Patent Application
20200241792
The current subject matter is directed to advanced database techniques in which large objects can be selective evicted from memory of an in-memory database.
SQL provides for various data types. One data type is referred to as a large object page (LOB) which allows for large pieces of information to be stored. As one example, SQL can provide for a table that has LOB columns to store large amounts of data (e.g., a maximum of 4 GB). This data can take various forms including text, audio/visual files in various media formats, and more.
In one aspect, pages are loaded into a resource container from physical disk storage into memory of an in-memory database. The resource container includes a resource provider and the pages include a plurality of large object pages (LOBs). Thereafter, the resource provider is used to allocate portions of the memory and control blocks to the pages. A job is triggered when an amount of memory allocated to LOBs by the resource provider exceeds a predefined first level. The job evicts LOBs from memory until the amount of memory allocated to LOBs is below a predefined second level.
The job can be an asynchronous job that loops over the resource container to evict LOBs. In some variations, only LOBs that are not used are evicted from the memory of the in-memory database. Further, in some variations, only LOBs that have not been used within a first predefined amount of time are evicted from the memory of the in-memory database. If the job is not able to evict a sufficient amount of LOBs after evicting only LOBs that have not been used with the first predefined amount of time, the job can evict LOBs that have not been used within a second predefined amount of time, the second predefined amount of time being more recent than the first predefined amount of time.
An out-of-memory error can be triggered if the amount of memory allocated to LOBs by the resource provider exceeds a predefined second level that is greater than the predefined first level.
The resource provider can specify an interface providing a page allocation method and a control block allocation method.
The control block can be a transient object encapsulating information characterizing the corresponding page.
Non-transitory computer program products (i.e., physically embodied computer program products) are also described that store instructions, which when executed by one or more data processors of one or more computing systems, cause at least one data processor to perform operations herein. Similarly, computer systems are also described that may include one or more data processors and memory coupled to the one or more data processors. The memory may temporarily or permanently store instructions that cause at least one processor to perform one or more of the operations described herein. In addition, methods can be implemented by one or more data processors either within a single computing system or distributed among two or more computing systems. Such computing systems can be connected and can exchange data and/or commands or other instructions or the like via one or more connections, including but not limited to a connection over a network (e.g., the Internet, a wireless wide area network, a local area network, a wide area network, a wired network, or the like), via a direct connection between one or more of the multiple computing systems, etc.
The details of one or more variations of the subject matter described herein are set forth in the accompanying drawings and the description below. Other features and advantages of the subject matter described herein will be apparent from the description and drawings, and from the claims.
Like reference symbols in the various drawings indicate like elements.
The current subject matter is directed to techniques for selectively restricting and/or evicting large object pages from memory of an in-memory database. Such an arrangement is advantageous in that it increases the overall performance of such a database. Such performance increase is accomplished by preventing large object pages from causing other pages which may be more important to a particular database operation or set of operations from being evicted from memory.
The index server 110 can contain in-memory data stores and engines for processing data. The index server 110 can also be accessed by remote tools (via, for example, SQL queries), that can provide various development environment and administration tools. Additional details regarding an example implementation of the index server 110 is described and illustrated in connection with diagram 300 of
The name server 115 can own information about the topology of the database system 105. In a distributed database system, the name server 115 can know where various components are running and which data is located on which server. In a database system 105 with multiple database containers, the name server 115 can have information about existing database containers and it can also hosts the system database. For example, the name server 115 can manage the information about existing tenant databases. Unlike a name server 115 in a single-container system, the name server 115 in a database system 105 having multiple database containers does not store topology information such as the location of tables in a distributed database. In a multi-container database system 105 such database-level topology information can be stored as part of the catalogs of the tenant databases.
The application services server 120 can enable native web applications used by one or more remote clients 150 accessing the database system 105 via a web protocol such as HTTP. The application services server 120 can allow developers to write and run various database applications without the need to run an additional application server. The application services server 120 can also be used to run web-based tools 155 for administration, life-cycle management and development. Other administration and development tools 160 can directly access the index server 110 for, example, via SQL and other protocols.
The extended store server 125 can be part of a dynamic tiering option that can include a high-performance disk-based column store for very big data up to the petabyte range and beyond. Less frequently accessed data (for which is it non-optimal to maintain in main memory of the index server 110) can be put into the extended store server 125. The dynamic tiering of the extended store server 125 allows for hosting of very large databases with a reduced cost of ownership as compared to conventional arrangements.
The DDI server 130 can be a separate server process that is part of a database deployment infrastructure (DDI). The DDI can be a layer of the database system 105 that simplifies the deployment of database objects using declarative design time artifacts. DDI can ensure a consistent deployment, for example by guaranteeing that multiple objects are deployed in the right sequence based on dependencies, and by implementing a transactional all-or-nothing deployment.
The data provisioning server 135 can provide enterprise information management and enable capabilities such as data provisioning in real time and batch mode, real-time data transformations, data quality functions, adapters for various types of remote sources, and an adapter SDK for developing additional adapters.
The streaming cluster 140 allows for various types of data streams (i.e., data feeds, etc.) to be utilized by the database system 105. The streaming cluster 140 allows for both consumption of data streams and for complex event processing.
As is illustrated in
Requests from the client applications 150 can be processed and executed by way of a request processing and execution control component 310. The database system 105 offers rich programming capabilities for running application-specific calculations inside the database system. In addition to SQL, MDX, and WIPE, the database system 105 can provide different programming languages for different use cases. SQLScript can be used to write database procedures and user defined functions that can be used in SQL statements. The L language is an imperative language, which can be used to implement operator logic that can be called by SQLScript procedures and for writing user-defined functions.
Once a session is established, client applications 150 typically use SQL statements to communicate with the index server 110 which can be handled by a SQL processor 312 within the request processing and execution control component 310. Analytical applications can use the multidimensional query language MDX (MultiDimensional eXpressions) via an MDX processor 322. For graph data, applications can use GEM (Graph Query and Manipulation) via a GEM processor 316, a graph query and manipulation language. SQL statements and MDX queries can be sent over the same connection with the client application 150 using the same network communication protocol. GEM statements can be sent using a built-in SQL system procedure.
The index server 110 can include an authentication component 304 that can be invoked with a new connection with a client application 150 is established. Users can be authenticated either by the database system 105 itself (login with user and password) or authentication can be delegated to an external authentication provider. An authorization manager 306 can be invoked by other components of the database system 105 to check whether the user has the required privileges to execute the requested operations.
Each statement can processed in the context of a transaction. New sessions can be implicitly assigned to a new transaction. The index server 110 can include a transaction manager 344 that coordinates transactions, controls transactional isolation, and keeps track of running and closed transactions. When a transaction is committed or rolled back, the transaction manager 344 can inform the involved engines about this event so they can execute necessary actions. The transaction manager 344 can provide various types of concurrency control and it can cooperate with a persistence layer 346 to achieve atomic and durable transactions.
Incoming SQL requests from the client applications 150 can be e received by the SQL processor 312. Data manipulation statements can be executed by the SQL processor 312 itself. Other types of requests can be delegated to the respective components. Data definition statements can be dispatched to a metadata manager 306, transaction control statements can be forwarded to the transaction manager 344, planning commands can be routed to a planning engine 318, and task related commands can forwarded to a task manager 324 (which can be part of a larger task framework) Incoming MDX requests can be delegated to the MDX processor 322. Procedure calls can be forwarded to the procedure processor 314, which further dispatches the calls, for example to a calculation engine 326, the GEM processor 316, a repository 300, or a DDI proxy 328.
The index server 110 can also include a planning engine 318 that allows planning applications, for instance for financial planning, to execute basic planning operations in the database layer. One such basic operation is to create a new version of a data set as a copy of an existing one while applying filters and transformations. For example, planning data for a new year can be created as a copy of the data from the previous year. Another example for a planning operation is the disaggregation operation that distributes target values from higher to lower aggregation levels based on a distribution function.
The SQL processor 312 can include an enterprise performance management (EPM) runtime component 320 that can form part of a larger platform providing an infrastructure for developing and running enterprise performance management applications on the database system 105. While the planning engine 318 can provide basic planning operations, the EPM platform provides a foundation for complete planning applications, based on by application-specific planning models managed in the database system 105.
The calculation engine 326 can provide a common infrastructure that implements various features such as SQLScript, MDX, GEM, tasks, and planning operations. The SQLScript processor 312, the MDX processor 322, the planning engine 318, the task manager 324, and the GEM processor 316 can translate the different programming languages, query languages, and models into a common representation that is optimized and executed by the calculation engine 326. The calculation engine 326 can implement those features using temporary results 340 which can be based, in part, on data within the relational stores 332.
Metadata can be accessed via the metadata manager component 306. Metadata, in this context, can comprise a variety of objects, such as definitions of relational tables, columns, views, indexes and procedures. Metadata of all these types can be stored in one common database catalog for all stores. The database catalog can be stored in tables in a row store 336 forming part of a group of relational stores 332. Other aspects of the database system 105 including, for example, support and multi-version concurrency control can also be used for metadata management. In distributed systems, central metadata is shared across servers and the metadata manager component 306 can coordinate or otherwise manage such sharing.
The relational stores 332 form the different data management components of the index server 110 and these relational stores can, for example, store data in main memory. The row store 336, a column store 338, and a federation component 334 are all relational data stores which can provide access to data organized in relational tables. The column store 338 can stores relational tables column-wise (i.e., in a column-oriented fashion, etc.). The column store 338 can also comprise text search and analysis capabilities, support for spatial data, and operators and storage for graph-structured data. With regard to graph-structured data, from an application viewpoint, the column store 338 could be viewed as a non-relational and schema-flexible in-memory data store for graph-structured data. However, technically such a graph store is not a separate physical data store. Instead it is built using the column store 338, which can have a dedicated graph API.
The row store 336 can stores relational tables row-wise. When a table is created, the creator can specify whether it should be row or column-based. Tables can be migrated between the two storage formats. While certain SQL extensions are only available for one kind of table (such as the “merge” command for column tables), standard SQL can be used on all tables. The index server 110 also provides functionality to combine both kinds of tables in one statement (join, sub query, union).
The federation component 334 can be viewed as a virtual relational data store. The federation component 334 can provide access to remote data in external data source system(s) 354 through virtual tables, which can be used in SQL queries in a fashion similar to normal tables.
The database system 105 can include an integration of a non-relational data store 342 into the index server 110. For example, the non-relational data store 342 can have data represented as networks of C++ objects, which can be persisted to disk. The non-relational data store 342 can be used, for example, for optimization and planning tasks that operate on large networks of data objects, for example in supply chain management. Unlike the row store 336 and the column store 338, the non-relational data store 342 does not use relational tables; rather, objects can be directly stored in containers provided by the persistence layer 346. Fixed size entry containers can be used to store objects of one class. Persisted objects can be loaded via their persisted object IDs, which can also be used to persist references between objects. In addition, access via in-memory indexes is supported. In that case, the objects need to contain search keys. The in-memory search index is created on first access. The non-relational data store 342 can be integrated with the transaction manager 344 to extends transaction management with sub-transactions, and to also provide a different locking protocol and implementation of multi version concurrency control.
An extended store is another relational store that can be used or otherwise form part of the database system 105. The extended store can, for example, be a disk-based column store optimized for managing very big tables, which ones do not want to keep in memory (as with the relational stores 332). The extended store can run in an extended store server 125 separate from the index server 110. The index server 110 can use the federation component 334 to send SQL statements to the extended store server 125.
The persistence layer 346 is responsible for durability and atomicity of transactions. The persistence layer 346 can ensure that the database system 105 is restored to the most recent committed state after a restart and that transactions are either completely executed or completely undone. To achieve this goal in an efficient way, the persistence layer 346 can use a combination of write-ahead logs, undo and cleanup logs, shadow paging and savepoints. The persistence layer 346 can provide interfaces for writing and reading persisted data and it can also contain a logger component that manages a recovery log. Recovery log entries can be written in the persistence layer 346 (in recovery log volumes 352) explicitly by using a log interface or implicitly when using the virtual file abstraction. The recovery log volumes 352 can include redo logs which specify database operations to be replayed whereas data volume 350 contains undo logs which specify database operations to be undone as well as cleanup logs of committed operations which can be executed by a garbage collection process to reorganize the data area (e.g. free up space occupied by deleted data etc.).
The persistence layer 346 stores data in persistent disk storage 348 which, in turn, can include data volumes 350 and/or recovery log volumes 352 that can be organized in pages. Different page sizes can be supported, for example, between 4k and 16M. Data can be loaded from the disk storage 348 and stored to disk page wise. For read and write access, pages can be loaded into a page buffer in memory. The page buffer need not have a minimum or maximum size, rather, all free memory not used for other things can be used for the page buffer. If the memory is needed elsewhere, least recently used pages can be removed from the cache. If a modified page is chosen to be removed, the page first needs to be persisted to disk storage 348. While the pages and the page buffer are managed by the persistence layer 346, the in-memory stores (i.e., the relational stores 332) can access data within loaded pages.
As noted above, the data volumes 350 can include a data store that together with undo and cleanup log and recovery log volumes 352 comprise the recovery log. Other types of storage arrangements can be utilized depending on the desired configuration. The data store can comprise a snapshot of the corresponding database contents as of the last system savepoint. The snapshot provides a read-only static view of the database as it existed as of the point (i.e., time, etc.) at which it was created. Uncommitted transactions, at such time, are not reflected in the snapshot and are rolled back (i.e., are undone, etc.). Database snapshots operate at the data-page level such that all pages being modified are copied from the source data volume to the snapshot prior to their being modified via a copy-on-write operation. The snapshot can store such original pages thereby preserving the data records as they existed when the snapshot was created.
System savepoints (also known in the field of relational database servers as checkpoints) can be periodically or manually generated and provide a point at which the recovery log can be truncated. The savepoint can, in some variations, include an undo log of transactions which were open in the savepoint and/or a cleanup log of transactions which were committed in the savepoint but not yet garbage collected (i.e., data which has been deleted by these transactions has been marked as deleted but has not been deleted in a physical manner to assure multiversion concurrency control).
The recovery log can comprise a log of all changes to the database system 105 since the last system savepoint, such that when a database server is restarted, its latest state is restored by replaying the changes from the recovery log on top of the last system savepoint. Typically, in a relational database system, the previous recovery log is cleared whenever a system savepoint occurs, which then starts a new, empty recovery log that will be effective until the next system savepoint. While the recovery log is processed, a new cleanup log is generated which needs to be processed as soon as the commit is replayed to avoid a growing data area because of deleted but not garbage collected data.
As part of a database system recovery/restart, after the savepointed state of data is restored, and before processing of the recovery log commences, all cleanup logs can be iterated through and, in implementations using a history manager, passed to the history manager for asynchronous garbage collection processing.
In addition, it can be checked if there are older versions of the cleanup log present in the savepoint which need to be processed synchronously with regard to the recovery log. In such cases, recovery log processing can wait until garbage collection of old versions of cleanup logs finish. However, recovery log processing can commence when there are newer versions of cleanup logs for garbage collection. In cases in which no old versions of cleanup logs exist, recovery log replay can start immediately after the cleanup log from the savepoint has been passed to the history manager.
A typical savepoint can have three phases. First, in the pre-critical phase all modified pages in the relational stores 332 (which are loaded into memory) can be iterated through and flushed to the physical persistence disk storage 348. Second, a critical phase can block all parallel updates to pages in the relational stores 332 and trigger all the remaining I/O (i.e., I/O for all pages that are still being modified when entering the critical phase) for the physical persistence disk storage 348 to ensure the consistent state of data. Lastly, a post-critical phase can wait for all remaining I/O associated with the physical persistence disk storage 348.
As referenced above, objects can be directly stored in containers provided by the persistence layer 346. One type of container is a resource container which, in turn, can include different types of pages such as, for example, normal data pages, converter pages, and LOBs. As noted above, LOB can comprise big data which is stored in a SQL table as data type LOB. LOB data can be very large but sometimes only rarely read. Therefore, if LOB data is needed to be read into memory from the physical persistence (i.e., the disk storage 348), the amount of LOB pages might want to be limited within resource container so that other resources do not get evicted from memory. If a large table with LOB is scanned, all data is read into memory but may not be required afterwards. Reading the LOB into memory might evict other data that is needed to be kept in memory to the disk storage 348.
There can be one resource container that specifies multiple resource providers (e.g., a default provider, an LOB provider, etc.). The container implementation (e.g. VirtualFile, MidSizeLOBContainer, etc.) specifies, which resource provider to be used. This resource provider can be used instead of a memory allocator for allocating pages and control block in memory. The resource provider can be used during page access (instead of using heap allocator) and can have elaborate logic inside. For example, the resource provider can be an interface that has its own memory allocator and allocator statistics can be used to determine how much memory is allocated to LOBs including, for example, various associated limits for LOBs. The control block can specify, for example, if a page is written to disk and can be a transient object that encapsulates information about the LOB page. For example, the control block contains information about whether a page is modified, whether the page is in I/O, etc.
A database administrator can set various levels/limits/thresholds in connection with the handling of LOBs. For example, a soft upper limit can be specified that triggers a cleanup of LOBs (e.g., the soft upper limit can be set to 200 GB, etc. When the size of the LOBs in the resource container exceeds the soft upper limit, the resource provider can trigger a job (e.g., an asynchronous job, etc.) that can, for example, loop over the resource container and evict LOB pages from memory. For example, a lower limit can be established the database administrator (e.g., 180 GB) such that the job attempts to evict LOB pages from memory to a lower threshold. If all LOB pages are referenced (i.e., they are being used, etc.) more memory can be allocated if needed and can be temporarily over the upper soft limit. The database administrator can also set a hard upper limit (e.g., 250 GB) which, when exceeded, can cause an out-of-memory error to be thrown.
The job can utilize various mechanisms in order to determine which LOB pages to evict from memory. For example, the job can attempt to evict (i.e., get rid) of LOB pages that are rarely used. Rarely used, in this context, can be set by a first time threshold. As an example, all LOB pages 8 hours or older can evicted. However, in some cases, there may not be any or an adequate number of LOB pages and so further LOB pages need to be evicted. The first time threshold can be modified and/or replaced with a second time threshold which is smaller than the first time threshold (e.g., 6 hours v. 8 hours, etc.) and the job can continue with evicting any LOB pages older than the second time threshold. The job can continue this process in an iterative manner until such time that there are sufficient LOB pages evicted from memory (e.g., as set by the soft upper limit, etc.).
Eviction of LOB pages can occur by having a resource disposition for such pages being set to temporary. Setting the LOB pages to temporary in this manner causes them to be evicted from memory.
In some cases, multiple pages can be stitched together to store one LOB (if the size of the LOB exceeds the page size utilized by the database system 105).
One or more aspects or features of the subject matter described herein can be realized in digital electronic circuitry, integrated circuitry, specially designed application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) computer hardware, firmware, software, and/or combinations thereof. These various aspects or features can include implementation in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device. The programmable system or computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
These computer programs, which can also be referred to as programs, software, software applications, applications, components, or code, include machine instructions for a programmable processor, and can be implemented in a high-level procedural language, an object-oriented programming language, a functional programming language, a logical programming language, and/or in assembly/machine language. As used herein, the term “computer-readable medium” refers to any computer program product, apparatus and/or device, such as for example magnetic discs, optical disks, memory, and Programmable Logic Devices (PLDs), used to provide machine instructions and/or data to a programmable processor, including a computer-readable medium that receives machine instructions as a computer-readable signal. The term “computer-readable signal” refers to any signal used to provide machine instructions and/or data to a programmable processor. The computer -readable medium can store such machine instructions non-transitorily, such as for example as would a non-transient solid-state memory or a magnetic hard drive or any equivalent storage medium. The computer -readable medium can alternatively or additionally store such machine instructions in a transient manner, for example as would a processor cache or other random access memory associated with one or more physical processor cores.
In one example, a disk controller 548 can interface one or more optional disk drives to the system bus 504. These disk drives can be external or internal floppy disk drives such as 560, external or internal CD-ROM, CD-R, CD-RW or DVD, or solid state drives such as 552, or external or internal hard drives 556. As indicated previously, these various disk drives 552, 556, 560 and disk controllers are optional devices. The system bus 504 can also include at least one communication port 520 to allow for communication with external devices either physically connected to the computing system or available externally through a wired or wireless network. In some cases, the communication port 520 includes or otherwise comprises a network interface.
To provide for interaction with a user, the subject matter described herein can be implemented on a computing device having a display device 540 (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information obtained from the bus 504 to the user and an input device 532 such as keyboard and/or a pointing device (e.g., a mouse or a trackball) and/or a touchscreen by which the user can provide input to the computer. Other kinds of input devices 532 can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback by way of a microphone 536, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input. In the input device 532 and the microphone 536 can be coupled to and convey information via the bus 504 by way of an input device interface 528. Other computing devices, such as dedicated servers, can omit one or more of the display 540 and display interface 514, the input device 532, the microphone 536, and input device interface 528.
To provide for interaction with a user, the subject matter described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) and/or a touchscreen by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.
In the descriptions above and in the claims, phrases such as “at least one of” or “one or more of” may occur followed by a conjunctive list of elements or features. The term “and/or” may also occur in a list of two or more elements or features. Unless otherwise implicitly or explicitly contradicted by the context in which it is used, such a phrase is intended to mean any of the listed elements or features individually or any of the recited elements or features in combination with any of the other recited elements or features. For example, the phrases “at least one of A and B;” “one or more of A and B;” and “A and/or B” are each intended to mean “A alone, B alone, or A and B together.” A similar interpretation is also intended for lists including three or more items. For example, the phrases “at least one of A, B, and C;” “one or more of A, B, and C;” and “A, B, and/or C” are each intended to mean “A alone, B alone, C alone, A and B together, A and C together, B and C together, or A and B and C together.” In addition, use of the term “based on,” above and in the claims is intended to mean, “based at least in part on,” such that an unrecited feature or element is also permissible.
The subject matter described herein can be embodied in systems, apparatus, methods, and/or articles depending on the desired configuration. The implementations set forth in the foregoing description do not represent all implementations consistent with the subject matter described herein. Instead, they are merely some examples consistent with aspects related to the described subject matter. Although a few variations have been described in detail above, other modifications or additions are possible. In particular, further features and/or variations can be provided in addition to those set forth herein. For example, the implementations described above can be directed to various combinations and subcombinations of the disclosed features and/or combinations and subcombinations of several further features disclosed above. In addition, the logic flows depicted in the accompanying figures and/or described herein do not necessarily require the particular order shown, or sequential order, to achieve desirable results. Other implementations may be within the scope of the following claims.
