SELF-DESCRIPTIVE SHARED MEMORY FOR METRICS COLLECTION

Abstract

A method of operating a data storage system to obtain metrics data describing performance and capacities of the data storage system and provide the metrics data to external metrics users includes, by a metrics data producer, storing metrics data in a data region of a shared memory, and populating a metadata region of the shared memory with metadata describing a structure of the data region. The method further includes, by a metrics data collector, reading the metadata of the shared memory and using the metadata to tailor a manner of access of the data region by the metrics data collector for correctly accessing the stored metrics data.

Description

BACKGROUND

The invention is related to the field of data storage systems.

SUMMARY

A method is disclosed of operating a data storage system to obtain metrics data describing performance and capacities of the data storage system and provide the metrics data to external metrics users. Generally, the method includes, by a metrics data producer, storing metrics data in a data region of a shared memory, and populating a metadata region of the shared memory with metadata describing a structure of the data region. The method further includes, by a metrics data collector, reading the metadata of the shared memory and using the metadata to tailor a manner of access of the data region by the metrics data collector for correctly accessing the stored metrics data.

In some embodiments, the metrics data is for a set of objects of a single type in the data storage system for which respective performance or capacity data is obtained, and the data region stores a plurality of same-structured multi-field records for respective ones of the objects, and the metadata region includes a plurality of field definition areas each storing a respective field definition describing structure and content of a respective field of the records, each field definition being used by the metrics data collector to locate and interpret the respective field in each of the records. Further, the field definitions may each include at least a field name, field location, and field size for the respective field, the field name identifying contents of the field, the field location identifying a relative location of the field in each of the records, the field size identifying a size of the field.

The metadata region may further include a general area containing data items describing general aspects of the metrics data common to the records for the set of objects, the data items including a fields number identifying the number of fields in each of the records, the fields number being used by the metrics data collector to parse a correct number of fields of each of the records. The data items of the general area may further include an update type and an update interval, the update type indicating whether the metrics data is updated at regular intervals or otherwise, the update interval identifying an interval length of the regular intervals, the update type and update interval being used by the metrics data collector to control timing of accessing the metrics data to obtain all metrics data as it is updated by the metrics data producer. The data items of the general area may further include a layout type identifying a high-level structure of the data region, the layout type being selected from an array type and a map type, the array type having the records arranged in a single area with no intra-record indirection, the map type having the records arranged across multiple areas with at least one intra-record indirection from one area to another.

In some embodiments, the metrics data producer and metrics data collector operate according to a protocol. Producer operations include, when initially creating a new record in the data region and when reusing a record of the data region, (i) generating a data generation identifier and writing it into the record, (ii) initializing other record fields, (iii) initializing corresponding metadata in the metadata region, and (iv) setting a metadata generation identifier to equal the data generation identifier; and when removing a record from the data region, changing the data generation identifier to an invalid value, and changing the metadata generation identifier to the invalid value. Collector operations include, for collecting a specific instance of the metrics data, copying the entirety of the metrics data from the shared memory to private memory of the collector, and scanning the processing the records of the metrics data by, for each record: (i) skipping further processing of a record when its metadata generation identifier is invalid and when its metadata generation identifier is not equal to its data generation identifier; and (ii) collecting record data for downstream use when the metadata generation identifier is valid and equal to the data generation identifier.

In other aspects, the metrics data collector may provide metrics data to the metrics users using one of a file and a database, and/or the metrics data producer may register itself by storing registration data in a collector configuration readable by the metrics data collector, the registration enabling the metrics data collector to identify the metrics data producer and begin collecting the metrics data in subsequent operation.

The technique may also be used by multiple metrics data producers and collectors for respective metrics data in the system. Each producer populates a respective data region and metadata region in the general manner described above, and each collector reads a respective metadata region and uses the respective data and metadata in the general manner described above.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, features and advantages will be apparent from the following description of particular embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views.

FIG. 1 is a block diagram of a data storage system;

FIG. 2 is a schematic diagram of an arrangement of communicating metrics data from a producer to a collector using a self-describing shared memory;

FIG. 3 is a schematic diagram of structure and content of a self-describing shared memory according to a first embodiment;

FIG. 4 is a schematic diagram of structure and content of a self-describing shared memory according to a second embodiment;

FIG. 5 is a high-level general flow diagram of a method of using a self-describing shared memory; and

FIG. 6 is a high-level general flow diagram of a method of ensuring consistency of data communicated from a producer to a collector.

DETAILED DESCRIPTION

Overview

The disclosed technique is directed to systems such as data storage systems that track and maintain performance and capacity metrics during operation, which can be used by a variety of automated or human entities for different purposes. A given system may track a large number (e.g., hundreds) of metrics, for which there are respective data producers and collectors, and for both software and hardware components in the system. The metrics data may be exposed to customers and/or maintenance personnel such as field support engineers. In one example, a telemetry service collects metrics data for offline analytics to provide potential service recommendations, such as extending disk capacity based on usage or detecting disks to be replaced for better overall system performance. In another example, a storage administrator uses performance metrics data to monitor current system throughput (input/output operations per second (IOPS)) or operations latencies to ensure proper system operation.

In known solutions, metrics data collectors send requests to metrics data producers to request metrics data. Each metrics data producer gathers the metrics data and sends it back to a metrics data collector. The producers work with the collectors together to create data transfer channels by which the producers transfer the metrics data to the collectors, according to an aligned protocol and predefined data format definition.

Depending on different designs for the protocol and the communication channel between the producer and the collector, there can be various problems or limitations with such known solutions:

1. Scalability

• • For existing solutions, the producer commonly sends the metrics data via either the intra-network or external network to the collector. Each collector is required to create a dedicated session for each producer. However, in a typical system there is an overall limit on the number of open network sockets that are permitted, so this technique can be limited and may not scale well.

2. Efficiency

• • For metric data transferred over the network, the producer's metrics data is first copied from the producer's memory to a socket buffer, and after being received on the collector side, it is copied from the socket buffer to the collector's memory. Hence at least two memory copies of the same metrics data are needed. Although there are known utilities/libraries that can be leveraged to transmit a producer's memory contents to memory of the collector, such a technique still requires that all the producers and collectors are bound to a specific library and must understand all its complexities. Also, network latency could be a concern when the metrics data size is large while the network bandwidth is constrained and potentially insufficient.

3. Maintenance

• • There are several ways to organize a format for the metrics data. The data format could be specific for each producer, as different producers generate different metrics. Such an approach can be straightforward and thus may be used in closely coupled systems. However, the collector needs to a priori understand the data format from each producer. Any metrics update or new metrics in producers would introduce changes to the collector. Hence, such an approach may not be extensible beyond relatively small and constrained systems.
  • As an example, one common format may be shared by all producers and collectors, e.g., by sharing the same data structure across different producers. This binds all the producers and collectors to a given language, which may not be practical for a system developed by a large, diverse organization or across multiple organizations. Alternatively, a format can be serialized and deserialized by different components in different languages, which is an approach used in the so-called “Protobuf” for exchanging data between endpoints. However, the Protobuf is generally used only for small amounts of data of control messages. This approach may not be usable with larger data sets that could be dozens or hundreds of megabytes in size.

Thus, what is needed is an efficient mechanism that can decouple producers and collectors to enable producers to introduce new metrics readily and independently, or update existing metrics, without requiring changes to the collectors for such new or updated metrics.

In general, the disclosed technique involves use of a self-descriptive shared memory structure used for data transfer from metrics data producers to metrics data collectors. The producer populates the shared memory according to a definition that is itself stored in the shared memory along with metrics data values, and the collector loads the definition to enable it to accurately parse the shared memory according to the definition to retrieve the metrics data. Each metrics type has one or more fields for values, which in some cases are counters for dynamic aspects of operations (e.g., counts of operations, data transfer, etc.). For a given metric there are thus two shared memory regions, a metadata region and a data region holding the metrics data to be collected. The metadata region includes common attributes and short descriptors describing the structure of the data region for use by the collectors to tailor their accesses to the data region to correctly access stored metrics data.

Also described is a technique for ensuring data consistency in operation. Metrics data in shared memory is created and updated by a producer. A collector reads the shared memory at regular collection intervals according to an interval size defined in the metadata region. To ensure that the collector reads the correct data, while maintaining a desired decoupling of the collector and the producer, the technique employs a time_of_day (TOD) variable, which is the timestamp in seconds, stored in the shared memory (e.g., at the beginning of the data region). Then for a metrics type that is updated at a fixed interval, the TOD value is first loaded and checked. If the target collection timestamp equals the TOD value, the metrics data is properly updated and valid, and can be collected. Otherwise, the metrics data is stale and invalid, and should not be collected. For a metrics type that is updated dynamically at any time, TOD variable can also be checked. However, the producer might change the metrics data quite frequently, and it may be inefficient to update the timestamp each time the producer updates the metrics data. Hence the collector might load and use the metrics data unless the TOD value is older than the target collection timestamp by more than some desired amount (e.g., more than 10 seconds).

Embodiments

FIG. 1 shows a data storage system 10 as including interface circuitry 12, storage devices (DEVs) 14, and processing circuitry 16, which is shown as including storage operations processing (STG OP PROC'G) 18 and a metrics subsystem 20. In operation as generally known, the data storage system 10 stores data of storage users 22 on the storage devices 14 and provides stored data to the users 22. These operations are referred to as storage operations (e.g., write and read operations), and they are executed by the storage operations processing circuitry 18. In addition, the metrics subsystem 20 collects and stores various operating metrics and capacities and makes the resulting stored metrics data available to metrics users 24, e.g., system administrators, etc. A “metric” is a monitored or measured value reflecting associated storage operations, such as a measure of data transfer rate, operation rate, etc., which typically fluctuates throughout operation and is thus monitored continually, typically at some interval (e.g., 1-second interval). A “capacity” is a static or dynamic measure of resource availability or utilization, e.g., a configured size of a logical volume or other storage object. Illustrative examples are described below.

FIG. 2 is a schematic diagram of the metrics system 20. At a high level, it includes metrics producers 30 (two producers 30-A, 30-B shown), metrics collectors 32 (one shown), and a shared memory 34 by which metric data produced by the producers 30 is communicated to the collectors 32. A metrics collector 32 may provide metrics data to the metrics users 24 in a variety of ways, including use of a file such as a comma-separated variable (CSV) file 36 or use of a database (DB) 38. In some embodiments it may be desired to deploy producers 30 and/or collectors 32 in respective containers such as docker containers, an example of which is shown as container 40 for producer 30-B. Also, the metrics subsystem 20 includes collector configuration data 42 used to communicate information about the producers 30 and their metrics data sets to a collector 32, as described more below.

The producers 30 communicate metrics data to a collector 32 using respective data sets in the shared memory 34, each of which has a “data” region for the metrics data and a “metadata” (M-D) region for metadata that describes the structure of the data region. In the example of FIG. 2, producer 30-A is shown as using an “A” data region 44-A and an “A” metadata region 46-A, and the producer 30-B is shown as using a “B” data region 44-B and a “B” metadata region 46-B. The collector 32 includes a metadata reader 48 and a data reader 50, along with post processing 52 and one or more writers for providing the output metrics data in a required format, e.g., a CSV writer 54 and a DB writer 56 as shown. The collector 32 is also shown as including a configuration reader 58 for reading the collector configuration data 42. As indicated at 59, the producers 30 register themselves by storing registration data in the collector configuration 42, enabling the collectors 32 to identify new producers 30 and begin collecting their respective metrics data in subsequent operation.

It will be appreciated that in general a system will include multiple producers 30 and collectors 32 for respective distinct metrics data, and that the structuring and functionality described herein generally applies to producer-collector pairs (or sets) and respective sets of metrics data, each having a respective data and metadata component. The distinct sets of metrics data will relate to respective distinct objects or components of interest in the system 10. As an example, while one set of metrics data may be Volume Space metrics such as in an example given below, the system may also produce and collect myriad other types of metrics data, each generally requiring a different producer/collector pair or set. For example, another type of metrics data may be performance data such as input/operations per second for different components within the system, and this data may be gathered using a respective distinct producer/collector pair or set (separate from the producer/collector for Volume Space Metrics, for example).

FIG. 3 is a schematic depiction of a data region 44 and metadata region 46 for an example metric, “Volume Space Metrics”. The arrangement of FIG. 3 is also an example of one type of structuring or layout of these regions, which is referred to as “Array” layout type. An alternative “Map” layout type is described below.

The data region 44 is shown as storing a time-of-day (TOD) variable 60 as well as a plurality of records for respective distinct storage volumes of the DSS 10. The records are indicated in FIG. 3 by respective indicators Index 0, Index 1, etc. which indicate the relative locations of the records in the data region 44. In this example, each record has three fields of different types shown as Field 1, Field 2, and Field 3. These are a Volume Identifier (VOL ID, a string), Committed Size (SIZE, an integer representing size in known units, e.g., MBs), and Logical Space (SPACE, an integer of similar type as SIZE) respectively. In typical use, a Volume ID will follow some system-defined scheme by which storage volumes are named. As a simple example, a system may simply assign monotonically increasing volume numbers as new volumes are created, in which case a Volume ID is a number of some known size and format. Other volume naming schemes are of course possible.

The metadata region 46 is shown as having a General area 62 and a succession of Field definition areas 64, which in this example are three in number (64-1, 64-2 and 64-3). The General 62 area includes several data items describing the metric data collection generally, including the number of fields that are contained in each record. Additional detailed description is given below. Each Field area 64 describes the structure of a corresponding field of a record. For this example having three fields for each record, there are three Field definition areas 64, each corresponding to a respective field type as indicated by the dotted arrows. The Field 1 definition 64-1 describes the structure of the first field (Field 1), which in this example is the Volume ID. The Field 2 definition 64-2 describes the structure of the second field (Field 2), which in this example is Committed Size. The Field 3 definition 64-3 describes the structure of the third field (Field 3), which in this example is Logical Space. It will be understood that each record contains the same three fields Field 1, Field 2 and Field 3 having the same defined structure. In the illustrated embodiment, each Field definition 64 includes five values, which in this example are Name, Usage, Type, Size and Offset for the respective field. These are described in more detail below. Overall, the Field definitions 64 collectively describe the structure of each record of the data region 44.

Table 1 below describes structure and content of the General area 62 for a given metric, such as the example Volume Space Metrics given above.

TABLE 1
Per-Metric General Metadata 62
Segment Name	Type	Description
metric_name	String	An identifying name of the metric/capacity,
		e.g., “Volume Space Metrics”
metric_path	String	The path or location of the data region for
(“Path”)		this metric, to enable a collector 32 to
		locate and load the metric data
layout_type	Enum	Type indicator for layout of data region, e.g.,
		“1” for Array-based and “2” for Map-based
update_type	Enum	Indicator for manner by which the producer
		30 updates the metric data, i.e., at a fixed
		interval or dynamically at any time. Use of
		this value helps ensure data consistency
		between producer and collector without
		dependency or synchronization
update_interval	Integer	Update interval (e.g., in seconds) for metric
(“Interval”)		using fixed-interval update. The producer
		updates the data at least once within
		the interval
entry_limit	Integer	The maximum record count of the metric
entry_size	Integer	The total memory size of one record of
(“Size”)		the metric
#_fields	Integer	The number of fields in each record of
		the metric

It will be appreciated that the general metadata 62 is used by a collector 32 to control some aspects of its accessing of the metrics data. A collector 62 uses the layout_type to understand high-level structure of the data region 44, i.e., whether the records are arranged in a single area without any internal indirection or in multiple areas with indirection, such as described below for an alternative Map-based structure. Update_type and update_interval are used to control the nature of timing of collector interrogating of metrics data as it is regularly updated during a period of operation. For example, for data subject to periodic updating, the collector 62 uses the update_interval to set its own timer for how often to perform a collection cycle. Entry_size and #_fields are used to control exactly how a collector 32 locates and parses the data as distributed in the various fields of each record, i.e., to locate the start and extent of the data of each field, and to segregate and identify collected data accordingly.

Table 2 below describes structure and content of each Field definition area 64 for the metric:

TABLE 2
Per-Field Metadata 64
Segment Name	Type	Description
field_name	String	The name of the field.
(“Name”)		In this example, these are Volume ID (VOL
		ID), Committed Size (SIZE) and Logical
		Space (SPACE) for the three fields Field 1,
		Field 2, and Field 3 respectively
field_usage	Enum	Flag indicating whether the field is a key or
(“Usage”)		a counter. In this example, Field 1 (Volume
		ID) is a Key, and the other two fields are
		Counters
field_data_type	Enum	The data type of the field, such as u64, into
(“Type”)		which the memory segment is expected to be
		translated.
		In this example, Field 1 is a string, and
		Fields 2 and 3 are u64 integers
field_size	Integer	The memory size (e.g., in bytes) of the field
(“Size”)
field_offset	Integer	The offset (e.g., in bytes) to the beginning
(“Offset”)		of the field in each record

As with the general metadata 62, it will be appreciated that the per-field metadata 64 is used by a collector 32 to control other aspects of its accessing of the metrics data, such as how to interpret the data (according to field_type), and the size and relative location of the field data in the record (field_size, field_offset).

FIG. 4 is a schematic depiction of the data region 44 and metadata region 46 for the same example metric, e.g., “Volume Space Metrics”, according to the alternative Map-type layout as mentioned above.

The data region 44 is shown as storing the time-of-day (TOD) variable 60 as well as a plurality of records for respective distinct storage volumes of the system. The data region 44 is structured to include a key section 72 and a data section 74 to contain the records. In this example, each record again has the three fields Volume Identifier (VOL ID), Committed Size (SIZE), and Logical Space (SPACE) respectively. The Volume Identifier (Field 1) is stored in the key section 72 along with a metadata generation identifier (M-D GEN ID) and an offset (OFFSET) pointing to the remaining fields in the data section 74. For each record, the data section 74 includes a data generation identifier (D GEN ID) along with the Size and Space fields (Field 2 and Field 3 respectively).

The metadata region 46 is shown as having a General area 76 and a succession of Field definition areas 78. These are generally the same as the corresponding fields 62, 64 of the Array layout of FIG. 3 as described above, with the respective offsets being interpreted as pointing to either the key section 72 or data section 74 as indicated by the dashed arrows.

The following describes a Producer/Collector access protocol using the shared memory 34 structured in the map-based manner of FIG. 4. The following description uses the terms “counter”, “entry”, and “tuple” to describe what has also been referred to as a “record”, i.e., the data content in the data region 44 for a specific metric (e.g., volume space metric).

Producer 30:

• • 1 When initially creating a new record, the producer 30 generates a uint64 data generation id and writes it into the record, initializes other record/counter fields, initializes the corresponding metadata, and sets the metadata generation id to equal the generation id.
  • 2 When removing a record, the producer 30 changes the data generation id to invalid (e.g., 0), then changes the metadata generation id to invalid (e.g., 0).
  • 3 When reusing a record, the producer 30 generates a new generation id and writes it into the record, initializes other record/counter fields, initializes the corresponding metadata, and sets the metadata generation id to equal the generation id.

Collector 32:

• • 1. The collector copies the entire metrics data collection in the shared memory 34 into its own internal memory at a given single TOD
  • 2. Scan ID array tuples one by one. For each tuple,
    • a. If metadata generation id is invalid (0), the tuple is skipped
    • b. If metadata generation id and data generation id aren't equal, the tuple is skipped
    • c. If metadata generation id and data generation id are equal, process the object data pointed by the tuple offset field.

Below is a simplified example for a set of three distinct volumes:

• • 1. Volume with id “11111111” is valid, metadata generation id and data generation id are equal, the collector will process this volume counter from copy.
  • 2. Volume with id “22222222” is being removed, its data generation id has been set to 0 (invalid) at the beginning. But its meta generation id still has a valid value (of 200, for example), i.e., it has not been cleared. An inconsistency is detected, and the collector won't process this volume counter.
  • 3. Volume with id “33333333” is being allocated, its data generation id (e.g., 300) has been allocated, its counter fields have been initialized to 0, its metadata (ID, offset) has been established but its generation id in metadata has not yet been updated. So, the metadata generation is invalid, and the collector won't process this volume counter.

Data Consistency

The metrics data in the shared memory 34 is created and updated by the producer 30. The collector 32 is reading the memory according to a collection interval defined in the metadata region 46. To decouple the collector 32 and the producer 30, conventional mechanisms of synchronizing memory reads and writes (e.g., semaphores) are avoided. To make sure the collector 32 reads the correct data, the system uses the additional memory segment time_of_day (TOD) 60, which is the timestamp in seconds, at the beginning of the data region 44.

Then for a metrics type that is updated at a fixed interval, the TOD 60 is first loaded and checked. If the target collection timestamp equals the TOD 60 in the shared memory, the metrics data is properly updated and valid, and should be collected. Otherwise, the metrics data is stale and invalid, and shouldn't be collected.

For a metrics type that is updated dynamically at any time, TOD 60 is also checked. However, the producer 30 could change the metrics data quite frequently, and it may be inefficient to update the timestamp each time when the producer updates the metrics data. Hence the collector 32 would load the metrics data unless the TOD 60 is older than the target collection timestamp by more than some predefined value (e.g., 10 seconds).

FIG. 5 is a general, high-level flow diagram of basic operation for obtaining metrics data describing performance and capacities of a system and provide the metrics data to external metrics users.

At 80, by a metrics data producer, metrics data is stored in a data region of a shared memory (e.g., 34), and a metadata region of the shared memory is populated with metadata describing a structure of the data region. Referring to the example of FIG. 3, the producer 30 populates the 3-field records of the respective storage volumes at their respective indexes (0, 1, . . . , N) in the data region, and populates the general are 62 and field areas 64 with information describing the metric generally (e.g., name, type, interval, etc.) and the 3-field structure and content of the stored metrics data).

At 82, by a metrics data collector, the metadata of the shared memory is read and used to tailor a manner of access of the data region by the metrics data collector for correctly accessing the stored metrics data. Referring to the example of FIG. 4, the collector 32 examines the metadata of the metadata region 46 to obtain the description of the metric and the structuring of the metric data stored in the data region 44, i.e., the presence of exactly 3 fields and their content and relative memory locations per the field definitions 64. This information enables the collector 32 to coherently extract all the individual records and to examine their contents per the field definitions 64. The collector 32 can then provide the metrics data to separate users, such as described above with reference to FIG. 2 for example.

FIG. 6 is a general, high-level flow diagram of a method of ensuring consistency of data being received by a producer from a collector via a shared memory at regular intervals. The method assumes use of a time-of-day (TOD) variable for storing a time value with a granularity at least as fine as an interval size of the intervals.

At 90, by the producer for new data to be written into the shared memory for a given interval, (1) the TOD variable is first set to an invalid value, (2) next the new data is written into the shared memory with a timestamp, and (3) next the TOD variable is set to a time value corresponding to the timestamp.

At 92, by the collector for the new data, (1) the TOD variable and timestamp are read from the shared memory and it is confirmed that the TOD variable has a valid time value and is equal to the timestamp. Assuming so, then (2) the new data is read from the shared memory, and the validity of the new data is confirmed by (a) reading the TOD variable again from the shared memory, and (b) confirming that the TOD variable has the same valid time value as in the reading step (1). Assuming this confirmation is successful, the new data is processed as new metrics data to be provided to separate users, such as described above with reference to FIG. 2 for example. If either confirmation step (1) or (2) fails, this signifies that the stored data is either stale or otherwise not usable, and thus the data is not processed as new metrics data for this interval or sampling instance.

Summary of Features

1. Decoupling

The self-descriptive memory layout makes it possible for a producer to add or update metrics counters independently, decoupling from a collector. This is a key value compared with existing solutions, where changes in producers generally require changes in the collector. This feature can be useful for example in a development setting, enabling a development team to quickly add new metrics support.

2. Language-Neutral

The producer and collector can be developed in different preferred languages. For example, a low-level producer could be creating shared memory regions in C, while a higher-level producer could use Java or Python to create shared memory regions. The collector could also be variably implemented. This feature can be advantageous for example in larger-scale development efforts, where different components are owned by different teams with different skillsets.

3. Scalable

The shared memory could be read simultaneously by different collectors without allocating additional system resources for each collector. This flexibility can be provided without the typical costs of Remote Procedure Call (RPC), for example, in which more collectors introduce more socket connections to the producers. This feature may be especially useful for embedded systems, where system resources are highly constrained.

4. Efficient

For each collection time, the metrics data may be copied only once and directly from the producers' memory.

5. Easily Integrated

The technique may require no additional third-party libraries. The producers only need to organize the metrics data in the required memory layout and update them regularly.

While various embodiments of the invention have been particularly shown and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the invention as defined by the appended claims.

Claims

1. A method of operating a data storage system to obtain metrics data describing performance and capacities of the data storage system and provide the metrics data to external metrics users, comprising: by a metrics data producer, storing metrics data in a data region of a shared memory, and populating a metadata region of the shared memory with metadata describing a structure of the data region; andby a metrics data collector, reading the metadata of the shared memory and using the metadata to tailor a manner of access of the data region by the metrics data collector for correctly accessing the stored metrics data.

2. The method of claim 1, wherein the metrics data is for a set of objects of a single type in the data storage system for which respective performance or capacity data is obtained, and the data region stores a plurality of same-structured multi-field records for respective ones of the objects, and wherein the metadata region includes a plurality of field definition areas each storing a respective field definition describing structure and content of a respective field of the records, each field definition being used by the metrics data collector to locate and interpret the respective field in each of the records.

3. The method of claim 2, wherein the field definitions each include at least a field name, field location, and field size for the respective field, the field name identifying contents of the field, the field location identifying a relative location of the field in each of the records, the field size identifying a size of the field.

4. The method of claim 2, wherein the metadata region further includes a general area containing data items describing general aspects of the metrics data common to the records for the set of objects, the data items including a fields number identifying the number of fields in each of the records, the fields number being used by the metrics data collector to parse a correct number of fields of each of the records.

5. The method of claim 4, wherein the data items of the general area further include an update type and an update interval, the update type indicating whether the metrics data is updated at regular intervals or otherwise, the update interval identifying an interval length of the regular intervals, the update type and update interval being used by the metrics data collector to control timing of accessing the metrics data to obtain all metrics data as it is updated by the metrics data producer.

6. The method of claim 4, wherein the data items of the general area further include a layout type identifying a high-level structure of the data region, the layout type being selected from an array type and a map type, the array type having the records arranged in a single area with no intra-record indirection, the map type having the records arranged across multiple areas with at least one intra-record indirection from one area to another.

7. The method of claim 1, wherein the metrics data producer and metrics data collector operate according to a protocol including: by the producer: when initially creating a new record in the data region and when reusing a record of the data region, (i) generating a data generation identifier and writing it into the record, (ii) initializing other record fields, (iii) initializing corresponding metadata in the metadata region, and (iv) setting a metadata generation identifier to equal the data generation identifier; andwhen removing a record from the data region, changing the data generation identifier to an invalid value, and changing the metadata generation identifier to the invalid value; andby the collector, for collecting a specific instance of the metrics data: copying the entirety of the metrics data from the shared memory to private memory of the collector; andscanning the processing the records of the metrics data by, for each record:(i) skipping further processing of a record when its metadata generation identifier is invalid and when its metadata generation identifier is not equal to its data generation identifier; and(ii) collecting record data for downstream use when the metadata generation identifier is valid and equal to the data generation identifier.

8. The method of claim 1, wherein the metrics data collector provides metrics data to the metrics users using one of a file and a database.

9. The method of claim 1, wherein the metrics data producer registers itself by storing registration data in a collector configuration readable by the metrics data collector, the registration enabling the metrics data collector to identify the metrics data producer and begin collecting the metrics data in subsequent operation.

10. The method of claim 1, wherein the metrics data is first metrics data of a first type, and the metrics data producer and metrics data collector are a first metrics data producer and first metrics data collector operating upon the first metrics data, and wherein the data storage system further includes a second metrics data producer and a second metrics data collector operating upon second metrics data of a second type, by: by the second metrics data producer, storing the second metrics data in a second data region of the shared memory, and populating a second metadata region of the shared memory second with second metadata describing a structure of the second data region; andby the second metrics data collector, reading the second metadata of the shared memory and using the second metadata to tailor a manner of access of the second data region by the second metrics data collector for correctly accessing the stored second metrics data.

11. A data storage system, comprising: interface circuitry for receiving storage commands from remote storage users;storage devices for storing user data; andprocessing circuitry configured to execute data storage operations using the storage devices in response to the storage commands, the processing circuitry including a metrics subsystem for gathering metrics data describing performance and capacities of the data storage system and for providing the metrics data to external metrics users, the metrics subsystem including one or more producers, one or more collectors, and a self-describing shared memory having a data region and a metadata region, the data region storing metrics data produced by the producers, the metadata region storing metadata describing structure of the data region and being used by the collectors to tailor their accesses to the data region to correctly access the stored metrics data.

12. The data storage system of claim 11, wherein the metrics data is for a set of objects of a single type in the data storage system for which respective performance or capacity data is obtained, and the data region stores a plurality of same-structured multi-field records for respective ones of the objects, and wherein the metadata region includes a plurality of field definition areas each storing a respective field definition describing structure and content of a respective field of the records, each field definition being used by the metrics data collector to locate and interpret the respective field in each of the records.

13. The data storage system of claim 12, wherein the field definitions each include at least a field name, field location, and field size for the respective field, the field name identifying contents of the field, the field location identifying a relative location of the field in each of the records, the field size identifying a size of the field.

14. The data storage system of claim 12, wherein the metadata region further includes a general area containing data items describing general aspects of the metrics data common to the records for the set of objects, the data items including a fields number identifying the number of fields in each of the records, the fields number being used by the metrics data collector to parse a correct number of fields of each of the records.

15. The data storage system of claim 14, wherein the data items of the general area further include an update type and an update interval, the update type indicating whether the metrics data is updated at regular intervals or otherwise, the update interval identifying an interval length of the regular intervals, the update type and update interval being used by the metrics data collector to control timing of accessing the metrics data to obtain all metrics data as it is updated by the metrics data producer.

16. The data storage system of claim 14, wherein the data items of the general area further include a layout type identifying a high-level structure of the data region, the layout type being selected from an array type and a map type, the array type having the records arranged in a single area with no intra-record indirection, the map type having the records arranged across multiple areas with at least one intra-record indirection from one area to another.

17. The data storage system of claim 11, wherein the metrics data producer and metrics data collector operate according to a protocol including: by the producer: when initially creating a new record in the data region and when reusing a record of the data region, (i) generating a data generation identifier and writing it into the record, (ii) initializing other record fields, (iii) initializing corresponding metadata in the metadata region, and (iv) setting a metadata generation identifier to equal the data generation identifier; andwhen removing a record from the data region, changing the data generation identifier to an invalid value, and changing the metadata generation identifier to the invalid value; andby the collector, for collecting a specific instance of the metrics data: copying the entirety of the metrics data from the shared memory to private memory of the collector; andscanning the processing the records of the metrics data by, for each record:(i) skipping further processing of a record when its metadata generation identifier is invalid and when its metadata generation identifier is not equal to its data generation identifier; and(ii) collecting record data for downstream use when the metadata generation identifier is valid and equal to the data generation identifier.

18. The data storage system of claim 11, wherein the metrics data collector provides metrics data to the metrics users using one of a file and a database.

19. The data storage system of claim 11, wherein the metrics data producer registers itself by storing registration data in a collector configuration readable by the metrics data collector, the registration enabling the metrics data collector to identify the metrics data producer and begin collecting the metrics data in subsequent operation.

20. The data storage system of claim 11, wherein the metrics data is first metrics data of a first type, and the metrics data producer and metrics data collector are a first metrics data producer and first metrics data collector operating upon the first metrics data, and wherein the data storage system further includes a second metrics data producer and a second metrics data collector operating upon second metrics data of a second type, by: by the second metrics data producer, storing the second metrics data in a second data region of the shared memory, and populating a second metadata region of the shared memory second with second metadata describing a structure of the second data region; andby the second metrics data collector, reading the second metadata of the shared memory and using the second metadata to tailor a manner of access of the second data region by the second metrics data collector for correctly accessing the stored second metrics data.