Repository analysis API

edit

Analyzes a repository, reporting its performance characteristics and any incorrect behaviour found.

resp = client.perform_request(
    "POST",
    "/_snapshot/my_repository/_analyze",
    params={
        "blob_count": "10",
        "max_blob_size": "1mb",
        "timeout": "120s"
    },
)
print(resp)
response = client.snapshot.repository_analyze(
  repository: 'my_repository',
  blob_count: 10,
  max_blob_size: '1mb',
  timeout: '120s'
)
puts response
const response = await client.transport.request({
  method: "POST",
  path: "/_snapshot/my_repository/_analyze",
  querystring: {
    blob_count: "10",
    max_blob_size: "1mb",
    timeout: "120s",
  },
});
console.log(response);
POST /_snapshot/my_repository/_analyze?blob_count=10&max_blob_size=1mb&timeout=120s

Request

edit

POST /_snapshot/<repository>/_analyze

Prerequisites

edit

Description

edit

There are a large number of third-party storage systems available, not all of which are suitable for use as a snapshot repository by Elasticsearch. Some storage systems behave incorrectly, or perform poorly, especially when accessed concurrently by multiple clients as the nodes of an Elasticsearch cluster do.

The Repository analysis API performs a collection of read and write operations on your repository which are designed to detect incorrect behaviour and to measure the performance characteristics of your storage system.

The default values for the parameters to this API are deliberately low to reduce the impact of running an analysis inadvertently and to provide a sensible starting point for your investigations. Run your first analysis with the default parameter values to check for simple problems. If successful, run a sequence of increasingly large analyses until you encounter a failure or you reach a blob_count of at least 2000, a max_blob_size of at least 2gb, a max_total_data_size of at least 1tb, and a register_operation_count of at least 100. Always specify a generous timeout, possibly 1h or longer, to allow time for each analysis to run to completion. Perform the analyses using a multi-node cluster of a similar size to your production cluster so that it can detect any problems that only arise when the repository is accessed by many nodes at once.

If the analysis fails then Elasticsearch detected that your repository behaved unexpectedly. This usually means you are using a third-party storage system with an incorrect or incompatible implementation of the API it claims to support. If so, this storage system is not suitable for use as a snapshot repository. You will need to work with the supplier of your storage system to address the incompatibilities that Elasticsearch detects. See Self-managed repository types for more information.

If the analysis is successful this API returns details of the testing process, optionally including how long each operation took. You can use this information to determine the performance of your storage system. If any operation fails or returns an incorrect result, this API returns an error. If the API returns an error then it may not have removed all the data it wrote to the repository. The error will indicate the location of any leftover data, and this path is also recorded in the Elasticsearch logs. You should verify yourself that this location has been cleaned up correctly. If there is still leftover data at the specified location then you should manually remove it.

If the connection from your client to Elasticsearch is closed while the client is waiting for the result of the analysis then the test is cancelled. Some clients are configured to close their connection if no response is received within a certain timeout. An analysis takes a long time to complete so you may need to relax any such client-side timeouts. On cancellation the analysis attempts to clean up the data it was writing, but it may not be able to remove it all. The path to the leftover data is recorded in the Elasticsearch logs. You should verify yourself that this location has been cleaned up correctly. If there is still leftover data at the specified location then you should manually remove it.

If the analysis is successful then it detected no incorrect behaviour, but this does not mean that correct behaviour is guaranteed. The analysis attempts to detect common bugs but it certainly does not offer 100% coverage. Additionally, it does not test the following:

  • Your repository must perform durable writes. Once a blob has been written it must remain in place until it is deleted, even after a power loss or similar disaster.
  • Your repository must not suffer from silent data corruption. Once a blob has been written its contents must remain unchanged until it is deliberately modified or deleted.
  • Your repository must behave correctly even if connectivity from the cluster is disrupted. Reads and writes may fail in this case, but they must not return incorrect results.

An analysis writes a substantial amount of data to your repository and then reads it back again. This consumes bandwidth on the network between the cluster and the repository, and storage space and IO bandwidth on the repository itself. You must ensure this load does not affect other users of these systems. Analyses respect the repository settings max_snapshot_bytes_per_sec and max_restore_bytes_per_sec if available, and the cluster setting indices.recovery.max_bytes_per_sec which you can use to limit the bandwidth they consume.

This API is intended for exploratory use by humans. You should expect the request parameters and the response format to vary in future versions.

Different versions of Elasticsearch may perform different checks for repository compatibility, with newer versions typically being stricter than older ones. A storage system that passes repository analysis with one version of Elasticsearch may fail with a different version. This indicates it behaves incorrectly in ways that the former version did not detect. You must work with the supplier of your storage system to address the incompatibilities detected by the repository analysis API in any version of Elasticsearch.

This API may not work correctly in a mixed-version cluster.

Implementation details

edit

This section of documentation describes how the Repository analysis API works in this version of Elasticsearch, but you should expect the implementation to vary between versions. The request parameters and response format depend on details of the implementation so may also be different in newer versions.

The analysis comprises a number of blob-level tasks, as set by the blob_count parameter, and a number of compare-and-exchange operations on linearizable registers, as set by the register_operation_count parameter. These tasks are distributed over the data and master-eligible nodes in the cluster for execution.

For most blob-level tasks, the executing node first writes a blob to the repository, and then instructs some of the other nodes in the cluster to attempt to read the data it just wrote. The size of the blob is chosen randomly, according to the max_blob_size and max_total_data_size parameters. If any of these reads fails then the repository does not implement the necessary read-after-write semantics that Elasticsearch requires.

For some blob-level tasks, the executing node will instruct some of its peers to attempt to read the data before the writing process completes. These reads are permitted to fail, but must not return partial data. If any read returns partial data then the repository does not implement the necessary atomicity semantics that Elasticsearch requires.

For some blob-level tasks, the executing node will overwrite the blob while its peers are reading it. In this case the data read may come from either the original or the overwritten blob, but the read operation must not return partial data or a mix of data from the two blobs. If any of these reads returns partial data or a mix of the two blobs then the repository does not implement the necessary atomicity semantics that Elasticsearch requires for overwrites.

The executing node will use a variety of different methods to write the blob. For instance, where applicable, it will use both single-part and multi-part uploads. Similarly, the reading nodes will use a variety of different methods to read the data back again. For instance they may read the entire blob from start to end, or may read only a subset of the data.

For some blob-level tasks, the executing node will abort the write before it is complete. In this case it still instructs some of the other nodes in the cluster to attempt to read the blob, but all of these reads must fail to find the blob.

Linearizable registers are special blobs that Elasticsearch manipulates using an atomic compare-and-exchange operation. This operation ensures correct and strongly-consistent behavior even when the blob is accessed by multiple nodes at the same time. The detailed implementation of the compare-and-exchange operation on linearizable registers varies by repository type. Repository analysis verifies that that uncontended compare-and-exchange operations on a linearizable register blob always succeed. Repository analysis also verifies that contended operations either succeed or report the contention but do not return incorrect results. If an operation fails due to contention, Elasticsearch retries the operation until it succeeds. Most of the compare-and-exchange operations performed by repository analysis atomically increment a counter which is represented as an 8-byte blob. Some operations also verify the behavior on small blobs with sizes other than 8 bytes.

Path parameters

edit
<repository>
(Required, string) Name of the snapshot repository to test.

Query parameters

edit
blob_count
(Optional, integer) The total number of blobs to write to the repository during the test. Defaults to 100. For realistic experiments you should set this to at least 2000.
max_blob_size
(Optional, size units) The maximum size of a blob to be written during the test. Defaults to 10mb. For realistic experiments you should set this to at least 2gb.
max_total_data_size
(Optional, size units) An upper limit on the total size of all the blobs written during the test. Defaults to 1gb. For realistic experiments you should set this to at least 1tb.
register_operation_count
(Optional, integer) The minimum number of linearizable register operations to perform in total. Defaults to 10. For realistic experiments you should set this to at least 100.
timeout
(Optional, time units) Specifies the period of time to wait for the test to complete. If no response is received before the timeout expires, the test is cancelled and returns an error. Defaults to 30s.

Advanced query parameters

edit

The following parameters allow additional control over the analysis, but you will usually not need to adjust them.

concurrency
(Optional, integer) The number of write operations to perform concurrently. Defaults to 10.
read_node_count
(Optional, integer) The number of nodes on which to perform a read operation after writing each blob. Defaults to 10.
early_read_node_count
(Optional, integer) The number of nodes on which to perform an early read operation while writing each blob. Defaults to 2. Early read operations are only rarely performed.
rare_action_probability
(Optional, double) The probability of performing a rare action (an early read, an overwrite, or an aborted write) on each blob. Defaults to 0.02.
seed
(Optional, integer) The seed for the pseudo-random number generator used to generate the list of operations performed during the test. To repeat the same set of operations in multiple experiments, use the same seed in each experiment. Note that the operations are performed concurrently so may not always happen in the same order on each run.
detailed
(Optional, boolean) Whether to return detailed results, including timing information for every operation performed during the analysis. Defaults to false, meaning to return only a summary of the analysis.
rarely_abort_writes
(Optional, boolean) Whether to rarely abort some write requests. Defaults to true.

Response body

edit

The response exposes implementation details of the analysis which may change from version to version. The response body format is therefore not considered stable and may be different in newer versions.

coordinating_node

(object) Identifies the node which coordinated the analysis and performed the final cleanup.

Properties of coordinating_node
id
(string) The id of the coordinating node.
name
(string) The name of the coordinating node
repository
(string) The name of the repository that was the subject of the analysis.
blob_count
(integer) The number of blobs written to the repository during the test, equal to the ?blob_count request parameter.
concurrency
(integer) The number of write operations performed concurrently during the test, equal to the ?concurrency request parameter.
read_node_count
(integer) The limit on the number of nodes on which read operations were performed after writing each blob, equal to the ?read_node_count request parameter.
early_read_node_count
(integer) The limit on the number of nodes on which early read operations were performed after writing each blob, equal to the ?early_read_node_count request parameter.
max_blob_size
(string) The limit on the size of a blob written during the test, equal to the ?max_blob_size parameter.
max_blob_size_bytes
(long) The limit, in bytes, on the size of a blob written during the test, equal to the ?max_blob_size parameter.
max_total_data_size
(string) The limit on the total size of all blob written during the test, equal to the ?max_total_data_size parameter.
max_total_data_size_bytes
(long) The limit, in bytes, on the total size of all blob written during the test, equal to the ?max_total_data_size parameter.
seed
(long) The seed for the pseudo-random number generator used to generate the operations used during the test. Equal to the ?seed request parameter if set.
rare_action_probability
(double) The probability of performing rare actions during the test. Equal to the ?rare_action_probability request parameter.
blob_path
(string) The path in the repository under which all the blobs were written during the test.
issues_detected
(list) A list of correctness issues detected, which will be empty if the API succeeded. Included to emphasize that a successful response does not guarantee correct behaviour in future.
summary

(object) A collection of statistics that summarise the results of the test.

Properties of summary
write

(object) A collection of statistics that summarise the results of the write operations in the test.

Properties of write
count
(integer) The number of write operations performed in the test.
total_size
(string) The total size of all the blobs written in the test.
total_size_bytes
(long) The total size of all the blobs written in the test, in bytes.
total_throttled
(string) The total time spent waiting due to the max_snapshot_bytes_per_sec throttle.
total_throttled_nanos
(long) The total time spent waiting due to the max_snapshot_bytes_per_sec throttle, in nanoseconds.
total_elapsed
(string) The total elapsed time spent on writing blobs in the test.
total_elapsed_nanos
(long) The total elapsed time spent on writing blobs in the test, in nanoseconds.
read

(object) A collection of statistics that summarise the results of the read operations in the test.

Properties of read
count
(integer) The number of read operations performed in the test.
total_size
(string) The total size of all the blobs or partial blobs read in the test.
total_size_bytes
(long) The total size of all the blobs or partial blobs read in the test, in bytes.
total_wait
(string) The total time spent waiting for the first byte of each read request to be received.
total_wait_nanos
(long) The total time spent waiting for the first byte of each read request to be received, in nanoseconds.
max_wait
(string) The maximum time spent waiting for the first byte of any read request to be received.
max_wait_nanos
(long) The maximum time spent waiting for the first byte of any read request to be received, in nanoseconds.
total_throttled
(string) The total time spent waiting due to the max_restore_bytes_per_sec or indices.recovery.max_bytes_per_sec throttles.
total_throttled_nanos
(long) The total time spent waiting due to the max_restore_bytes_per_sec or indices.recovery.max_bytes_per_sec throttles, in nanoseconds.
total_elapsed
(string) The total elapsed time spent on reading blobs in the test.
total_elapsed_nanos
(long) The total elapsed time spent on reading blobs in the test, in nanoseconds.
details

(array) A description of every read and write operation performed during the test. This is only returned if the ?detailed request parameter is set to true.

Properties of items within details
blob

(object) A description of the blob that was written and read.

Properties of blob
name
(string) The name of the blob.
size
(string) The size of the blob.
size_bytes
(long) The size of the blob in bytes.
read_start
(long) The position, in bytes, at which read operations started.
read_end
(long) The position, in bytes, at which read operations completed.
read_early
(boolean) Whether any read operations were started before the write operation completed.
overwritten
(boolean) Whether the blob was overwritten while the read operations were ongoing.
writer_node

(object) Identifies the node which wrote this blob and coordinated the read operations.

Properties of writer_node
id
(string) The id of the writer node.
name
(string) The name of the writer node
write_elapsed
(string) The elapsed time spent writing this blob.
write_elapsed_nanos
(long) The elapsed time spent writing this blob, in nanoseconds.
overwrite_elapsed
(string) The elapsed time spent overwriting this blob. Omitted if the blob was not overwritten.
overwrite_elapsed_nanos
(long) The elapsed time spent overwriting this blob, in nanoseconds. Omitted if the blob was not overwritten.
write_throttled
(string) The length of time spent waiting for the max_snapshot_bytes_per_sec (or indices.recovery.max_bytes_per_sec if the recovery settings for managed services are set) throttle while writing this blob.
write_throttled_nanos
(long) The length of time spent waiting for the max_snapshot_bytes_per_sec (or indices.recovery.max_bytes_per_sec if the recovery settings for managed services are set) throttle while writing this blob, in nanoseconds.
reads

(array) A description of every read operation performed on this blob.

Properties of items within reads
node

(object) Identifies the node which performed the read operation.

Properties of node
id
(string) The id of the reader node.
name
(string) The name of the reader node
before_write_complete
(boolean) Whether the read operation may have started before the write operation was complete. Omitted if false.
found
(boolean) Whether the blob was found by this read operation or not. May be false if the read was started before the write completed, or the write was aborted before completion.
first_byte_time
(string) The length of time waiting for the first byte of the read operation to be received. Omitted if the blob was not found.
first_byte_time_nanos
(long) The length of time waiting for the first byte of the read operation to be received, in nanoseconds. Omitted if the blob was not found.
elapsed
(string) The length of time spent reading this blob. Omitted if the blob was not found.
elapsed_nanos
(long) The length of time spent reading this blob, in nanoseconds. Omitted if the blob was not found.
throttled
(string) The length of time spent waiting due to the max_restore_bytes_per_sec or indices.recovery.max_bytes_per_sec throttles during the read of this blob. Omitted if the blob was not found.
throttled_nanos
(long) The length of time spent waiting due to the max_restore_bytes_per_sec or indices.recovery.max_bytes_per_sec throttles during the read of this blob, in nanoseconds. Omitted if the blob was not found.
listing_elapsed
(string) The time it took to retrieve a list of all the blobs in the container.
listing_elapsed_nanos
(long) The time it took to retrieve a list of all the blobs in the container, in nanoseconds.
delete_elapsed
(string) The time it took to delete all the blobs in the container.
delete_elapsed_nanos
(long) The time it took to delete all the blobs in the container, in nanoseconds.