column values of at least one additional column distinct from the set of columns.
In various examples, the plurality of CNF filtering predicates are based on a set of columns, and the at least one source element sources column values of at least one column included in the set of columns. In various examples, the plurality of other IO elements implementing the disjunction of the plurality of CNF filtering predicates implements one of the plurality of CNF filtering predicates via a first source element sourcing a given column of the set of columns, and via a filtering element serially after the first source element to identify only rows having column values for the given column satisfying a corresponding filtering predicate in the one of the plurality of CNF filtering predicates. In various examples, the at least one source element includes a second source element, serially after the first source element and the filtering element, again sourcing the given column of the set of columns based on processing input rows satisfying other ones of the plurality of CNF filtering predicates in addition to processing input rows satisfying the one of the plurality of CNF filtering predicates.
In various examples, the at least one type of optimization includes a shared filtering-based optimization based on the filtering predicates of the query denoting that multiple ones of the plurality of CNF filtering predicates include a same filtering predicate. In various examples, the IO pipeline wherein the at least one shared IO element includes a single set of IO elements implementing the same filtering predicate for use in implementing the multiple ones of the plurality of CNF filtering predicates based on applying the shared filtering-based optimization.
In various examples, the single set of IO elements implementing the same filtering predicate is an index element for accessing index data of a given column to which the same filtering predicate is applied based on the given column having the index data indexing column values of the given column.
In various examples, the single set of IO elements implementing the same filtering predicate includes a source element to source column values for a given column to which the same filtering predicate is applied, and further includes a filter element to filter rows based on identifying ones of the rows sourced via the source element that meet requirements of the same filtering predicate.
In various examples, the plurality of CNF filtering predicates are implemented via parallelized IO pipeline branches of the IO pipeline that each filter a same set of input rows, and wherein the single set of IO elements is shared across the parallelized IO pipeline branches of the multiple ones of the plurality of CNF filtering predicates.
In various examples, the disjunction of the plurality of CNF filtering predicates are applied via the IO pipeline based on a union operator of the IO pipeline applied to a plurality of parallelized IO pipeline branches that each include at least one serialized IO pipeline operator implementing a corresponding one of the plurality of CNF filtering predicates. In various examples, each of the plurality of parallelized IO pipeline branches implements the corresponding one of the plurality of CNF filtering predicates to apply filtering to a same plurality of input rows processed by all of the plurality of parallelized IO pipeline branches.
In various examples, the plurality of CNF filtering predicates each include a corresponding set of predicates to which a conjunction is applied. In various examples, at least one of the corresponding set of predicates is a shared predicate included in all corresponding sets of predicates for all of the plurality of CNF filtering predicates. In various examples, applying the at least one type of optimization includes including at least one additional IO pipeline operator implementing the shared predicate serially after the union operator in the IO pipeline as the at least one shared IO element. In various examples. none of the plurality of parallelized IO pipeline branches implement the shared predicate in their respective at least one serialized operator. In various examples, the shared predicate is a time column-based filtering predicate applied to a time column of the plurality of rows storing temporal values.
In various examples, the filtering predicates of the query indicate multiple ones of the plurality of CNF filtering predicates share a same predicate in respective sets of predicates to which a conjunction is applied. In various examples, applying the at least one type of optimization includes sharing the at least one shared IO element across at least two of the plurality of parallelized IO pipeline branches that implement the multiple ones of the plurality of CNF filtering predicates.
In various examples, the query indicates generation of the query resultant based on column values of at least one column for rows meeting filtering predicates of the query. In various examples, applying the at least one type of optimization includes including at least one source element serially after the union operator as the at least one shared IO element in the IO pipeline to source the column values of the at least one column for rows meeting any of the plurality of CNF filtering predicates. In various examples, one of the plurality of parallelized IO pipeline branches also includes an additional source element to source the column values of a corresponding column included in the at least one column to implement the corresponding one of the plurality of CNF filtering predicates.
In various examples, any one of more of the various examples listed above are implemented in conjunction with performing some of all steps of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps of
In various embodiments, a database system includes at least one processor and at least one memory that stores operational instructions. In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to perform some or all steps of
In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to: determine a query for execution, determine a disjunction of a plurality of conjunctive normal form (CNF) filtering predicates based on filtering predicates of the query: generate an IO pipeline based on the disjunction of the plurality of CNF filtering predicates that includes, based on applying at least one type of optimization, at least one shared IO element common to multiple ones of plurality of the CNF filtering predicates; and/or execute the IO pipeline in accordance with execution of the query to generate a filtered subset of a plurality of rows that each meet at least one of the plurality of CNF filtering predicates.
Some or all of the method of
Some or all of the steps of
Step 4982 includes generating an initial query operator execution flow for a corresponding query for execution that includes a pre-data materialization portion serially before a post-data materialization portion that applies filtering based on a plurality of conjunctive normal form (CNF) filtering predicates. Step 4984 includes converting the initial query operator execution Dow into an updated query operator execution flow semantically equivalent to the initial query operator execution flow based on including filtering for a disjunction of the plurality of CNF filtering predicates within the pre-data materialization portion of the updated query operator execution flow as a pre-data materialization condition. Step 4986 includes executing the updated query operator execution flow in accordance with execution of the corresponding query.
Performing step 4986 can include performing step 4988 and/or step 4990. Step 4988 includes executing the pre-data materialization portion to produce a plurality of data blocks for processing that all meet the pre-data materialization condition. Step 4990 includes executing the post-data materialization portion to generate a query resultant for the corresponding query based on processing the plurality of data blocks via a plurality of operators.
In various examples, the method farther includes determining whether the initial query operator execution flow meets a pre-materialization filter rewrite requirement. In various examples, the initial query operator execution flow is converted into the updated query operator execution flow in response to determining the initial query operator execution flow meets the pre-materialization filter rewrite requirement.
In various examples, the pre-materialization filter rewrite requirement includes: a first requirement that the post-data materialization portion of the initial query operator execution flow implements sending of a same set of data blocks to each of a plurality of parallelized operator branches that collectively implement the plurality of CNF filtering predicates: a second requirement that original data values of the same set of data blocks are unaltered in the sending to the plurality of parallelized operator branches, and/or a third requirement that the same set of data blocks sent to the plurality of parallelized operator branches are unaltered from materialization of the same set of data blocks via the pre-data materialization portion of the of the initial query operator execution flow.
In various examples, the method further includes: generating a second initial query operator execution flow for a second corresponding query for execution: determining whether the second initial query operator execution flow meets the pre-materialization filter rewrite requirement and/or foregoing converting the initial query operator execution flow into an updated query operator execution flow based on applying filtering for the disjunction of the plurality of CNF filtering predicates within the pre-data materialization portion of the updated query operator execution flow based on determining the second initial query operator execution flow fails to meet the pre-materialization filter rewrite requirement.
In various examples, processing the plurality of data blocks in executing the executing the post-data materialization portion of the updated query operator execution flow includes filtering of the plurality of data blocks via the plurality of CNF filtering predicates as indicated in the post-data materialization portion of the initial query operator execution flow.
In various examples, at least one of the plurality of data blocks passing the pre-data materialization condition based on the plurality of CNF filtering predicates is filtered out via at least one of the plurality of CNF filtering predicates of the post-data materialization portion.
In various examples, an arrangement of the plurality of CNF filtering predicates in the initial query operator execution Now is not in accordance with CNF. In various examples, the disjunction of the plurality of CNF filtering predicates in the pre-data materialization portion of the updated query operator execution flow is not in accordance with CNF requirements based on the disjunction of the plurality of CNF filtering predicates.
In various examples, the pre-data materialization portion of the initial query operator execution flow includes an initial pre-data materialization condition based on applying additional CNF filtering predicate distinct from the plurality of CNF filtering predicates. In various examples, the pre-data materialization condition of the pre-data materialization portion of the updated query operator execution flow is based on applying a conjunction of the additional CNF filtering predicate with the disjunction of the plurality of CNF Altering predicates.
In various examples, converting the initial query operator execution flow into the updated query operator execution flow includes generating the pre-data materialization portion of the updated query operator execution flow to apply the pre-data materialization condition via applying a corresponding disjunction to a corresponding plurality of CNF filtering predicates. In various examples, the corresponding disjunction of the corresponding plurality of CNF filtering predicates is semantically equivalent with the conjunction of the additional CNF filtering predicate with the disjunction of the plurality of CNF filtering predicates. In various examples, each of the corresponding plurality of CNF filtering predicates is a conjunction of a corresponding one of the plurality of CNF filtering predicates with the additional CNF filtering predicate.
In various examples, each of the plurality of CNF filtering predicates includes a single corresponding disjunction. In various examples, converting the initial query operator execution flow into the updated query operator execution flew includes generating the pre-data materialization portion of the updated query operator execution flow to apply the pre-data materialization condition as the conjunction of the additional CNF filtering predicate with the disjunction of the plurality of CNF filtering predicates. In various examples, the pre-data materialization condition is a single CNF filtering predicate in accordance with CNF requirements based on being implemented as a conjunction of CNF filtering predicates.
In various examples, converting the initial query operator execution flow into the updated query operator execution flow includes generating an IO pipeline implementing the pre-data materialization portion of the updated query operator execution flow to apply the pre-data materialization condition via a anion operator applied to output of a plurality of IO pipeline branches that each include at least one IO pipeline operator.
In various examples, any one of more of the various examples listed above are implemented in conjunction with performing some or all steps of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database systems, cause the one or more computing devices to perform any or all of the method steps of
In various embodiments a database system includes at least one processor and at least one memory that stores operational instructions. In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to perform some or all steps of
In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to: generate an initial query operator execution flow for a corresponding query for execution that includes a pre-data materialization portion serially before a post-data materialization portion, where the post-data materialization portion applies filtering based on a plurality of conjunctive normal form (CNF) filtering predicates: convert the initial query operator execution flow into an updated query operator execution flow semantically equivalent to the initial query operator execution flow based on including filtering for a disjunction of the plurality of CNF filtering predicates within the pre-data materialization portion of the updated query operator execution flow as a pre-data materialization condition; and/or executing the updated query operator execution flow in accordance with execution of the corresponding query based on executing the pre-data materialization portion to produce a plurality of data blocks for processing that all meet the pre-data materialization condition and/or based on executing the post-data materialization portion to generate a query resultant for the corresponding query based on processing the plurality of data blocks via a plurality of operators.
In some embodiments, the database system: 10 can be operable to store compressed columns for one or more tables. where the data stored in these compressed column are compressed fixed-length values generated from the original (e.g. variable-length) values of the corresponding column via a compression scheme In the case where query expressions for execution indicate initial filtering expressions 5020 for compressed columns of the database system, a compressed column filter conversion module can generate an updated filtering expression 5021 (e.g. semantically equivalent to the initial filtering expressions 5020 to render correct execution of the corresponding query. These updated filtering expressions can be generated based on reducing and/or eliminating the need to generate/access the original, decompressed values of the corresponding column, where the filtering expression is updated to render semantic equivalence with the initial filtering expressions 5020 via access to the compressed values of the corresponding column in the compressed column. Some of all features and/or functionality of
In various embodiments, one or more columns of one or more relational database tables/one or more datasets are compressed via a compression scheme. For example, these columns are compressed during data ingress, where segments are generated to include compressed values for these columns, and where the decompressed values are recoverable in other memory and/or require performance of any decompression algorithm/process in accordance with the compression scheme.
SQL query commands can request filtering based on these compressed columns of the database system. In order to evaluate an arbitrary filter condition dependent upon a compressed column, the query processing system can initially plan to decompress the required column data mid-query (e.g. as the compressed values are accessed via row reads, where these compressed values are then decompressed in accordance with the compression scheme. However, the process of decompression can take time and memory, and the decompressed data takes up more memory and is slower to process than compressed data.
As discussed in conjunction with
Database storage 2450 that includes the records 2422 (e.g. rows) accessed during query executions via query execution module 2504 can store one or more compressed columns 5005 for one or more of its relational database tables/data sets. Some or all columns 2707 of some or all database tables 2712 can be compressed columns 5005, where their respective values 2708 are compressed values (e.g. fixed-length integer values or another type of compressed value) generated from their original, uncompressed values (e.g. variable-length values). For example a given column 2707 is stored as a compressed column 5008 based on storing string values, storing any type of variable-length values: storing large values/corresponding to a large datatype. having low cardinality/a large number of duplicated values where storage of each unique, decompressed value elsewhere is manageable: storing a datatype that is not able/configured to be compressed in accordance with the compression scheme: or other reasons. Some columns 2707 are optionally uncompressed, for example, based on: its values 2708 already being of fixed-length: having high cardinality: storing a datatype that is not able/configured to be compressed in accordance with the compression scheme: or other reasons.
When such compressed columns are indicated in filtering expressions 5020 in query expressions 2511 for execution (e.g. requested by a user/received from a client device corresponding to a requesting entity), the conditions by which rows be filtered based on compressed columns can always/typically be based on comparison with literal values that are not compressed (e.g, literal values that include one or more uncompressed values corresponding to possible values/the datatype of the original, decompressed values of the given column prior to compression). In particular a given user requesting the query may not be aware of whether the given column is compressed, as column compression can be an internal storage/processing optimization by the database system 10 that need not be known communicated to users requesting queries. Alternatively or in addition, as a compressed column is still treated as storing its original uncompressed value despite its values being compressed in storage, queries against these columns that correctly filter the columns can be required to be/processed as if being written in the accordance with comparison to literals/values in the uncompressed form.
Thus a filtering expression 5020 indicating filtering operators 5022 be applied to filtering operands 5023.0, which can indicate one or more filtering operators 5022 be applied to values of a given compressed column 5005. 1 (and/or one or more additional compressed columns and/or uncompressed columns) and one or more uncompressed values 5012. For example, the uncompressed values 5012 are implemented as and/or within one or more literals 5011.0 of the filtering expression 5020, where the literal values of literal 5011.0 are explicitly written/included in the query expression 2511. The filtering operators 5022 denoting a type of comparison of a given value 5008 of a given row of compressed column to the uncompressed values 501210 dictate whether the given row be included/excluded from the filtered for set 5033. The filtering operators/uncompressed values can be implemented via any type of filtering operators and/or literals/values/filtering conditions applied in query filtering predicates. where in this case, they are applied to a compressed column 5005. Note that, as the uncompressed columns are specified, correct execution of the given filtering expression as written would require that the values of the compressed column 5005 be first decompressed into corresponding decompressed values for each row, where the decompressed values are then compared with the uncompressed values 5012 (e.g. a literal in the expression) in accordance with the specified filtering operators 5022.
The compressed column filter conversion module 5010 can be implemented to generate a corresponding updated filtering expression 5021 from the given, initial filtering expression 5020 expressly included in and/or derived from the query expression 2511, for example, as part of implementing an optimizer upon the query expression 2511 in conjunction with generating/optimizing the corresponding query operator execution flow 2517 for execution. The updated filtering expression 5021 can be semantically equivalent to the initial filtering expression 5020 where, instead of requiring comparison of decompressed values of the compressed column with the uncompressed values, the compressed values of the compressed column are compared with corresponding compressed values 5013. This can be based on replacing the one or more uncompressed values 5012 in the literal 50110 with corresponding compressed values 5013 in a modified literal 5011.1.
These one or more compressed values 5013 can be determined based on determining which values map to the one or more compressed values 5012 in corresponding compression scheme data 5015, denoting the compression scheme by which the compressed column 5005.1 is compressed. For example a given uncompressed values 5012 of literal 5011.0 is replaced with a corresponding compressed value 5013 in literal 5011.1 based on determining which compressed value 5013 maps to the given uncompressed values 5012 (e.g., if the uncompressed values 5012 were to be compressed as a compressed value 5008 in the compressed column 5005 by applying the compression scheme denoted in compression scheme data 5015, its value would be this given compressed value 5013, and/or if this given compressed value 5013 were to be decompressed, its value after decompression would be the uncompressed value 5012). Thus, the corresponding compressed values 5013 were optionally never expressed explicitly in the query expression and are instead derived based on applying the compression scheme data 5015 to the uncompressed values 5012.
The updated filtering expression 5021 can optionally have other changes, for example, as required to render semantic equivalence, as necessary to compensate for the use of compressed values 5013 rather than uncompressed values 5012 as its one or more literals 5011. The filtering operators 5022. I can optionally be different from the filtering operators 5022.0. Note that some or all changes in filtering operators 5022.1 can alternatively or additionally be based on applying other forms of optimization/rewritting of the query expression, optionally independently of any needs induced by the conversion of literal values. For comparison with the compressed column. Examples of various means of generating updated filtering expressions 5021 from particular types of initial filtering expressions 5020 are described in further detail in conjunction with
As illustrated in
As illustrated in
The filtered row set 5033 can be a proper subset of the row set 5032 when one or more rows were filtered out based on only some rows of the row set 5032 satisfying the requirements of the filter. The filtered row set 5033 can be equivalent to the row set 5032 ne rows were filtered out based on all rows of the row set 5032 satisfying the requirements of the filter. The filtered row set 5033 can be a null set based on all rows being filtered out based on based on no rows of the row set 5032 satisfying the requirements of the filter. The filtered row set 5033 can be included in the query resultant, and/or can further processed (e.g. further filtered/manipulated/combined with other sets) to ultimately generate the query resultant.
In some embodiments, columns are compressed as compressed columns 5005 in accordance with implementing some or all features or functionality of whole column compression of
In some embodiments, columns are compressed as compressed columns 5005 based on a globally maintained dictionary (e.g, dictionary structure 5016), for example, in conjunction with applying Global Dictionary Compression (GDC). Applying Global Dictionary Compression can include replaces variable length column values with fixed length integers on disk (e.g., in database storage 2450), where the globally maintained dictionary is stored elsewhere, for example, via different (e.g. slower/less efficient) memory resources of a different type/in a different location from the database storage 2450 that stores the compressed columns 5005 accessed during query execution.
The dictionary structure can store a plurality of fixed-length, compressed values 5013 (e.g., integers) each mapped to a single uncompressed value 5012 (e.g. variable-length values, such as strings). The mapping of compressed values 5013 to uncompressed values 5012 can be in accordance with a one-to-one mapping. The mapping of compressed values 501310 uncompressed values 5012 can be based on utilizing the fixed-length values 5013 as keys of a corresponding map and/or dictionary data structure, and/or can be based on utilizing the uncompressed values 5012 as keys of a corresponding map and/or dictionary data structure.
A given uncompressed value 5012 that is included in many rows of one or more tables can be replaced (i.e. “compressed”) via a same corresponding compressed value 5013 mapped to this uncompressed value 5012 as the compressed value 5008 for these rows in compressed column 5005 in database storage. As new rows are received for storage over time, their column values for one or more compressed columns 5005 can be replaced via corresponding compressed values 5008 based on accessing the dictionary structure and determining whether the uncompressed value 5012 of this column is stored in the dictionary structure 5016. If yes, the compressed value 5013 mapped to the uncompressed value 5012 in this existing entry is stored as compressed value 5008 in the compressed column 5005 in the database storage 2450. If no, the dictionary structure 5016 can be updated to include a new entry that includes the uncompressed value 5012 and a new compressed value 5013 (e.g, different from all existing compressed values in the structure) generated for this uncompressed value 5012, where this new compressed value 5013 is stored as is applied as compressed value 5008 in the database storage 2450.
The dictionary structure 5016 can be stored in dictionary storage resources 2514, which can be different types of resources from and/or can be stored in a different location from the database storage 2450 storing the compressed columns for query execution. In some embodiments, the dictionary storage resources 2514 storing dictionary structure 5016 can be considered a portion/type of memory as of database storage 2450 that are accessed during query execution as necessary for decompressing column values. In some embodiments, the dictionary storage resources 2514 storing dictionary structure 5016 can be implemented as metadata storage resources, for example, implemented by a metadata consensus state mediated via a metadata storage cluster of nodes maintaining system metadata such as GDCs of the database system 10.
The dictionary structure 5016 can correspond to a given column 5005, where different columns optionally have their own dictionary structure 5016 build and maintained. Alternatively a common dictionary structure 5016 can optionally be maintained for multiple columns of a same table/same dataset, and/or for multiple columns across different tables/different datasets. For example a given uncompressed value 5012 appearing in different columns 5005 of the same or different table is compressed via the same fixed-length value 5013 as dictated by the dictionary structure 5016.
In some embodiments, dictionary compression via dictionary structure 5016 can implement the compression scheme utilized to generate (e.g. compress/decompress the values of) compressed columns 5005 of
In some embodiments, dictionary compression via dictionary structure 5016 can implement the compression scheme utilized to generate (e.g. compress/decompress the values of) compressed columns 5005 of
In some embodiments, dictionary compression via dictionary structure 5016 can implement the compression scheme utilized to generate (e.g. compress/decompress the values of) compressed columns 5005 of
The compressed column filter conversion module 5010 can generate updated filtering expression 5021 based on updating one or more literals 5011.1 from corresponding literals 5011.0 based on replacing uncompressed values 5012 with compressed values 5013 mapped to these compressed values based on accessing dictionary structure 5016 and determining which fixed-length compressed value 5013 is mapped to each given uncompressed value 5012.
In some embodiments, the rewriting of filtering expression 5020 an updated filtering expression 5021 via generic rewrite-based filter conversion 5051 can be based on attempting to rewrite all filters of the form {{decompressed column FILTER_OPERATION [literal containing decompressed literal values]} } by replacing the decompressed literal value(s) with their compressed versions. This rewrite can frequently result in completely discarding the need to decompress the columns at any point during the query execution Filter expressions on compressed, smaller data can generally evaluate faster than their decompressed equivalents. Since decompressed data does not exist on disk, these compressed filters can also be applied before data materialization and benefit from available indexes.
As illustrated in
This direct replacement of literal 5011 with the compressed values mapped to the original uncompressed values in generating updated filtering expression 5021, without changing other attributes of the expression (e.g. without changing filtering operators 5022) can render semantically equivalence in some cases, but optionally not all cases. For some types of queries, the filtering operators 5022 themselves are also updated, as discussed in further detail with other examples described herein.
As illustrated in
In other embodiments, full value of literal 5014 is not itself a compressible data type 5036.A1, but instead stores multiple compressible data type 5036.A1 as a corresponding container-type data structure. Rather than the compression scheme data denoting a mapping of the literal to a compressed value in its entirety, individual values within the literal are themselves compressed into compressed values, for example, within a corresponding container-type.
In some embodiments, the compressed column 5005 similarly stores array structures 5042 that are compressed in this fashion of uncompressed values (e.g. strings) of array element being compressed into corresponding compressed values (e.g. integers mapped to the strings). For example, the array-based filtering expression 5041 is based on comparing array structures of compressed column 5005 with the array structure 5042 of literal value 5014.
As a particular example, array columns can utilize GDC via replacing arbitrarily deep non-array/non-tuple elements of the array with compressed integers. As such the rewrites for array filters follows the same pattern as non-container filter expressions, but the literal elements inside of the array are replaced with their compressed versions instead of the entire array literal value being replaced.
In other embodiments, the compressed column 5005 stores a different data type from array structures 5042. For example. the compressed column 5005 stores compressible data type 5036.A1, which are compressed auto compressed values 5013 (e.g. compressed column 5005 is a column of strings). For example, the array-based filtering expression 5041 is based on comparing the values of compressed column 5005 with the array elements of array structure 5042 of literal value 5014 (e.g. filter based on whether the column value is included as an army element of army structure implementing literal value 5014).
As illustrated in the embodiment of
As illustrated in example of
As illustrated in this example, the multiple data types of the tuple optionally include multiple compressible data types. including compressible data type 5036.A1 and compressible data type 5036. A2. The compression scheme data 5015 can denote mapping of multiple different types into fixed-length values, utilized to rewrite the literal value 5014.1 from 5014.0 as having corresponding compressed values 5013 for each uncompressed value 5012. For example a same dictionary structure 5005 and/or other common scheme is applied, where all fixed-length values are unique across the different datatypes based on being applied to multiple different types of variable-length values/multiple different compressible data types. Alternatively, a different dictionary structure 5005 and/or separate scheme is applied for different data types, where a given fixed-length value is optionally mapped to multiple given uncompressed values of different types.
As illustrated in example of
As illustrated in example of
In some embodiments, the compressed column 5005 similarly stores tuple structures 5044 that are compressed in this fashion, while maintaining the structure/ordering of the tuple, where uncompressed values (e.g. strings, other variable-length values, etc.) included/nested in elements of the tuple stricture 5044 are compressed into corresponding compressed values (e.g. integers mapped to the strings/other compressible data types), and where other values having other, non-compressible types that are not compressed in compression scheme (e.g., integers, chars, binary values, other fixed-length values, etc.) included/nested in elements of the tuple structure 5044 are not compressed into corresponding compressed values, and instead maintain their value in their respective position of the tuple structure 5044 of the compressed column for each corresponding row. For example, the tuple-based filtering expression 5043 is based on comparing tuple structures 5044 of compressed column 5005 with the tuple structure 5044 of literal value 5014.
As a particular example, tuple columns can utilize GDC via replacing arbitrarily deep non-array/non-tuple elements of the tuple with compressed integers. As such, the rewrites for army filters can follows the same pattern as non-container filter expressions, but the literal elements inside of the tuple are replaced with their compressed versions instead of the entire tuple literal value being replaced.
In other embodiments, the compressed column 5005 stores a different data type from array structures 5042. For example. the compressed column 5005 stores compressible data type 5036.A1, which are compressed into compressed values 5013 (e.g. compressed column 5005 is a column of strings). For example, the array-based filtering expression 5041 is based on comparing the values of compressed column 5005 with the array elements of array structure 5042 of literal value 5014 (e.g. filter based on whether the column value is included as a given tuple element of tuple structure implementing literal value 5014).
While
Global dictionary compressed columns can provide complete knowledge of all values present in a column upon query submission. For example, global dictionaries provide a perfect snapshot of the available, unique values of a column. By the nature of the compression, the total number of unique values can be expected/guaranteed to be rather low. This knowledge can be leveraged to enable rewriting of some GDC filters to always true or always false. In other embodiments, other information regarding which values are included in a given compressed column at a given time can be determined/maintained in cases where other compression schemes different from GDC are applied to enable the functionality of
Determining whether the some or all uncompressed values 5012.x of literal value 5014.0 exist in the corresponding compressed column 5005 can be based on accessing other portions of/information indicated by compression scheme data 5015. For example, the compression scheme data 5015 can otherwise denote the full set of compressed values 5013 that are included in one or more rows of a given column 5005, and/or can denote the full set of uncompressed values 5012 that are compressed as compressed values in one or more rows of a given column 5005.
If the corresponding compressed value 5013.x does not exist in the given column the updated filtering expression 5021 can be rewritten as always FALSE or always TRUE, for example, based on this rendering semantic equivalence with the given filtering expression 5020, as denoted by the initial filtering operators 5022, for example, based on the initial filtering operators 5022 denoting an equality-based condition, contains-based condition, and/or overlap-based condition. If the value does exist in the given column, such optimization is not applied, where the updated filtering expression 5021 is generated via another means of conversion. This can include replacing of literal value 5014.1 with compressed value 5013.x determined to be mapped to the uncompressed value 5012.x, based on compressed value 5013.x being determined to be included in the corresponding column 5005. This can optionally further include updating the filtering operators 5022.1 from filtering operators 5022.0 as necessary. and/or keeping the filtering operators 5022. 1 the same as filtering operators 5022.0 is semantically equivalent and/or if optimal.
As a particular example, updated filtering expression 5021 can be rewritten to always false when the filter checks for equality of the literal value 5014 with the values of column 5005, but the literal value 5014 does not exist in the column 5005 (and thus all rows are guaranteed to fail this condition). As another particular example, updated filtering expression 5021 can be rewritten to always false when the filter is an array-based filter checking for array equality between array structures of compressed column 5005 with an array structure of the literal 5014, but at least one of the literal array values do not exist in the column. As another particular example, updated filtering expression 5021 can be rewritten to always false when the filter is an array-based filter checking whether array structures of compressed column 5005 contain a single value implementing given literal 5014, or all values of a set of values (e.g., of an array structure) implementing given literal 5014, but the given literal 5014, and/or at least one of the literal array element values of the array structure implementing the literal, do not exist in the column. As another particular example, updated filtering expression 5021 can be rewritten to always false when the filter is an array-based filter checking whether values of compressed column 5005 are contained within an array structure implementing given literal 5014, but none of the literal array element values of the literal array structure exist in the column. As another particular example, updated filtering expression 5021 can be rewritten to always false when the filter checks for array overlap between array structures of compressed column 5005 with an array structure of the literal 5014, but none of the non-null literal values within the array exist in the column.
Some cases of the values of literal 5014 not existing in the column (e.g. negations applied to the above cases, or other cases) could be processed similarly to render the expression always true. Similar optimizations can be applied for similar cases applied to other container types (e.g. tuples) and/or any other cases where one or more values denoted by the literal 5014 being known to not exist in the column would render all rows as being guaranteed to pass/fail the respective expression
Cases where an always true/always false rewrite is applied can be ideal in skipping costly time spent evaluating filters. as well as further optimize the rest of the plan based on the knowledge that no, or all, rows will be propagated upstream as output of the filter, improving the technology of database systems by further improving query execution efficiency.
In some embodiments a potential rewrite relies on a value that does not exist yet in the global dictionary (e g. {GDC col NOT EQUALS unseen_value}}). When this occurs, and/or when the amount of key-value pairs that must be add the dictionary is stall a new key-value entry for that value can be added to the dictionary, and this compressed key can be utilized immediately.
In some embodiments, the dictionary structure 5016 (and/or other portions of compression scheme data 5015) differentiate such values from values actually included in the column 5005 (e.g. when knowledge of which values are included in a column is leveraged for rewrites to always true or always false as discussed in conjunction with
The literal values 5014.0 can indicate one or more range and/or regex values 5019 (e.g. upper and/or lower bounds on a value range for ordered datatype values: a text pattern/text substring that strings must match with, must include; etc.).
In some embodiments, the compressed values are not necessarily ordered and/or otherwise would net render range/regex operations being applied correctly if applied to the compressed values rather than the uncompressed values. Alternatively in addition, the nature of the number of unique values being small can enable simplification of a corresponding expression based on testing for equality with a small set of values that correspond to all values included in rows for the column meeting the respective range/regex filter.
The dictionary stricture search module 5038 can be applied to identify which uncompressed values included within the column 5005.1 meet the corresponding condition (e.g., the range condition based on having values falling within the specified bounds, and/or the regex condition, based on having values that match the text pattern/include the specified substring). This can be based on searching the dictionary structure 5016 for such values and/or otherwise accessing compression scheme data 5015 to determine, from the set of values that exist in the column, which of these values meet the specified condition. Thus accessing can be performed in a same or similar fashion as discussed in conjunction with
This search can render a satisfying compressed value set 5033 indicating all compressed fixed-length values 5013 (e.g., in this case. value 5013.a, 5013.b, etc.) that correspond to all uncompressed values 5012 that exist in the column 5005.1 and also meet the condition. A satisfying set-based conversion 5047 can be applied to generate updated filtering expression 5021 a disjunction of equality predicates, where each predicate requires equality with a corresponding compressed value 5013 in the satisfying compressed value set 5033, and where all values in the satisfying compressed value set 5033 have a corresponding equality based predicate in the disjunction. Thus, when applied to the column 5005.1 a given row passes if its value matches one of these values, and is thus included in the range/meets the regex condition. The disjunction can be written via a plurality of disjunction operators 5053 (e.g. “OR”. “[ ]”, etc.) applied to the set of predicates, where each predicate is written via applying an equality operator 5058 (e.g. “=”. “EQUALS”. “==”, etc.) to the column 5005. 1 and a corresponding compressed value 5013 in the satisfying compressed value set 5033.
As a particular example, the expression “colA <=5” to be applied to the decompressed values of a given compressed column colA could be rewritten to “colA=3| colA=5” when applied to the corresponding compressed column, based on 3 and 5 being determined to be the only uncompressed values in colA that are less than or equal to 5. The values of 3 and 5 could then be replaced with the respective compressed values for these uncompressed values, where the final rewrite can correspond to “colA=[compressed value mapped to 3] | colA=[compressed value mapped 105]” to reduce the need to access the uncompressed values in the dictionary structure as discussed previously. Note that while this case expressed the literals as small integer values. these values could optionally be implemented as values of a variable-length data type that is compressed in compressed column 5005, and/or otherwise serves as an example for how this mechanism of optimization be applied to other datatypes that are compressed.
As another example, the expression “colB LIKE ‘str %’ to be applied to the decompressed values of a given compressed column colB could be rewritten to “colB=‘straight’| colA=‘strict’” when applied to the corresponding compressed column. based on ‘straight’ and ‘strict’ being determined to be the only uncompressed values in colB that stad with the substring str (e.g. where ‘%’ is a wildcard special character) The values of ‘straight’ and ‘strict’ could then be replaced with the respective compressed values for these uncompressed values, where the final rewrite can correspond to “colB=[compressed value mapped to ‘straight] ∥ colB=[compressed value mapped to ‘strict’]” to reduce the need to access the uncompressed values in the dictionary structure as discussed previously. This case can optionally be applied based on colB storing strings that are compressed based on being variable length values.
In some embodiments, rewrites dependent upon knowledge of values existing within a compressed column, such as the example cases of
Inequality filters (e.g., via applying a NOT EQUALS operator) on nullable columns can implicitly filter out NULLs, for example, according to the SQL standard. In some embodiments compressed columns 5005 in their compressed state are never nullable—a special, con-null null-mapped value 5013.0 (e.g., the integer “0”, or another non-null value reserved for mapping exclusively to NUL.L.) can represents the NULL value for additional compression, where no non-null values of the column are mapped to this value 5013.0 reserved for null. As such the initial filter expression “decompressed_column NOT EQUALS decompressed_value” is not equivalent with “compressed_column NOT EQUALS compressed_value”, as the latter expression will include rows with value 0 in its filtered row set ourpur, despite these values needing to be filtered out to render semantic equivalence with the first expression, due to how the value of NULL, is treated, and due to the fact that an integer value of zero is not implicitly treated this way when relying upon SQL rules alone.
Thus, the simple replacement of the uncompressed value 5012.x of literal 5014.0 with compressed value 5013.x is not sufficient in rendering semantic equivalence alone, and the filtering expression must further be updated. In particular, a conjunction (e.g. “AND”. ‘&&’, etc.) with an additional predicate requiring that the column value is also not equal to the reserved null value can be applied, as illustrated in the example of
In the case where the conjunction is applied as discussed in conjunction with
In some embodiments, if this expansion adds too many additional individual filter expressions, the rewrite will not be applied due to the inefficiency of evaluating large filter expressions. For example a tradeoff of efficiency in applying the compressed columns directly without needed to read the decompressed values vs. needing to evaluate a large expression due to CNF expansion can be applied, where the rewrite via using compressed values is foregone in some cases when the number of expansions and/or expected execution time and/or other metric resulting from expansion of the expression into CNF due to the conjunction with the predicate testing for inequality with the non-mall value (or other expansion required for other conversions for other expressions) compares unfavorably to a corresponding inefficiency threshold (e.g. exceeds a threshold max #conversions. threshold max estimated execution time, threshold max #parallelized tracks, etc.).
Further examples of compensating for null-values can be relevant for other types of operators. Special cases for handing pull operators for existential quantifier logic and/or universal quantifier logic is discussed in conjunction with
Filters can be wrapped in a for some or for all operator that evaluates a filter expression for each value in an array, for example, to implement a corresponding existeunal quantifier operation or a corresponding universal quantifier operation. respectively. As used herein. “for all” can correspond to a type of universal quantifier operator evaluating a universal quantifier condition, where a row passes the filter only if all of its array elements of an array stricture for a corresponding array column meet the respective condition. As used herein. “for_some” can correspond to a type of existential quantifier operator evaluating an existential quantifier condition, where a row passes the filter only if at least one of its element of an array structure for a corresponding array column meet the respective condition. Different keywords for such operators implementing this functionality can be applied in other embodiments.
The generic rewrite logic discussed previously can be applied to these types of filters as well, where the operators need not be changed and only the literals are changed as discussed in conjunction with
Filters of the form “for_some(compressed array column) NOT EQUALS non-array literal” can implicitly filter out NULLs. Simply re-using the distributive law as applied for the not equals case discussed in conjunction with
The “DOES NOT ONLY CONTAIN” operator can be implemented as type of negation of a universal quantifier operator requiring not every value is one of a set of values in the corresponding array (e.g. thus implicitly being implemented as an existential operator of the negated condition, requiring at least one value is not one of the set of values.
In this case, the array includes only the compressed value 5013.x mapped to the uncompressed values 5012.x and the null-mapped value 5013.0, where the condition denoted in array-based filtering expression 50411 thus requires at least one given value of the column's array structure is not equal to the compressed value 5013.x mapped to the uncompressed values 5012.x, and this given value is also not equal to the null-mapped value 5013.0 (e.g., the integer value of 0, or other non-null fixed-length value that is mapped to NULL in the compression scheme data 5015). The intention in this case is to fail the filter if the compressed row only contains the matching GDC key (e.g, mapped compressed value 5013.x) or the compressed key NULL value (and thus none of its uncompressed array values are not equal to the value 5012.x and also not equal to null). This further filters the NULL case appropriately, rendering semantic equivalence with the array-based filtering expression 5041.0.
Filters of the form “for all(compressed array column) EQUALS/RANGE/REGEX non-array literal” can rewritten as “compressed array column ONLY CONTAINS array [all passable compressed literal values]”. The passable compressed literal values can be identified as discussed in conjunction with
The “ONLY CONTAINS” operator can be implemented as type of universal quantifier operator requiring every value is one of a set of values in the corresponding army. Thus, in this case, if any of the compressed values 5013 in the satisfying compressed value set 5033 array are equal to each given array element of the given row's army structure of the given column 5005.1, the filter passes as required.
In some embodiments, similar logic can be applied to handle the existential quantifier case. For example. filters of the form “for some(compressed array column) EQUALS/RANGE/REGEX non-array literal” can rewritten as “compressed array column CONTAINS array[all passable compressed literal values]”, where the array structure implementing the updated literal 5014.1 is similarly generated to include all of the compressed values 5013 in the satisfying compressed value set 5033 identified via dictionary stricture search module 5038 and/or other means of searching respective portions of compression scheme data 5015 as discussed in conjunction with
The “CONTAINS” operator can be implemented as type of existential quantifier operator requiring at least one value is one of a set of values in the corresponding array. Thus, in this case, if any of the compressed values 5013 in the satisfying compressed value set 5033 array are equal to at least one given array element of the given row's array structure of the given column 5005.1, the filter passes as required.
Null array structures (e.g. array structures 5042 that are null, array structures 5042 that only contain the value null, and/or array structures 5042 that are empty), to which a for all/for some operator is applied in an array-based filtering expression 5041.0 can be rewritten as an array 5042. 1 that includes the null-mapped value 5013.0 as its single value. This can be necessary in handling the special properties of NULL and ensuring that all rows are filtered correctly. In particular because literal NULLs have special properties, the generic rewrite cannot be applied in this case. This means of instead rewriting the literal null as a non-null array containing the compressed value mapped to NULL as as singular value can render semantic equivalence, for example based on the filter basically decompose into “non-array GDC column [NOT] EQUALS 0”. For example, because the array for_some/all iterates over only has I value, it guarantees that the NULLs that become 0 s match on either side.
For example, filters of the form “for_some (NULL array literal) EQUALS non-array decompressed column” can be rewritten as “for_some(array [0]) EQUALS non-army column” in the case where the null-mapped value 5013.0 is equal to zero.
As another example, filters of the form “for_all (NULL array literal) EQUALS non-array decompressed column” can be rewritten as “for all(array[0]) EQUALS non-array column” in the case where the null-mapped value 5013.0 is equal to zero.
As another example, filters of the form “for_some (NULL array literal) NOT EQUALS non-array decompressed column” can be rewritten as “for_some(array [0]) NOT EQUALS non-array column” in the case where the null-mapped value 5013.0 is equal to zero.
As another example, filters of the form “for_all (NULL array literal) NOT EQUALS non-array decompressed column” can be rewritten as “for all(army[0]) NOT EQUALS non-array column” in the case where the null-mapped value 5013.0 is equal to zero.
Some or all of the method of
Some or all of the steps of
Step S082 includes determining a query for execution indicating filtering of a set of rows based on at least one uncompressed value for a compressed column of the set of rows. Step S084 determining an initial filtering expression for the query indicating the at least one uncompressed value for the compressed column as a literal value of the initial filtering expression. Step S086 includes converting the initial filtering expression for the query into an updated filtering expression for the query indicating filtering of the set of rows based on at least one compressed fixed-length value mapped to the at least one uncompressed value in accordance with a compression scheme, for example, where the updated filtering expression is semantically equivalent to the initial filtering expression. Step S088 includes identifying a filtered subset of the set of rows in conjunction with executing the query based on applying the updated filtering expression.
In various examples, the updated filtering expression is semantically equivalent to the initial filtering expression based on applying of the updated filtering expression to the compressed column being semantically equivalent to applying of the initial filtering expression to corresponding uncompressed values of the compressed column due to the corresponding uncompressed values being mapped to the compressed fixed-length values in accordance with the compression scheme.
In various examples, the compressed column is stored as a plurality of compressed fixed-length values in first storage resources. In various examples, the plurality of compressed fixed-length values are mapped to a plurality of uncompressed column values of the compressed column in a dictionary structure stored via second storage resources in accordance with a dictionary compression scheme. In various examples, converting the initial filtering expression into the updated filtering expression includes determining the at least one compressed fixed-length value is mapped to the at least one uncompressed value based on accessing the dictionary structure in the second storage resources.
In various examples, applying the updated filtering expression in conjunction with executing the query to filter the set of rows based on the at least one compressed fixed-length value is based on: accessing compressed fixed-length values of the set of cows in the first storage resources, and/or comparing the compressed fixed-length values of the set of rows to the at least one compressed fixed-length value of the updated filtering expression to identify the filtered subset of the set of rows.
In various examples, an index structure indexing the plurality of compressed fixed-length values of compressed column is stored via third storage resources. In various examples, applying the updated filtering expression in conjunction with executing the query to filter the set of rows based on the at least one compressed fixed-length value is based on identifying the filtered subset of the set of rows based on accessing the at least one compressed fixed-length value in the index structure. In various examples. the third storage resources are shared storage resources with and/or a same type of storage resources as the first storage resources.
In various examples, evaluation of the initial filtering expression requires decompression of compressed fixed-length values of the compressed column based on the initial filtering expression indicating the at least one uncompressed value. In various examples, evaluation of the updated filtering expression requires no decompression of any compressed fixed-length values of the compressed column based on the updated filtering expression indicating the at least one compressed fixed-length value mapped 10 the at least one uncompressed value in the dictionary structure. In various examples, executing the query includes no decompression of compressed fixed-length values of the compressed column based on applying the updated filtering expression.
In various examples, the compression scheme is based on compressing variable-length: values into a corresponding fixed-length integer value. In various examples, the variable-length values include string values. In various examples, the variable-length values include multiple different data types.
In various examples, the literal value of the initial filtering expression is an array that includes the at least one uncompressed value as at least one array element. In various examples, converting the initial filtering expression into the updated filtering expression includes replacing the at least one uncompressed value of the array with the at least one compressed fixed-length value mapped to the at least one uncompressed value.
In various examples, the literal value of the initial filtering expression is a tuple that includes the at least one uncompressed value as at least one tuple element. In various examples, converting the initial filtering expression into the updated filtering expression includes replacing the at least one uncompressed value of the tuple with the at least one compressed fixed-length value mapped to the at least one uncompressed value.
In various examples, the tuple further includes at least one additional value that is not compressed. In various examples. converting the initial filtering expression into the updated filtering expression includes maintaining the at least one additional value that is not compressed based on no compressed fixed-length values being mapped to the at least one additional value. In various examples, the at least one additional value includes one or more fixed-length values of one or more fixed-length data types. In various examples, the at least one uncompressed value was compressed into the compressed fixed length values via the compression scheme, and the at least one additional value was not.
In various examples, the tuple includes at least one container structure (e.g. an array/list/set, embedded tuple, etc.) as one of its values/elements. In various examples, one or more uncompressed values of the at least one uncompressed values are included in the at least one container structure of the tuple. In various examples, converting the initial filtering expression into the updated filtering expression includes replacing the uncompressed values within corresponding container structures of the tuple with the corresponding compressed fixed-length values mapped to these uncompressed value.
In various examples, converting the initial filtering expression into the updated filtering expression includes determining whether the literal value is included in a dictionary structure associated with the compression scheme as a corresponding uncompressed value. In various examples, when the literal value is included in an entry of the dictionary structure, the initial filtering expression is generated to indicate a corresponding compressed fixed-length value mapped to the literal value in the entry as the at least one compressed fixed-length value. In various examples, when the literal value is not included in any entry of the dictionary structure, the updated filtering expression is generated to indicate one of: an always true condition, or an always false condition.
In various examples, converting the initial filtering expression into the updated filtering expression includes accessing compression scheme data associated with the compression scheme to determine whether the at least one compressed value mapped to the at least one uncompressed value of the literal value exists in the compressed column. In various examples, when the compression scheme data indicates the at least one compressed value exists in the compressed column, the initial filtering expression is generated to indicate the at least one compressed value. In various examples, when the compression scheme data indicates the at least one compressed value does not exist in the compressed column, the updated filtering expression is generated to indicate one of: an always true condition, or an always false condition. In various examples, conversion in this fashion is based on an operator type of the initial filtering expression being one: an equality type a contains type, or an overlap type.
In various examples, converting the initial filtering expression into the updated filtering expression includes determining whether the literal value is included in a dictionary structure associated with the compression scheme. In various examples, when the literal value is included in an existing entry in the dictionary structure, the initial filtering expression is generated to indicate an existing compressed fixed-length value mapped to the literal value in the existing entry. In various examples, when the literal value is not included in any entry in the dictionary structure, the dictionary structure is updated to include a new entry that includes the literal value mapped to a new compressed fixed-length value, and the updated filtering expression is generated to indicate the new compressed fixed-length value mapped to the literal value in the new entry. In various examples, such conversion is based on an operator type of the initial filtering expression being an inequality operator type.
In various examples, converting the initial filtering expression into the updated filtering expression includes accessing compression scheme data associated with the compression scheme to determine a set of compressed fixed-length values corresponding to all uncompressed values meeting requirements of the initial filtering expression and having corresponding compressed values in the column, where the at least one compressed fixed-length value of the updated filtering expression includes the set of compressed fixed-length values.
In various examples, the requirements of the initial filtering expression are based on an operator type of the initial filtering expression being a range type operator applied to the literal value or a regex type operator applied to the literal value. In various examples, based on the operator type of the initial filtering expression being the range type or the regex type, the converting the initial filtering expression into the updated filtering expression includes generating the updated filtering expression to indicate a disjunction of a set of predicates corresponding to the set of compressed fixed-length values, where each of the set of predicates denotes equality with a corresponding one of the set of compressed fixed-length values.
In various examples, converting the initial filtering expression into the updated filtering expression includes determining a set of uncompressed values in a dictionary structure associated with the compression scheme that meet requirements of the initial filtering expression. In various examples, the at least one compressed fixed-length value of the updated filtering expression includes a set of compressed fixed-length values mapped to the set of uncompressed values in the dictionary structure.
In various examples, the requirements of the initial filtering expression are based on an operator type of the initial filtering expression being a range type applied to the literal value, or a regex type applied to the literal value. In various examples. based on the operator type of the initial filtering expression being the range type operator or the regex type operator, the updated filtering expression is generated to indicate a disjunction of a set of predicates corresponding to the set of uncompressed values. in various examples, each of the set of predicates denotes equality with a corresponding one of the set of compressed fixed-length values.
In various examples, the requirements of the one of the updated filtering expression are based on an operator type of the initial filtering expression being one of: an equality type operator applied to the literal value, a range type operator applied to the literal value, or a regex type operator applied to the literal value, and wherein, based on the operator type of the initial filtering expression being the one of the equality type operator, the mange type operator of the regex type operator, and based on the initial filtering expression indicating a first type of universal quantifier operator be applied to array structures of the compressed column requiring, all array elements meeting a condition applied to the literal value in accordance with the operator type, the converting the initial filtering expression into the updated filtering expression include generating the updated filtering expression to apply a second type of universal quantifier operator to array structures of the compressed column requiring, to have a corresponding row be included in the filtered subset of the set of rows, the array structures include only elements included in the set of compressed fixed-length values In various examples, the first type of universal quantifier operator is a for all operator applied to array structures of the compressed column requiring all array elements meeting a condition applied to the literal value in accordance with the operator type. In various example, the second type of universal quantifier operator is an ONLY CONTAINS operator applied to structures of the compressed column requiring the array structures include only elements included in the set of compressed fixed-length values.
In various examples, converting the initial filleting expression into the updated filtering expression includes performing a null value handling-based conversion to generate the updated filtering expression to include, in addition to the at least one compressed fixed-length value mapped to the at least one uncompressed value, an additional, non-null compressed fixed-length value mapped to a null value in the compression scheme. In various example, the additional compressed fixed-length value mapped to the null value is an integer value of zero, where all other compressed fixed-length values are non-zero integer values mapped to non-null variable-length values compressed via the compression scheme. In various examples, the null value handling-based conversion is performed despite the bull value not being expressly indicated in the initial filtering expression.
In various examples, the converting the initial filtering expression into the updated filtering expression includes, based on the initial filtering expression indicating an inequality operator be applied to the compressed column and the literal value. generating the updated filtering expression to include a conjunction of two predicates, wherein a first one of the two predicates indicates the inequality operator be applied to the compressed column and a compressed value mapped to the literal value, and
wherein a second one of the two predicates indicates the inequality operator be applied to the compressed column and the additional, non-null compressed fixed-length value mapped to the anti value
In various examples, based on the initial filtering expression indicating an existential quantifier operator be applied to array structures of the compressed column requiring at least one array element having inequality with the literal value, the converting the initial filtering expression into the updated filtering expression includes generating the updated filtering expression to apply a negated universal quantifier operator to array structures of the compressed column requiring, to have a corresponding row be included in the filtered subset of the set of rows, that the at least one array element include elements not included in an array that includes the compressed fixed-length value mapped to the literal value, and the additional, non-null compressed fixed-length value mapped to the bull value. In various examples, the negated universal quantifier operator is a DOES NOT ONLY CONTAIN operator.
In various examples, the literal value is a null array (e.g., is null itself, is empty, or contains the value of null as a single array element, such as its only array element and or as all of its multiple array elements).
In various examples, based on the initial filtering expression indicating an existential quantifier operator be applied to the cull array requiring at least one array element of the null array having inequality with the compressed column, the updated filtering expression is generated based on applying the existential quantifier operator to a non-bull array that includes a non-uall compressed fixed-length value mapped to the null value as its only array element, based on requiring at least one array element of the non-null array having inequality with a corresponding compressed value of the compressed column to have the corresponding row be included in the filtered subset of the set of rows. In various examples, the existential quantifier operator is a for_some operator.
In various examples, based on the initial filtering expression indicating the existential quantifier operator be applied to the pull array requiring the at least one array element of the bull array having equality with the compressed column, the updated filtering expression is generated based on applying the existential quantifier operator to the non-null array that includes the non-null compressed fixed-length value mapped to the null value as its only array element based on requiring the at least one array element of the non-null array having equality with the corresponding compressed value of the compressed column to have the corresponding row be included in the filtered subset of the set of rows. In various examples, the existential quantifier operator is a for some operator.
In various examples, based on the initial filtering expression indicating a universal quantifier operator be applied to the null array requiring all array elements of the null array having inequality with the compressed column, the updated filtering expression is generated based on applying the universal quantifier operator to the non-null array that includes the non-null compressed fixed-length value mapped to the null value as its only array element based on requiring the all array elements of the non-null array having inequality with the corresponding compressed value of the compressed column to have the corresponding row be included in the filtered subset of the set of rows. In various examples, the universal quantifier operator is a for all operator.
In various examples, based on the initial filtering expression indicating the universal quantifier operator be applied to the null array requiring the all array elements of the null array having equality with the compressed column, the updated filtering expression is generated based on applying the universal quantifier operator to the non-null army that includes the non-null compressed fixed-length value mapped to the null value as its only array element based on requiring the all array elements having equality with the corresponding compressed value of the compressed column to have the corresponding row be included in the filtered subset of the set of rows. In various examples, the universal quantifier operator is a for_all operator.
In various embodiments, any one of more of the various examples listed above are implemented in conjunction with performing some or all steps of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps of
In various embodiments a database system includes at least one processor and at least one memory that stores operational instructions. In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to perform some or all steps of
In various embodiments, the operational instructions, when executed by the at least one processor, cause the database system to: determine a query for execution against that includes a filtering predicate indicating filtering of a set of rows based on at least one uncompressed value of a compressed column of the set of rows; determine, based on the filtering predicate, an initial filtering expression indicating the at least one uncompressed value for the compressed column; convert the initial filtering expression into an updated filtering expression indicating filtering of the set of rows based on at least one compressed fixed-length value mapped to the at least one at least one uncompressed value in accordance with a dictionary compression scheme, wherein the updated filtering expression is semantically equivalent to the initial filtering expression: and/or identify a filtered subset of the set of rows in conjunction with executing the query based on applying the updated filtering expression.
In some embodiments, compressed column slabs generated via implementing WCC as discussed herein are seekable to support efficient lookup of single rows in the slab. For example, at a high level: first, a compression lookup structure is loaded and searched to identify the disk blocks that contain the frame holding the target row: next, those blocks are read off disk; finally, the frame is decompressed in a streaming fashion to find and/or materialize the target row. Embodiments of finding/decompressing rows in compressed column slab are discussed in further detail in conjunction with
Some or all features and/or functionality of the columns compressed via whole-column compression and/or corresponding compressed column slabs of
A plurality of uncompressed column slab data 2611 can be determined from a given dataset: 2505, where some or all of this uncompressed column slab data 2611 is compressed. In particular, for each segment 2424.i a plurality of uncompressed column slab data 2611.j 1-2611.i.C can be determined, for example, based on the column values for the records 2422 assigned to be included in the corresponding segment, for example, as discussed in conjunction with
The column slab compression module 2616 can be implemented to compress some or all of the uncompressed column slab data 2611 of some or all segments 2424 to generate a plurality of compressed column slabs, which can be written to the respective segment for the respective column accordingly. Thus, each segment can have some or all of its C column slabs 2610 implemented as compressed column slabs 2612 generated by column slab compression module 2616 from a corresponding uncompressed column slab data 2611. In some embodiments, one or more uncompressed column slab data 2611 for one or more columns and/or one or more segments remains uncompressed, where respective column slabs 2610 are generated from uncompressed column slab data 2611 without applying the WCC scheme via column slab compression module 2610.
Compressed column slabs 2612 can be generated on a per-segment, per column-slab basis. Different compressed column slabs 2612 can thus be generated separately and/or independently from other compressed column slabs 2612. Some or all compressed column slabs 2612 can optionally be generated via parallelized processing resources, for example, operating without coordination.
When a segment is generated and whole-column compression is enabled for at least one column in that segment, the segment writer (e.g. segment generator 2507) can train a compression dictionary on a subset of that column's data. For example. as illustrated in
Each compression dictionary 2622 can be trained in conjunction with implementing a corresponding compression scheme/compression library. This compression scheme/compression library can be a 3rd party compression scheme/compression library that is developed/established separately from database system 10, but is utilized by database system 10. This compression scheme/compression library can alternatively be a custom compression scheme/compression library configured for database system 10, for example, developed/established in conjunction with developing/establishing database system 10.
The compression scheme/compression library utilized to train each compression dictionary 2622 can be implemented via a non-adaptive dictionary compression algorithm, an adaptive dictionary compression algorithm a lossless compression algorithm, a fixed byte pair encoding, byte pair encoding, and/or other dictionary compression. The compression scheme/compression library utilized to train each compression dictionary 2622 can be based on implementing some or all features and/or functionality of the Lempel Ziy Algorithm and/or adaptations of the Lempel Ziv Algorithm. The compression scheme/compression library utilized to train each compression dictionary 2622 can be implemented as, based on and/or in a same or similar fashion as: zstd. zlib, LZ4, Smaz, LZ77, LZSS, LZRWI-A, LZJB, BARF, LZF, FastLZ, miniLZO, QuickLZ, LZS, Snappy, PalmDoc, LZSA, LZSA1, LZSA2, LZW, IZX, ROLZ, ACB, DTE, SCZ byte pair encoding, ISSDC digram coding. LZ78, GIF, LXMW, LZAP, LZWL statistical Lempel Ziv, and/or another scheme and/or library.
A given compression dictionary 2622.1k can be trained from a proper subset of the data in the corresponding uncompressed column slab data, for example, corresponding to column values for the column k for only a proper subset of the rows assigned to the segment i. In some embodiments, this proper subset of data utilized to train a given compression dictionary 2622.i.k is the first set of data (e.g., the first set of column values, serially) in the uncompressed column slab data 2611. In other embodiments, this proper subset of data utilized to train a given compression dictionary 2622.i.k is a random selection set of data (e.g. a randomly selected set of column values) in the uncompressed column slab data 2611.
This trained compression dictionary can be written to output column data. For example, the trained compression dictionary 2622.i is written to the corresponding segment 2424.i, for example, within the corresponding column slab 2612.i.
A header can also be written to output column data. The header can be generated and/or determined by the column slab as dictated/defined by the compression library utilized to generated the trained compression dictionary 2622. While not illustrated in
Once the given compression dictionary 2622, i.k is trained and/or is written to output column data, the segment generator 2507 can begin compressing the uncompressed column slab data 2611.i.k in a streaming fashion, generating one or more compression frames 2624. As illustrated in
In some embodiments, decompression can only begin at frame boundaries, and it is ideal to minimize extra block IO needed to read each row. In such embodiments, to minimize extra block IO needed to read rows and thus improve IO efficiency. the segment generator 2507 can be configured to attempt to adaptively determine the number of input rows needed to generate a frame spanning one to two fixed-length memory blocks of the segment 2424. When the desired size is reached, the frame can be closed and a new one is begun. Each frame can be required to contain an integer number of rows, where a column value is never split across frames Thus, the frames holding very large values may span many (e.g, more than 2 blocks), despite this attempt to minimize blocks per frame being employed. An example compressed columns slab illustrating the spanning of frames across memory blocks is discussed in conjunction with
The compression frame generator 2623 can generate a compression lookup structure 2613, which can be written to the segment 2424 and/or other memory resources. When each frame is closed/completed, the compression lookup structure 2613 can be updated with an entry associating the frame with its corresponding blocks.
This entry and/or other mapping data within the compression lookup structure 2613 can farther denote which rows are included within the given frame, for example, by simply denoting the number/identifier of a starting row of the frame and/or the range of rows in the frame if rows are written sequentially by identifier/number, and/or by otherwise denoting a list/set of rows included in the frame. In cases where rows are written consecutively, only the starting row is necessary, as the ending row/intermediate rows in the frame are frame is optionally not necessary, as this information is inherently denoted by the starting row of the next frame as indicated in the compression lookup structure 2613.
This entry and/or other mapping data within the compression lookup structure 2613 can further denote an offset/location within the respective starting block for a given frame denoting where within the block the given frame starts, and/or can further indicate an off′set/location within the respective ending block for a given frame denoting where within the block the given frame ends. In cases where frames are written consecutively, the ending point of the frame is optionally not necessary, as this information is inherently denoted by the starting off′set of the next frame as indicated in the compression lookup structure 2613.
In some embodiments, as illustrated in
In other embodiments a single lookup structure 2613 is generated to include lookup information for multiple different compressed column slabs 2612 for different columns of the same segment, multiple different compressed column slabs 2612 of different segments for the same column, and/or multiple different compressed column slabs 2612 of different segments and different columns. Such shared lookup strictures 2613 can be accessible in a corresponding segment and/or other memory resources if shared across multiple segments.
Once all data of the uncompressed column slab data 2611.i.k has been compressed and thus included in compression frames 2624 written to the compressed columns slab 2612, the final frame 2424.L can be closed.
Different compressed column slabs 2612 for different columns in the same segment, and/or for the same or different column in different segments, can have the same or different numbers of frames generated. Different compressed column slabs 2612 for different columns in the same segment, and/or for the same or different column in different segments, can have the same or different storage sizes. For example, the number of frames and/or final size of different compressed column slabs 2612 are different based on reflecting different column values rendering different compression ratios, based on having compressed data for different columns of different data types having different sized, based on applying different compression dictionaries 2622 trained separately, or other differences.
The given compressed column slab 2612 can begin with a header and the compression dictionary 2622 written within two blocks, or any number of blocks. The first frame can optionally start at the next new block as illustrated in
The frame header can be defined by the 3rd party compression library utilized to train the compression dictionary and/or the compressed payload (e.g., the plurality of frames 2624.1-2624.1.) itself. When decompressed, the payload can be identical to an uncompressed fixed or variable-length column slab on-disk format.
The trained compression dictionary 2622 can optionally span Jess blocks or more blocks. In some embodiments, the compression dictionary has a predetermined, configurable size and/or size range. In some embodiments, the trained compression dictionary 2622 has a size corresponding to a configured size and/or size range within the 32-128 KiB, or other required and/or suggested size bounds. The configured fixed-size and/or size range can be automatically selected via an optimization or other process, can be configured via user input, can be received, can be accessed in memory, and/or can otherwise be determined. Configuration of dictionary size is discussed in further detail in conjunction with
Compression dictionaries for different column slabs of the same or different segments can be of the same or different size. For example, different segments/different columns can optionally be user-configured and/or automatically configured to have compression dictionaries of different sizes. For example, columns having larger data types for its values and/or more cardinality across its values can optionally have larger compression dictionaries, for example, to optimize compression of the larger values. As another example, different segments have different sized dictionaries configured due to automatically detected differences in Storage constraints, processing capabilities, or other performance differences across different nodes/computing devices generating. storing and/or accessing these different segments 2424.
In this example, the first frame 2624. 1 spans more than one block and less than two blocks; the second frame 2624.2 spans more than two blocks, and the third frame 2624.3 spans less than one block. For example, these differences are based on the frames being variable-length based on applying the compression dictionary and/or the requirement that all column values be written within a single frame. Some or all of the spans of frames 2624 can be close to one to two blocks based on targeting one to two blocks as the frame size in minimizing IO during row reads as discussed previously. In other embodiments. greater/smaller numbers of blocks can be configured as the target/average frame size based on other IO optimizations or other storage/processing optimizations.
The compression lookup structure 2613 can be stored separately from the compressed column slab 2612, for example. within the segment 2424 and/or in another accessible location. WCC can employ a compression lookup sirneture 2613 that points le compression frames (which can be variable-length and may span blocks). Each block in the column span can appear as an entry in the structure, and each entry can identify the start row of the frame beginning in that block (if any), and the block relative byte offset where that frame begins.
As illustrated in
In some embodiments. WCC can be applied to any type of column, including fixed and variable-length scalars, arrays. and/or tuple columns having a plurality of tuple components. The given column 2707.k described herein can optionally be a tuple column and/or an array column. The dataset: 2505 can include one or more tuple columns, one or more array columns, and/or a combination of both, where some or all of its tuple columns and army columns are compressed via WCC for some or all segments.
In some embodiments. WCC can be implemented on a per-tuple component basis. In some embodiments, the column slab for a tuple column includes different frames generated separately for different tuple elements 2739. In some embodiments. different compression dictionaries are trained separately for some or all different tuple elements/sub-elements, where multiple tuple component slabs implement a corresponding compressed column slab for the tuple that is stored in a given segment. For example, each tuple component slab for the tuple column can optionally include its own header and/or its own trained compression dictionary 2622, as well as its own set of frames compressing only the given tuple element for each row Each such tuple component slab can have blocks/frame locations/rows mapped via their own lookup structure 2613, or shared lookup structure can be applied for some or all different elements of the tuple. This can be ideal in optimizing compression of like components appearing as a given element 2739 across the maple structures of different rows, which can be unrelated to other elements 2739 of a given tuple based on training compression dictionaries separately for different components, for example, motivated similarly to training different compression dictionaries for different columns.
In some embodiments. WCC can be implemented on a per-array component basis in a similar fashion for some or all array columns. Alternatively, in cases where array elements are the same data type and/or are not necessarily mapped to distinct types of data for different indexes of the array, arrays are compressed as a whole and/or their elements undergo compression via a same compression dictionary trained upon some or all elements of a subset of array structures corresponding to a subset of rows of the uncompressed column slab.
In some embodiments. WCC can be implemented for tuples as a whole for some or all tuple columns, where the compression dictionary is trained upon entire tables and compressed the tuples accordingly via a single dictionary.
In some embodiments. WCC can be exclusive of other fixed-length or variable-length compression. Furthermore, in some embodiments. WCC can be used in conjunction with global dictionary compression (GDC). For example, when enabled on a GDC column, first GDC is applied to compress a variable-length value into an integer value, and then WCC is applied on the column stream of integers to compress them on disk.
As illustrated in
In some embodiments, dictionary compression (GDC) module 2635 can determine which integer key maps to a given value of a given column undergoing GDC, and/or can optionally add a new entry if a new value is encountered to map this new value to a new integer key. The integer keys can be unique to ensure the values 2939 is recoverable as needed.
Thus, a given uncompressed column slab data 2611 for the given column can include the corresponding GDC pre-compressed column data 2632 (e.g., integer values) for the respective set of rows assigned to the segment. In some embodiments. the entire column underwent GDC via GDC module prior to grouping of rows into segments groups. In other embodiments, the column undergoes GDC via GDC module after being grouped into segment groups.
Such uncompressed column slab data 2611.i.k of a GDC column k, if compressed via WCC, can thus be processed to train a corresponding compression dictionary 2622.i.k, which is different from the dictionary structure 2636. In particular, this corresponding compression dictionary 2622a.k is trained from the integer values of the corresponding GDC pre-compressed column data 2632.i.k to render further compression of this set of integer values of the given column for the set of rows included in the given segment.
In some embodiments, all rows of the dataset 2505 have the given column GDC compressed via the dictionary structure 2636. However, the given column may be selectively further compressed via WCC for some segments, but not for others, based on WCC being applied on a per-segment basis, while GDC is optionally applied across all rows of a dataset regardless of what segments they ultimately are stored in. Furthermore, for each given segment that is further WCC compressed a different compression dictionary 2622 is generated and applied to further compress the column in the given segment, where multiple compression dictionaries 2622 are thus generated for this same column if multiple segments have this column undergo WCC. despite the same, single dictionary structure 2636 having been applied to compress this column via GDC across all segments.
In some embodiments, the dataset has multiple GDC compressed columns, such as variable-length columns or fixed-length columns compressed as fixed-length integer values via dictionary structure 2636, where any of these columns can similarly be further compressed for some or all segments 2424 via WCC, and/or where one or more of these columns are not further compressed for some or all segments 2424 via WCC
In some embodiments, the dataset has one or more variable-length columns or fixed-length columns not compressed via GDC, where the uncompressed column slab data 2611 for these columns are thus still the original variable-length column values and/or original fixed-length values, which are compressed directly via WCC rather than first being converted into column values.
The dictionary structure 2636 implemented by GDC module can be stored in any memory resources of database system 10. The dictionary structure 2636 can be applied across multiple columns, where different variable-length columns of the same or different dataset 2505 have their integer keys mapped to their original values via the same dictionary structure 2636. Alternatively. different dictionary structures 2636 are implemented for some or all different columns and/or for some or all different datasets 2505.
Once a WCC-compressed frame is identified and decompressed to recover the corresponding column values of a GDC compressed columns, the respective integer values are optionally further decompressed via the dictionary structure 2636 to determine the original variable-length value.
In some embodiments, WCC can be configured via user input, for example, as compression configuration data 2619. For example, this configuration is facilitated via user input, for example, by an administrator, end user, software engineer, or other user communicating with database system 10. As a particular example. Whole-column compression can have one or more configurable parameters that can be specified, for example, via the Data Definition Language (DDL) or another programming language/other instructions. Alternatively, some or all of the compression configuration data 2619 is automatically generated by database system 10.
In some embodiments a first parameter corresponding to compression level can be configured as compression level parameter data 2628, which can be configured as a numeric value that lets users adjust the compression ratio vs, heap memory and CPU usage, for example, to be consumed when training the respective compression dictionary. The compression level parameter data 2628 can be configured as other one or more values/instructions that configure how much compression is employed and/or how much processing/memory resources are utilized to generate the compression dictionary and/or the resulting compressed data slab.
Alternatively or in addition, a second parameter corresponding to dictionary size can be configured as dictionary size parameter data 2629, which can be configured as a value denoting the size (e.g., the fixed-size, and/or maximum/minimum size bounds) of the compression dictionary. In general, larger dictionaries provide better compression, but require more memory to train.
Alternatively or in addition, one or more other parameters of compression configuration data 2619 can be specified via user input and/or automatically. For example, the particular compression library/compression scheme to be applied can be configured to select which compression library/compression scheme is used by column slab compression module. As another example, the target frame size (e.g., one to two blocks) can be configured. Any other parameters specifying size/means by which columns slabs are compressed can be configurable parameters of compression configuration data 2619.
Some or all such parameters of the compression configuration data 2619 can be changed over time, for example, based on further user input updating one or more parameters of the compression configuration data 2619 and/or the database system 10 determining to automatically update one or more parameters, for example, as automatically identified to improve system performance.
Compression metadata 2631 can be maintained in each segment, enabling different segments to have different compression schemes for their respective column slabs. This metadata can be accessed to identify which columns are compressed in the segment, the scheme utilized to compress all columns and/or individual columns, and/or can specify some or all the compression configuration data 2619 that was applied to different individual columns and/or that was applied to the segment as a whole.
In some embodiments, some or all of the compression configuration data 2619 can be applied across a system level. where all compressed columns slabs across different columns and different segments are compressed via the same parameters as specified in compression configuration data 2619. In some embodiments, some or all of the compression configuration data 2619 can be applied across a per-segment, per-column or per-tuple-component basis. For example, different compression level parameter data 2628, different dictionary size parameter data 2629, and/or other different parameters of compression configuration data 2619 can be applied across different segments, different columns, and/or different tuple components. For example a first column is configured differently from a second column, and the first column is compressed in a first corresponding fashion across some or all segments, while the second column is compressed in a different, second corresponding fashion across some or all segments. As another example, a first segment is configured differently from a second segment, and the compressed columns of the first segment are all compressed in a first corresponding fashion, while the compressed columns of the second segment are all compressed in a second corresponding fashion.
As another example, a first tuple component of a given tuple column is configured differently from a second tuple component of the given tuple column, and the compressed column for the tuple column (across a given segment, or some or all segments), is generated based on compressing the first tuple component in a first corresponding fashion, and based on compressing the second tuple component in a different, second corresponding fashion. In some embodiments, the column slab for the tuple includes different frames generated separately for different tuple components, each in accordance with different compression parameters. In some embodiments, different compression dictionaries are trained separately for different tuple components, each in accordance with different compression parameters.
As illustrated in the example of
Different segments can be configured differently to have different ones of its columns compressed via WCC. This configuration is optionally specified by compression configuration data 2619 denoting different configuration for different columns, and/or other instructions that are user specified and/or automatically determined. Compression metadata 2631 can optionally be stored in and/or mapped to each segment to denote which columns of the corresponding segment are compressed vs. uncompressed.
In some embodiments, one or more segments have all of their columns compressed via WCC. In some embodiments. one or more segments have none of their columns compressed via WCC. In some embodiments, at least two segments have different non-null proper subsets of columns compressed via WCC and/or have different numbers of columns compressed via WCC.
In some embodiments, at least one column 2707 is consistent across all segments, where at least one columns is WCC compressed for all segments 2424, or is not WCC compressed for all segments 2424. In some embodiments, all columns 2707 are consistent across all segments, where every column is either WCC compressed in all segments segment 2424 or not WCC compressed in all segments segment 2424.
In some embodiments, at least one column 2707 is not consistent across all segments, where at least one column is WCC compressed in at least one segment 2424, and is also not WCC compressed for at least one other segment 2424. In some embodiments, no columns 2707 are consistent across all segments, where every column is WCC compressed in at least one segment 2424, and is also not WCC compressed for at least one other segment 2424.
As illustrated in the example of
Different segments can be configured differently to have different ones of its columns compressed via different parameters under WCC. This configuration is optionally specified by compression configuration data 2619 denoting different configuration for different columns, and/or other instructions that are user specified and/or automatically determined. Compression metadata 2631 can optionally be stored in and/or mapped to each segment to denote how different columns of the corresponding segment are compressed under WCC.
In some embodiments, some or all columns compressed are via WCC for a given segment, and all of the columns compressed under WCC are compressed via the same compression parameters. In some embodiments, a first segment segments has all of its WCC compressed column slabs compressed via first compression parameters applied across its column slabs, and a second segment segments has all of its WCC compressed column slabs compressed via second compression parameters applied across its column slabs, where the second compression parameters are different from the first compression parameters.
In some embodiments, some or all columns compressed are via WCC for a given segment, but some or all of different columns of the given segment are compressed under WCC via different compression parameters from each other. In some embodiments, at least two segments 2424 can have different sets of different compression parameters applied across its column slabs and/or can have different numbers of different compression parameters applied across its column slabs. In some embodiments, a first segment segments has its WCC compressed column slabs compressed via a corresponding set of compression parameters (which can be the same or different), and a second segment segments has all of its WCC compressed column slabs compressed via this same corresponding set of compression parameters (e.g. column I is compressed via compression parameters A for both segments, column 2 is compressed via compression parameters B for both rows, etc.).
In some embodiments, at least one column 2707 is compressed consistently across all segments, where at least one column is WCC compressed via the same compression parameters for all segments 2424. In some embodiments, at least one column 2707 is compressed consistently across all segments where it is compressed under WCC, where at least one column is WCC compressed via the same compression parameters for all segments 2424 in which it is WCC compressed, but is not compressed in some segments 2424. In some embodiments, all columns 2707 are compressed consistently across all segments when compressed under WCC, where all columns are each WCC compressed via the same compression parameters for all segments 2424 in which they are WCC compressed, which are optionally different from that of other columns.
In some embodiments a first columns is compressed consistently across all segments via first parameters, and a second column also compressed consistently across all segments via these first parameters. In some embodiments, a first columns is compressed consistently across all segments via first parameters, and a second column is compressed consistently across all segments via second parameters different from the first parameters.
In some embodiments, at least one column 2707 is not compressed consistently across all segments in which it is compressed under WCC, where at least one column is WCC compressed in at least one segment 2424 via first compression parameters, and this at least one column is WCC compressed in at least one other segment 2424 via second compression parameters. In some embodiments. no column 2707 is compressed consistently across all segments in which it is compressed under WCC, where any given column is WCC compressed in at least one segment 2424 via corresponding compression parameters, and the given column is WCC compressed in at least one other segment 2424 via ether compression parameters.
In some embodiments, at least two columns can have different sets of different compression parameters applied across all segments and/or can have different numbers of different compression parameters applied across all segments.
During query execution for a query requiring access to a given column k that is WCC compressed as a compressed column slab in one or more segments 2424, the IO level 2415 can implement segment readers 2650. A segment reader 2560 can be operable to read whole-column compressed data of at least one column slab of at least one segment. In particular a given segment reader 2560 can be operable to perform a compressed column slab read process 2650.i. k to read column k front segment i, rendering generation of row data from an incoming row list. This row data can be further filtered/processed at the IO level and/or can be emitted to operators 2420 for processing, for example, in conjunction with other data for other columns.
In some embodiments, some or all other segment readers for other segments do not perform the compressed column slab read process 2660 for column k based on column k not being compressed and being able to be read directly. I.
In some embodiments, the segment reader 2650.i, and/or some of all other segment readers, performs additional compressed column slab read processes 2660 for additional columns based on these additional columns being compressed via WCC and also requiring access in conjunction with execution of the given query.
Some or all of the plurality of segment readers 2650 can optionally be implemented independently and/or in parallel. Some or all of the plurality of segment readers 2650 can optionally be implemented as respective IO pipelines 2835 for the respective segments, for example, as discussed in conjunction with
An incoming tow list 2657 can specify which rows require being read, for example, for ultimate decompression of the respective values in conjunction with execution of the query. This incoming row list 2657 optionally specifies a filtered, proper subset of all rows of the segment based on prior filtering having been applied (e.g, based on applying other query predicates, based on accessing probabilistic index data for the given column, based on accessing the index data and/or values for other columns to filter the row list based on predicates for other columns, based on this row being specified in the query and/or in user input directly. etc.) Alternatively, the incoming row list 2657 optionally specifies all rows of the segment.
A lookup structure loader 2671 can be implemented to load some or all of the compression lookup structure 2613.i.k to local memory or other memory accessible by the read process 2660 for access to identify frame locations of each row in the row list 2657. Alternatively, the compression lookup structure 2613.i.k is already loaded based on having been cached, for example, in conjunction with executing another query. Alternatively, the compression lookup structure 2613.i.k is not loaded, but instead accessed directly in segment 2424 to return frame location data for each row in the row list as needed.
A dictionary loader 2672 can be implemented to load some of all of the compression structure 2613.i.k to local memory or other memory accessible by the read process 2660 for access to generate row data for each row in the row list 2657. Alternatively. the compression structure 2613.i.k is already loaded based on having been cached, for example, in conjunction with executing another query. Alternatively, the compression structure 2613.i.k is not loaded, but instead accessed directly in segment 2424 to return compression data for rows in the row list as needed.
A row list processing module 2673 can be implemented to process the row IDs included in the row list 2657 in conjunction with accessing the lookup structure 2613.i.k and/or dictionary 2622.i.k, for example, in local memory based on having been loaded and/or via corresponding accesses to the segment 2424 in database storage 2450. A frame identifier 2674 can be implemented to access the lookup structure 2613.i.k (e.g., in local storage based on having been loaded) to identify, for each row in the row list, the frame location of a corresponding compression frame.
For example, for a given row ID j, this includes searching the lookup structure to identify a starting block denoting the start of a frame p that has the largest row number that is still less than the given row j ID (e.g., via a binary search or other search). determining the row is thus included this frame p, identifying the offset of this frame in the corresponding starting block as specified in the respective entry of the lookup structure 2613, identifying the block and corresponding offset for the start of the next frame based on entries for one or more subsequent blocks in the lookup structure 2613.
A frame loader 2675 can utilize the frame location data for each row to load the identified frame p for each row. For a given frame p location, the frame 2624.p is loaded, for example, by reading from the offset in the identified starting block to the offset in the identified ending block where the next frame begins. In cases where multiple rows are included in the same frame. this same frame is optionally loaded only once.
A row data generator 2676 can process the frame 2624 in conjunction with processing dictionary 2622.i.k to generate row data 2659 for row j. In some embodiments, the row data 2659 is the original, decompressed column value of row j for column k. In other embodiments, the row data 2659 is a view, such as instructions or other data, that can render fast decompression of the frame to render recovery of the original, decompressed column value of row j for column k at a later time, as needed. For example. the row data 2659 includes and/or indicates: a relevant portion of the dictionary 2622 and/or memory location data to access the loaded dictionary 2622 when decompression is performed: the frame 2624.p and/or memory location data to access the loaded frame 2624.p when decompression is performed; information denoting, which row in the loaded frame is row j (e.g. a number of rows from the starting row to row j is the ID for row j minus the ID for the starting row of the frame as specified in the lookup table): and/or other information.
Decompressing the frame 2624.p to recover the column value of row j for column k (e.g. at a later time, or directly by row generator 2676) can include accessing the compression dictionary 2622.i.k (e.g., in local memory based on having been loaded) to decompress the loaded frame 2624, p in accordance with the respective compression library/compression scheme applied to trait the compression dictionary 2622.i.k. The loaded frame 2624.p can be decompressed starting from the beginning of the frame. In some embodiments, rather than decompressing the whole frame, only a first portion of the frame is decompressed up until row i (e.g, based on decompressing the determined number of values of row j from the start row).
In embodiments where multiple rows included in the same frame 2624 require decompression, the frame is optionally decompressed only once to render recovery of the multiple respective column values. In such cases, rather than decompressing the whole frame, only a first portion of the frame is decompressed up until the last row/row with the highest ID included in the column to ensure all necessary rows are decompressed, without requiring full decompression of the frame.
In embodiments where the column k was also GDC compressed, the dictionary structure 2636 can be accessed as necessary to further decompress the integers as the original column values.
In this example, frame identifier 2674 identifies location data for frame I and frame 2 based on accessing lookup structure 2613 and determining frame I includes rows 111 and 150, and that frame 2 includes row 265. The frame loader loads these frames 1 and 2 starting from the specified block at the specified offset. Frame O is not loaded based on the row list not including any rows from row 0 to row 100.
The row data generator 2676 can generate row data by decompressing, or generating a view to enable decompression of the identified rows of the row list. When ultimately decompressing the column values for rows 111 and 150. frame 2624.1 is optionally decompressed once to read both of these rows, up until row 150 is decompressed, as no rows after row 150 are required. This can include reading only the first 51 values of the frame based on the frame starting at row 100, and based on the compression being applied serially in accordance with applying the respective compression scheme, where the 5 1st value and the 12th value are returned as the column values for row 111 and 150. Frame 2 and/or other frames can be decompressed similarly based on which rows within the frame require having values materialized.
In some embodiments, the segment reader 2560 can optionally be implemented for a given segment in conjunction with executing a corresponding IO pipeline 2835 for the given segment 2424. In the case where the segment contains whole-column compressed data as one or more of its column slabs requiring access in conjunction with a corresponding query, IO pipeline 2835 can include a compressed pipeline element 3017 for column k that, when executed, renders execution of compressed column slab read process 2660. In such embodiments, the functionality of segment reader 2560 illustrated in conjunction with
In some embodiments, this compressed pipeline element 3017 is implemented as a type of source element 3014 that generates row data for specified rows of a given columns. However, the output of the compressed pipeline element optionally does net emit materialized column values like source elements applied to uncompressed columns, and can instead emit views for the requested rows that can be later processed to find, decompress, and/or materialize the column values for the requested rows from the loaded frames, for example, lazily and/or on-demand.
In particular, this element 3017 can be operable to generate a set of row data 2659.1-J for an incoming set of rows t-J indicated in incoming row list 2657 based on: reading the compression dictionary off disk, loading the corresponding compression lookup structure partition (which may be cached) and searching it for the frame and corresponding disk blocks holding the needed row data: issuing IO for the blocks containing the matching frames; and/or returning a view that can find, decompress, and materialize rows from the loaded frames lazily and/or on-demand as corresponding row data 2659.
In some embodiments, for each row materialized: a portion of corresponding compression frame can be decompressed. starting from the beginning of the frame. Decompressed column data is streamed into the provided output buffer, avoiding unnecessary copies.
In some embodiments, the incoming row list processed by compressed pipeline element 3017 of
In some embodiments, only some of the rows of the incoming row list having row data generated is ultimately materialized, for example, based on filtering being applied to the set of rows 1-J to filter some or all of these rows out in conjunction with applying the query predicates. Alternatively, all of the rows of the incoming row list having row data generated are ultimately materialized.
In some embodiments, additional compressed pipeline elements 3017 are applied for other WCC compressed columns for example, as specified in the query for being projected and/or being filtered based on their values Such other compressed pipeline elements 3017 are optionally applied serially before, serially after, and/or in parallel with the given compressed pipeline elements 3017 of
In some embodiments, other segments are processed via different IO pipelines that optionally do not include the compressed pipeline element 3017 for column k, for example, based on the column k not being WCC compressed in these other segments. In some embodiments, other segments are processed via different IO pipelines that optionally include the compressed pipeline element 3017 for column k, but are configured in a different fashion from the IO pipeline for segment i based on other differences between the segments.
In some embodiments, the rows are materialized within the IO pipeline to render further filtering of the rows, for example, via filtering elements 3016 that compare the decompressed values to a value specified by the query predicates or otherwise evaluate the decompressed values against the query predicate. Alternatively, the rows are matenalized later via other operators 2520 that process the respective view.
Ultimately, the materialized, decompressed values can be further processed/manipulated/aggregated via operators 2520 and/or can be emitted as projected values in the resultant, as specified by the query.
Some or all of the method of
Step 2582 includes determining a dataset for storage. For example, the dataset includes, for each of a plurality of rows of the dataset a plurality of column values corresponding to a plurality of columns of the dataset. Step 2584 includes generating a plurality of column slab data from the dataset. For example, each of the plurality of column slab data includes column values for one of the plurality of columns from a proper subset of rows of the plurality of rows assigned to one of a plurality of segments. Step 2586 includes training each of a plurality of compression dictionaries from a corresponding one of the plurality of column slab data. Step 2588 includes generating each segment of the plurality of segments based on writing a set of compressed column slabs to the each segment, for example, based on compressing each of a corresponding set of the plurality of column stab data as a set of variable-length compression frames written to the each segment based on applying a corresponding one of the plurality of compression dictionaries.
In various examples, the proper subset assigned to the one of the plurality of segments is one of a plurality of proper subsets of rows assigned to the plurality of segments. In various examples, each of the plurality of proper subsets of rows is assigned for storage in exactly one of the plurality of segments, and/or each of the plurality of plurality of proper subsets are mutually exclusive and/or collectively exhaustive with respect to the plurality of rows. In various examples, each of the plurality of rows is assigned to have its column values stored in exactly one segment. In various example, a given row's column values for all columns of the plurality of columns are stored in a same segment of the plurality of segments, for example, across multiple corresponding column slab data.
In various examples, the plurality of column slab data are generated from the dataset based on performing a cluster key-based grouping process to group rows into different record groups, where each record group is processed to generate a corresponding segment group.
In various examples, the dataset corresponds to a portion of a full dataset (e.g. a most recently received set of rows and/or a set of rows identified to be converted into segments at a given time) and/or corresponds to a full dataset. In various examples. the full dataset can correspond to one or more database tables, such as one or more relational database tables, for example, where its rows have the column values for some or all of the set of columns. In various examples, the full dataset corresponds to non-relational tow data and/or other records having values for a set of fields (e.g. columns).
In various examples, the set of compressed column slabs includes only one compressed column slab. In various example. the set of compressed column slabs includes multiple compressed column slabs. In various examples, the set of compressed column slabs corresponds to a set of columns that includes all of the plurality of columns, or only a proper subset of the plurality of columns. In various examples, some or all different ones of the plurality of segments have respective sets of compressed column slabs that correspond to the same set of columns, or different sets of columns.
In various examples, each of the plurality of compression dictionaries are trained from a proper subset of column values in the corresponding one of the plurality of column slab data. In various examples, the proper subset of column values includes one of: a first set of column values from a full set of column values in the corresponding one of the plurality of column slab data and/or a randomly selected set of column values from the full set of column values in the corresponding one of the plurality of column slab data. In various examples, the proper subset of column values in the corresponding one of the plurality of column slab data corresponds to column values of only a proper subset of the proper subset of rows assigned to the corresponding segment.
In various examples, each of the set of compressed column slabs is generated to include: a header: the corresponding one of the plurality of compression dictionaries: and/or compressed data generated based on a compressing one of the plurality of column slab data based on applying the corresponding one of the plurality of compression dictionaries. In various examples, the header is defined by and/or otherwise based on a third-party compression library and/or third-party compression scheme. In various examples, the header is defined by and/or otherwise based on a custom compression library and/or custom compression scheme.
In various examples, each of the set of variable-length compression frames includes a corresponding subset of a plurality of subsets of the proper subset of rows assigned to the each segment. In various examples, the plurality of subsets are mutually exclusive and collectively exhaustive with respect to the proper subset.
In various examples, generating the each of the plurality of segments is further based on writing a set of compression lookup structures corresponding to the set of compressed column slabs.
In various examples, the set of variable-length compression frames are written across a set of fixed-length blocks of the segment.
In various examples, each compression lookup structure of the set of compression lookup structures indicates, for each of the set of fixed-length blocks of the corresponding compressed column slab in which a new frame of the set of variable-length compression frames starts: a frame identifier identifying the new frame; a row identifier for identifying a starting from of the new frame: and/or an offset identifying a staring location of the new frame within the each of the set of fixed-length blocks.
In various examples, at least one frame of the set of variable-length compression frames of the each compressed column slab spans more than two blocks of the set of fixed-length blocks. In various examples, a corresponding compression lookup Structure of the set of compression lookup structures indicates a corresponding at least one of the set of fixed-length blocks of the corresponding compressed column slab is entirely consumed by compressed data of a frame of the at least one frame that started in a prior one of the set of fixed-length: blocks based on spanning more than two blocks.
In various examples, the method further includes determining compression level parameter data, for example, based on the compression level parameter data being configured via user input. In various examples, the method further includes determining dictionary size parameter data, for example, based on the dictionary size parameter data being configured via the same or different user input. In various examples, the plurality of compression dictionaries are trained based on applying the compression level parameter data and the dictionary size parameter data.
In various examples, generating the each of the plurality of segments is further based on writing compression metadata on the each segment indicating segment compression data for the each segment. In various examples, a first corresponding set of the plurality of column slab data of a first segment of the plurality of segments are compressed in accordance with first segment compression data. In various examples, a second corresponding set of the plurality of column slab data of a second segment of the plurality of segments are compressed in accordance with second segment compression data that is different from the first segment compression data.
In various examples, the first segment compression data is different from the second segment compression data based on the first segment compression data denoting compression of a first subset of columns of the plurality of columns, the second segment compression data denoting compression of a second subset of columns of the plurality of columns, wherein the first subset has a non-null set difference with the second subset.
In various examples, the first segment compression data is different from the second segment compression data based on the first segment compression data denoting compression of a first number of columns of the plurality of columns, the second segment compression data denoting compression of a second number of columns of the plurality of columns, wherein the first number is different from the second number.
In various examples, the first segment compression data is different from the second segment compression data based on the first segment compression data denoting compression of one of the plurality of columns in accordance with first compression parameters, and the second segment compression data denoting compression of the one of the plurality of columns in accordance with second compression parameters different from the first compression parameters.
In various examples, the compressing of each of the corresponding set of the plurality of column slab data is in accordance with a first compression type. In various examples, determining the plurality of column slab data includes generating a set of pre-compressed column data as a subset of the plurality of column slab data by applying a second compression type to column values of at least one of the plurality of columns for rows assigned to at least one segment of the plurality of segments. In various examples, a corresponding subset of the plurality of compression dictionaries are each trained from a corresponding one of the set of pre-compressed column data. In various examples, the at least one of the of a plurality of segments are generated based on writing the set of compressed column slabs to the at least one segment based on further compressing each corresponding one of the set of pre-compressed column data in accordance with the first compression type as the set of variable-length compression frames written to the each segment based on applying the corresponding one of the plurality of compression dictionaries.
In various examples, the second compression type is a global dictionary compression type. In various examples, same global compression dictionary is utilized to generate the set of pre-compressed column data for the at least one of the plurality of columns for all of the plurality of segments. In various examples, the same global compression dictionary is utilized to generate the pre-compressed column data for multiple ones of the plurality of columns.
In various examples, a first set of column slab data is generated for a first column of the plurality of columns storing a first data type. In various examples a second set of column slab data is generated for a first column of the plurality of columns storing a second data type. In various examples, the first data type and the second data type are different data types of a set of data types that includes: at least one fixed-length data type; at least one variable-length data type; at least one array data type: and/or at least one tuple data type. For example, the first data type and the second data type are: different fixed-length data types: different variable-length data types: different array data types: and/or different tuple data types. As another example, the first data type is a fixed-length data type and the second data type is a variable-length data type: the first data type is an array data type and the second data type is not an array data type; and/or the first data type is a tuple data type and the second data type is not a tuple data type.
In various examples, the method further includes determining a query having query predicates indicating a first column compressed as compressed column slabs in a set of segments of the plurality of segments. In various examples, the method further includes, for each segment of the set of segments, determining row data for rows satisfying the query predicates. Determining the row data for the rows satisfying the query predicates can be based on; reading the compression dictionary from the each segment: determining a set of rows of the each segment for access; identifying ones of the set of variable-length compression frames of the compressed column slab written for the first column that include ones of the set of rows: and/or generating the row data based on reading only the ones of the set of variable-length compression frames identified to include the ones of the set of rows. In various examples, the method further includes generating a query resultant for the query based on processing the row data for all segments of the set of segments.
In various examples, the method further includes reproducing column values of the first column, for each segment, based on utilizing the compression dictionary to decompress at least one of the set of variable-length compression frames indicated it the row data generated for the each segment.
In various examples, reproducing the column values of the first column is based on decompressing only a portion of one variable-length compression frame of the set of variable-length compression frames, starting from a start of the variable-length compression frame and ending before an end of the variable-length compression frame, based on all ones of the set of rows compressed in the one variable-length compression frame being serially included within the portion of the one variable-length compression frame.
In various examples, identifying the ones of the set of variable-length compression frames of the compressed column slab that include ones of the set of rows is based on accessing a compression lookup structure for the compressed column slab mapping row identifiers of the set of rows to corresponding ones of the set of variable-length compression frames, and further mapping memory location data to corresponding ones of the set of variable-length compression frames.
In various examples, the method further includes executing a query based on processing compressed column slabs stored it at least some of the plurality of segments based on performing some or all steps of
In various embodiments, any one of more of the various examples listed above are implemented in conjunction with performing some or all steps of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps of
In various embodiments a storage system, such as a database systems, includes at least one processor and at least one memory that stores operational instructions. In various embodiments, the operational instructions, when executed by the at least Que processor, cause the storage system to perform some or all steps of
In various embodiments, the operational instructions, when executed by the at least one processor, cause the storage system to determine a dataset for storage that includes, for each of a plurality of rows of the dataset a plurality of column values corresponding to a plurality of columns of the dataset, generate a plurality of column slab data from the dataset, where each of the plurality of column slab data includes column values for one of the plurality of columns from a proper subset of rows of the plurality of rows assigned to one of a plurality of segments: train each of a plurality of compression dictionaries from a corresponding one of the plurality of column slab data: and/or generate each segment of the plurality of segments based on writing a set of compressed column slabs to the each segment based on compressing each of a corresponding set of the plurality of column slab data as a set of variable-length compression frames written to the each segment based on applying a corresponding one of the plurality of compression dictionaries.
Some or all of the method of
Step 2581 includes determining a query having query predicates indicating a first column compressed as compressed column slabs in a set of segments. Step 2583 includes, for each segment of the set of segments, determining row data for rows satisfying the query predicates based on processing a corresponding compressed column slab of the each segment corresponding to the first column Step 2585 includes generating a query resultant for the query based on processing the row data for all segments of the set of segments.
Performing step 2583 can include performing some of all of steps 2587. 2589, and/or 2591. Step 2587 includes determining a set of rows of the each segment for access. Step 2589 includes identifying ones of a set of variable-length compression frames of the compressed column slab written for the first column that include ones of the set of rows based on accessing a lookup structure corresponding to the corresponding compressed column slab. Step 2589 includes generating the row data based on reading only the ones of the set of variable-length compression frames identified to include the ones of the set of rows, where the row data is generated based on a compression dictionary corresponding to the compressed column slab.
In various examples, the method further includes loading the lookup structure from storage resources in conjunction with accessing the each segment. In various examples, the lookup structure is read from the each segment based on being stored within the each segment. In various examples, the lookup structure is read from the compressed column slab based on being stored within the compressed column slab.
In various examples, the method further includes loading the corresponding compression dictionary from storage resources in conjunction with accessing the each segment. In various examples, the corresponding compression dictionary is read from the each segment based on being stored within the each segment. In various examples, the corresponding compression dictionary is read from the compressed column slab based on being stored within the compressed column slab.
In various examples, the query predicates further indicate at least one additional column of the plurality of columns. In various examples, the at least one additional column is compressed, where the method further includes, for each segment of the set of segments, processing at least one additional corresponding compressed column slab of the each segment corresponding to the at least one additional column. In various examples, the at least one additional is uncompressed, where the method further includes, for each segment of the set of segments, processing at least one corresponding uncompressed column slab of the each segment corresponding to the at least one additional column.
In various examples, determining the set of rows of the each segment for access is based on applying at least one prior IO pipeline element of an IO pipeline generated for the each segment. In various examples, the set of rows is a row list emitted based on having applied at least one: filtering operator, source operator, index element, intersection element, union element, or other IO pipeline element for the first column or for other columns In various examples, the same IO pipeline is applied across all segments. In various examples, different IO pipelines are generated for different segments. In various examples, the IO pipelines are different for different segments based on at least one segment having different ones of the sets of columns compressed.
In various examples, the first column is uncompressed as uncompressed column slabs in a second set of segments. In various examples, the method further includes, for each additional segment of the second set of segments, determining additional row data for rows satisfying the query predicates based on processing a corresponding uncompressed column slab of the each additional segment corresponding to the first column. In various examples, the a query resultant for the query is generated further based on processing the additional row data for all additional segments of the second set of segments.
In various examples, the row data is generated based on decompressing the column values for the set of rows based on applying the compression dictionary to the ones of the set of variable-length compression frames. In various examples, the row data indicates the decompressed column values based on the ones of the set of variable-length compression frames being decompressed.
In various examples, the row data is generated as view that can enable finding, decompressing, and/or materializing of rows from the loaded frames at a later time (e.g., if the corresponding column values are determined to be necessary for generation of the query resultant), for example, on-demand. In various examples, the column values for all of the set of rows is not decompressed, for example, based on column values of the first column not requiring materialization (e.g., the row identifiers are used to filter rows based on predicates applied to the first column, where other column values of other columns are projected in the resultant and/or are processed to generate the resultant), and/or based on at least some rows of the first column being filtered out via other filtering (e.g, based on other predicates), where only the column values of the first column of the remaining rows are materialized based on the view and/or other relevant information indicated in the tow data.
In various examples, the method further includes generating the compressed column slab for each segment of the set of segments based on performing some or all of the method of
In various embodiments, any one of more of the various examples listed above are implemented in conjunction with performing some or all steps of
In various embodiments, at least one memory device, memory section, and/or memory resource (e.g., a non-transitory computer readable storage medium) can store operational instructions that, when executed by one or more processing modules of one or more computing devices of a database system, cause the one or more computing devices to perform any or all of the method steps of
In various embodiments, a storage system, such as a database system, includes at least one processor and at least one memory that stores operational instructions. In various embodiments, the operational instructions, when executed by the at least one processor, cause the storage system to perform some or all steps of
In various embodiments, the operational instructions, when executed by the at least one processor cause the storage system to: determine a query having query predicates indicating a first column compressed as compressed column slabs in a set of segments: determine, for each segment of the set of segments, row data for rows satisfying the query predicates based on processing a corresponding compressed column slab of the each segment corresponding to the first column; and/or generating a query resultant for the query based on processing the row data for all segments of the set of segments. In various embodiments, processing the corresponding compressed column slab of the each segment is based on: determining a set of rows of the each segment for access: identifying ones of a set of variable-length compression frames of the compressed column slab written for the first column that include ones of the set of rows based on accessing a lookup structure corresponding to the corresponding compressed column slab; and/or generating the row data based on reading only the ones of the set of variable-length compression frames identified to include the ones of the set of rows, where the row data is generated based on a compression dictionary corresponding to the compressed column slab.
It is noted that terminologies as may be used herein such as bit stream, signal, signal sequence, etc. (or their equivalents) have been used interchangeably to describe digital information whose content corresponds to any of a number of desired types (e.g., data. video, speech, text. graphics, audio, etc, any of which may generally be referred to as data).
As may be used herein, the terms “substantially” and “approximately” provides an industry-accepted tolerance for its corresponding term and/or relativity between items. For some industries, an industry-accepted tolerance is less than one percent and, for other industries , the industry-accepted tolerance is IO percent or more. Other examples of industry-accepted tolerance range from less than one percent to fifty percent. Industry-accepted tolerances correspond to, but are not limited to, component values, integrated circuit process variations, temperature variations, rise and fall times, thermal noise, dimensions, signaling errors. dropped packets, temperatures, pressures, material compositions, and/or performance metrics. Within an industry, tolerance variances of accepted tolerances may be more or less than a percentage level (e.g., dimension tolerance of less than +/−1%). Some relativity between items may range from a difference of less than a percentage level to a few percent. Other relativity between items may range from a difference of a few percent to magnitude of differences.
As may also be used herein the term(s) “configured to”. “operably coupled to”. “coupled to”, and/or “coupling” includes direct coupling between items and/or indirect coupling between items via an intervening item (e.g., an item includes, but is not limited to, a component, an element, a circuit, and/or a module) where, for an example of indirect coupling, the intervening item does not modify the information of a signal but may adjust its current level, voltage level, and/or power level. As may further be used herein, inferred coupling (i.e., where one element is coupled to another element by inference) includes direct and indirect coupbag between two items in the same manner as “coupled to”
As may even further be used herein, the term “configured to”. “operable to”. “coupled to”, or “operably coupled to” indicates that an item includes one or more of power connections, input(s), outputs), etc., to perform, when activated, one or more its corresponding functions and may further include inferred coupling to one or more other items. As may still further be used herein, the term “associated with”, includes direct and/or indirect coupling of separate items and/or one item being embedded within another item.
As may be used herein, the term “compares favorably”, indicates that a comparison between two or more items, signals. etc., indicates an advantageous relationship that would be evident to one skilled in the art in light of the present disclosure, and based, for example, on the nature of the signals/items that are being compared. As may be used herein, the term “compares unfavorably”, indicates that a comparison between two or more items, signals, etc. fails to provide such an advantageous relationship and/or that provides a disadvantageous relationship. Such an item/signal can correspond to one or more numeric values, one or more measurements, one or more counts and/or proportions, one or more types of data, and/or other information with attributes that can be compared to a threshold, to each other and/or to attribute of other information to determine whether a favorable or unfavorable comparison exists. Examples of such an advantageous relationship can include: one item/signal being greater than (or greater than or equal to) a threshold value, one item/signal being less than (or less than or equal to) a threshold value, one item/signal being greater than (or greater than or equal to) another item/signal, one item/signal being less than (or less than or equal to) another item/signal, one item/signal matching another item/signal, one item/signal substantially matching another item/signal within a predefined or industry accepted tolerance such as 1%. 596, 10% or some other margin, etc. Furthermore, one skilled in the art will recognize that such a comparison between two items/signals can be performed in different ways. For example. when the advantageous relationship is that signal 1 has a greater magnitude than signal 2 a favorable comparison may be achieved when the magnitude of signal 1 is greater than that of signal 2 or when the magnitude of signal 2 is less than that of signal 1. Similarly, one skilled in the art will recognize that the comparison of the inverse or opposite of items/signals and/or other forms of mathematical or logical equivalence can likewise be used in an equivalent fashion. For example, the comparison to determine if a signal X>5 is equivalent to determining if −X<−5, and the comparison to determine if signal A marches signal B can likewise be performed by determining −A matches −B or no:(A) matches not(B). As may be discussed herein, the determination that a particular relationship is present (either favorable or unfavorable) can be utilized to automatically trigger a particular action. Unless expressly stated to the contrary, the absence of that particular condition may be assumed to imply that the particular action will not automatically be triggered. In other examples, the determination that a particular relationship is present (either favorable or unfavorable) can be utilized as a basis or consideration to determine whether to perform one or more actions. Note that such a basis or consideration can be considered alone or in combination with one or more other bases or considerations to determine whether to perform the one or more actions. In one example where multiple bases or considerations are used to determine whether to perform one or more actions, the respective bases or considerations are given equal weight in such determination. In another example where multiple bases or considerations are used to determine whether to perform one or more actions, the respective bases or considerations are given unequal weight in such determination.
As may be used herein, one or more claims may include, in a specific form of this generic form, the phrase “at least one of a, b, and c” or of this generic form “at least one of a, b, or c”, with more or less elements that “a″b” and “c”. In either phrasing, the phrases are to be interpreted identically. In particular, “at least one of a, b, and c” is equivalent to “at least one of a. b, or c” and shall mean a, b, and/or c. As an example, it means: “a” only.“o” only. “c” only, “a” and “b”. “a” and “c”, “b” and “c”, and/or “a”. “b”, and “c”.
As may also be used herein, the terms “processing module”, “processing circuit”. “processor”, “processing circuitry”. and/or “processing unit” may be a single processing device or a plurality of processing devices. Such a processing device may be a microprocessor, micro-contoller, digital signal processor, microcomputer, central processing unit, field programmable gate array, programmable logic device, state machine, logic circuity, analog circuity, digital circuitry, and/or any device that manipulates signals (analog and/or digital) based on hard coding of the circuitry and/or operational instructions. The processing module, module, processing circuit, processing circuitry, and/or processing unit may be, or further include, memory and/or an integrated memory element, which may be a single memory device a plurality of memory devices, and/or embedded circuitry of another processing module, module, processing circuit, processing circuitry, and/or processing unit. Such a memory device may be a read-only memory, random access memory, volatile memory, non-volatile memory, static memory, dynamic memory. flash memory, cache memory, and/or any device that stores digital information Note that if the processing module, module, processing circuit, processing circuitry, and/or processing unit includes more than one processing device, the processing devices may be centrally located (e.g., directly coupled together via a wired and/or wireless bus structure) or may be distributedly located (e.g., cloud computing via indirect coupling via a local area network and/or a wide area network). Further note that if the processing module, module, processing circuit, processing circuitry and/or processing unit implements one or more of its functions via a state machine, analog circuitry, digital circuitry, and/or logic circuitry, the memory and/or memory element storing the corresponding operational instructions may be embedded within or external to, the circuitry comprising the state machine, analog circuitry, digital circuitry, and/or logic circuitry. Still further note that the memory element may store, and the processing module, module. processing circuit, processing circuitry and/or processing unit executes, hard coded and/or operational instructions corresponding to at least some of the steps and/or functions illustrated in one or more of the Figures. Such a memory device or memory element can be included in an article of manufacture.
One or more embodiments have been described above with the aid of method steps illustrating the performance of specified functions and relationships thereof. The boundaries and sequence of these functional building blocks and method steps have been arbitrarily defined herein for convenience of description Alternate boundaries and sequences can be defined so long as the specified functions and relationships are appropriately performed. Any such alternate boundaries or sequences are thus within the scope and spirit of the claims. Further, the boundaries of these functional building blocks have been arbitrarily defined for convenience of description. Alternate boundaries could be defined as long as the certain significant functions are appropriately performed. Similarly. flow diagram blocks may also have been arbitrarily defined herein to illustrate certain significant functionality.
To the extent used, the flow diagram block boundaries and sequence could have been defined otherwise and still perform the certain significant functionality. Such alternate definitions of both functional building blocks and flow diagram blocks and sequences are thus within the scope and spirit of the claims. One of average skill in the art will also recognize that the functional building blocks, and other illustrative blocks, modules and components herein, can be implemented as illustrated or by discrete components, application specific integrated circuits, processors executing appropriate software and the like or any combination thereof.
In addition, a flow diagram may include a “start” and/or “continue” indication. The “start” and “continue” indications reflect that the steps presented can optionally be incorporated in or otherwise used in conjunction with one or more other routines. In addition, a flow diagram may include an “end” and/or “continue” indication. The “end” and/or “continue” indications reflect that the steps presented can end as described and shown or optionally be incorporated in or otherwise used in conjunction with one or more other routines. In this context. “start” indicates the beginning of the first step presented and may be preceded by other activities not specifically shown. Further, the “continue” indication reflects that the steps presented may be performed multiple times and/or may be succeeded by other activities not specifically shown. Further, while a flow diagram indicates a particular ordering of steps, other orderings are likewise possible provided that the principles of causality are maintained.
The one or more embodiments are used herein to illustrate one or more aspects, one or more features, one or more concepts, and/or one or more examples A physical embodiment of an apparatus, an article of manufacture a machine, and/or of a process may include one or more of the aspects. features, concepts, examples, described with reference to one or more of the embodiments discussed herein. Further. from figure to figure, the embodiments may incorporate the same or similarly named functions, steps, modules, etc, that may use the same or different reference numbers and, as such, the functions, steps, modules. etc, may be the same or similar functions, steps, modules, etc, or different ones.
Unless specifically stated to the contra, signals to, from, and/or between elements in a figure of any of the figures presented herein may be analog or digital, continuous time or discrete time, and single-ended or differential. For instance, if a signal path is shown as a single-ended path, it also represents a differential signal path. Similarly, if a signal path is shown as a differential path, it also represents a single-ended signal path. While one or more particular architectures are described herein. other architectures can likewise be implemented that use one or more data buses not expressly shown, direct connectivity between elements, and/or indirect coupling between other elements as recognized by one of average skill in the art.
The term “module” is used in the description of one or more of the embodiments. A module implements one or more functions via a device such as a processor or other processing device or other hardware that may include or operate in association with a memory that stores operational instructions. A module may operate independently and/or in conjunction with software and/or firmware. As also used herein a module may contain one or more sub-modules, each of which may be one or more modules.
As may further be used herein a computer readable memory includes one or more memory elements. A memory element may be a separate memory device, multiple memory devices, or a set of memory locations within a memory device. Such a memory device may be a read-only memory, random access memory, volatile memory, non-volatile memory, static memory, dynamic memory, flash memory, cache memory, a quantum register or other quantum memory and/or any other device that stores data in a non-transitory manner. Furthermore, the memory device may be in a form of a solid-state memory a hard drive memory or other disk storage, cloud memory, thumb drive, server memory, computing device memory, and/or other non-transitory medium for storing data. The storage of data includes temporary storage (i.e., data is lost when power is removed from the memory element) and/or persistent storage (i.e., data is retained when power is removed from the memory element). As used herein, a transitory medium shall mean one or more of: (a) a wired or wireless medium for the transportation of data as a signal from one computing device to another computing device for temporary storage or persistent storage: (b) a wired or wireless medium for the transportation of data as a signal within a computing device from one element of the computing device to another element of the computing device for temporary storage or persistent storage; (e) a wired or wireless medium for the transportation of data as a signal from one computing device to another computing device for processing the data by the other computing device; and (d) a wired or wireless medium for the transportation of data as a signal within a computing device from one element of the computing device to another element of the computing device for processing the data by the other element of the computing device. As may be used herein, a non-transitory computer readable memory is substantially equivalent to a computer readable memory. A non-transitory computer readable memory can also be referred to as a non-transitory computer readable storage medium.
One or more functions associated with the methods and/or processes described herein can be implemented via a processing module that operates via the non-human “artificial” intelligence (AI) of a machine. Examples of such A1 include machines that operate via anomaly detection techniques, decision trees, association rules, expert systems and other knowledge-based systems, computer vision models, artificial neural networks, convolutional neural networks, support vector machines (SVMs). Bayesian networks, genetic algorithms. feature learning, sparse dictionary learning, preference learning, deep learning and other machine learning techniques that are trained using training data via unsupervised, semi-supervised, supervised and/or reinforcement learning, and/or other A1. The human mind is net equipped to perform such AI techniques, not only due to the complexity of these techniques, but also due to the fact that artificial intelligence, by its very definition—requires “artificial” intelligence—i.e, machine/non-human intelligence.
One or more functions associated with the methods and/or processes described herein can be implemented as a large-scale system that is operable to receive, transmit and/or process data on a large-scale. As used herein, a large-scale refers to a large number of data, such as one or more kilobytes, megabytes, gigabytes, terabytes or more of data that are received, transmitted and/or processed. Such receiving, transmitting and/or processing of data cannot practically be performed by the human mind on a large-scale within a reasonable period of time, such as within a second, a millisecond, microsecond a real-time basis or other high speed required by the machines that generate the data, receive the data, convey the data, store the data and/or use the data.
One or more functions associated with the methods and/or processes described herein can require data to be manipulated in different ways within overlapping time spaus. The human mind is not equipped to perform such different data manipulations independently, contemporaneously, in parallel, and/or on a coordinated basis within a reasonable period of time, such as within a second a millisecond, microsecond a real-time basis or other high speed required by the machines that generate the data, receive the data, convey the data, store the data and/or use the data.
One or more functions associated with the methods and/or processes described herein can be implemented in a system that is operable to electronically receive digital data via a wired or wireless communication network and/or to electronically transmit digital data via a wired or wireless communication network. Such receiving and transmitting cannot practically be performed by the human mind because the human mind is not equipped to electronically transmit or receive digital data, let alone to transmit and receive digital data via a wired or wireless communication network.
One or more functions associated with the methods and/or processes described herein can be implemented in a system that is operable to electronically store digital data in a memory device. Such storage cannot practically be performed by the human mind because the human mind is not equipped to electronically store digital data.
One or more functions associated with the methods and/or processes described herein may operate to cause an action by a processing module directly in response to a triggering event—without any intervening human interaction between the triggering event and the action. Any such actions may be identified as being performed “automatically”. “automatically based on” and/or “automatically in response to” such a triggering event. Furthermore, any such actions identified in such a fashion specifically preclude the operation of human activity with respect to these actions—even if the triggering event itself may be causally connected to a human activity of some kind.
While particular combinations of various functions and features of the one or more embodiments have been expressly described herein, other combinations of these features and functions are likewise possible. The present disclosure is not limited by the particular examples disclosed herein and expressly incorporates these other combinations.
The present U S. Utility Patent application claims priority pursuant 10.35 U.S C. § 119(c) IO U.S. Provisional Application No. 63/380,414, entitled “QUERY FILTER PROCESSING IN DATABASE SYSTEMS”. filed 10/21/2022, which is hereby incorporated herein by reference in its entirety and made part of the present U.S. Utility Patent Application for all purposes.
Number | Date | Country | |
---|---|---|---|
20240134858 A1 | Apr 2024 | US |
Number | Date | Country | |
---|---|---|---|
63380414 | Oct 2022 | US |