Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Disable shard idle with segment replication. #4118

Merged
merged 1 commit into from
Aug 4, 2022

Conversation

mch2
Copy link
Member

@mch2 mch2 commented Aug 3, 2022

Signed-off-by: Marc Handalian handalm@amazon.com

Description

This change disables shard idle when segment replication is enabled.
Primary shards will only push out new segments on refresh, we do not want to block this based on search behavior.
Replicas will only refresh on an externally provided SegmentInfos, so we do not want search requests to hang waiting for a refresh.

To achieve this - on replicas NRTReplicationEngine will never return that a refresh is needed, all refreshes will originate from within the engine itself. This also avoids the step of reading and comparing isCurrent on the latest reader, which for segrep will use StandardDirectoryReader without a writer and performs an IO operation to read the on-disk SegmentInfos. For Primary shards, the index setting is read during scheduleRefresh that controls if a refresh is pending.

This PR also introduces a method in IndexShardTestCase to improve the quality of segrep unit level tests by actually performing copy events.

Issues Resolved

closes #3834

Check List

  • New functionality includes testing.
    • All tests pass
  • New functionality has been documented.
    • New functionality has javadoc added
  • Commits are signed per the DCO using --signoff

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

@mch2 mch2 requested review from a team and reta as code owners August 3, 2022 22:05
This change disables shard idle when segment replication is enabled.
Primary shards will only push out new segments on refresh, we do not want to block this based on search behavior.
Replicas will only refresh on an externally provided SegmentInfos, so we do not want search requests to hang waiting for a refresh.

Signed-off-by: Marc Handalian <handalm@amazon.com>
@github-actions

This comment was marked as outdated.

@github-actions
Copy link
Contributor

github-actions bot commented Aug 3, 2022

Gradle Check (Jenkins) Run Completed with:

@codecov-commenter
Copy link

Codecov Report

Merging #4118 (5d353e8) into main (6f43dbc) will increase coverage by 0.02%.
The diff coverage is 83.82%.

@@             Coverage Diff              @@
##               main    #4118      +/-   ##
============================================
+ Coverage     70.61%   70.63%   +0.02%     
- Complexity    56983    57083     +100     
============================================
  Files          4595     4595              
  Lines        274234   274299      +65     
  Branches      40181    40185       +4     
============================================
+ Hits         193638   193749     +111     
+ Misses        64342    64303      -39     
+ Partials      16254    16247       -7     
Impacted Files Coverage Δ
...pensearch/common/settings/IndexScopedSettings.java 100.00% <ø> (ø)
...in/java/org/opensearch/index/shard/IndexShard.java 68.88% <0.00%> (-0.18%) ⬇️
.../java/org/opensearch/snapshots/RestoreService.java 55.37% <0.00%> (ø)
...org/opensearch/index/shard/IndexShardTestCase.java 93.78% <85.00%> (-1.62%) ⬇️
...org/opensearch/cluster/metadata/IndexMetadata.java 83.85% <100.00%> (ø)
.../main/java/org/opensearch/index/IndexSettings.java 86.35% <100.00%> (ø)
.../opensearch/index/engine/NRTReplicationEngine.java 74.78% <100.00%> (+2.75%) ⬆️
...ation/OpenSearchIndexLevelReplicationTestCase.java 90.22% <100.00%> (+0.06%) ⬆️
...a/org/opensearch/client/cluster/SniffModeInfo.java 0.00% <0.00%> (-58.83%) ⬇️
.../java/org/opensearch/node/NodeClosedException.java 50.00% <0.00%> (-50.00%) ⬇️
... and 485 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

Copy link
Collaborator

@Bukhtawar Bukhtawar left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should add this to the documentation to call out refresh optimisations for idle shards will no longer be honoured with segrep.

@mch2 mch2 merged commit d09c285 into opensearch-project:main Aug 4, 2022
Rishikesh1159 pushed a commit to Rishikesh1159/OpenSearch that referenced this pull request Aug 17, 2022
This change disables shard idle when segment replication is enabled.
Primary shards will only push out new segments on refresh, we do not want to block this based on search behavior.
Replicas will only refresh on an externally provided SegmentInfos, so we do not want search requests to hang waiting for a refresh.

Signed-off-by: Marc Handalian <handalm@amazon.com>
Rishikesh1159 added a commit that referenced this pull request Aug 17, 2022
…aining segment replication changes (#4243)

* [Segment Replication] Checkpoint Replay on Replica Shard (#3658)

* Adding Latest Recevied checkpoint, replay checkpoint logic along with tests

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* [Segment Replication] Wire up segment replication with peer recovery and add ITs. (#3743)

* Add null check when computing max segment version.

With segment replication enabled it is possible Lucene does not set the SegmentInfos
min segment version, leaving the default value as null.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Update peer recovery to set the translogUUID of replicas to the UUID generated on the primary.

This change updates the UUID when the translog is created to the value stored in the passed segment userdata.
This is to ensure during failover scenarios that the replica can be promoted and not have a uuid mismatch with the value stored in user data.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Wire up Segment Replication under the feature flag.

This PR wires up segment replication and adds some initial integration tests.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Add test to ensure replicas use primary translog uuid with segrep.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Update SegmentReplicationIT to assert previous commit points are valid and SegmentInfos can be built.
Fix nitpicks in PR feedback.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Fix test with Assert.fail to include a message.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Disable shard idle with segment replication. (#4118)

This change disables shard idle when segment replication is enabled.
Primary shards will only push out new segments on refresh, we do not want to block this based on search behavior.
Replicas will only refresh on an externally provided SegmentInfos, so we do not want search requests to hang waiting for a refresh.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Fix isAheadOf logic for ReplicationCheckpoint comparison (#4112)

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* Fix waitUntil refresh policy for segrep enabled indices. (#4137)

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Add IndexShard#getLatestReplicationCheckpoint behind segrep enable feature flag (#4163)

* Add IndexShard#getLatestReplicationCheckpoint behind segrep enable feature flag

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* Address review comment. Move tests to SegmentReplicationIndexShardTests

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* Add segrep enbaled index settings in TargetServiceTests, SourceHandlerTests

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* [Segment Replication] Fix OngoingSegmentReplications to key by allocation ID instead of DiscoveryNode. (#4182)

* Fix OngoingSegmentReplications to key by allocation ID instead of DiscoveryNode.

This change fixes segrep to work with multiple shards per node by keying ongoing replications on
allocation ID.  This also updates cancel methods to ensure state is properly cleared on shard cancel.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Clean up cancel methods.

Signed-off-by: Marc Handalian <handalm@amazon.com>

Signed-off-by: Marc Handalian <handalm@amazon.com>

* [Bug] [Segment Replication] Update store metadata recovery diff logic to ignore missing files causing exception (#4185)

* Update Store for segment replication dif

Signed-off-by: Poojita Raj <poojiraj@amazon.com>

Signed-off-by: Poojita Raj <poojiraj@amazon.com>

* Update recoveryDiff logic to ingore missing files causing exception on replica during copy

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* Address review comments

Signed-off-by: Suraj Singh <surajrider@gmail.com>

Signed-off-by: Poojita Raj <poojiraj@amazon.com>
Signed-off-by: Suraj Singh <surajrider@gmail.com>
Co-authored-by: Poojita Raj <poojiraj@amazon.com>

* [Segment Replication] Adding PrimaryMode check before publishing checkpoint and processing a received checkpoint. (#4157)

* Adding PrimaryMode check before publishing checkpoint.

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Applying spotless check

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Moving segrep specific tests to SegmentReplicationIndexShardTests.

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Adding logic and tests for rejecting checkpoints if shard is in PrimaryMode.

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Applying ./gradlew :server:spotlessApply.

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Applying ./gradlew :server:spotlessApply

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Changing log level to warn in shouldProcessCheckpoint() of IndexShard.java class.

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* Removing unnecessary lazy logging in shouldProcessCheckpoint().

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>

* [Segment Replication] Wait for documents to replicate to replica shards (#4236)

* [Segment Replication] Add thread sleep to account for replica lag in delete operations test

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* Address review comments, assertBusy on doc count rather than sleep

Signed-off-by: Suraj Singh <surajrider@gmail.com>

Signed-off-by: Suraj Singh <surajrider@gmail.com>

* Segment Replication - Remove unnecessary call to markAllocationIdAsInSync. (#4224)

This PR Removes an unnecessary call to markAllocationIdAsInSync on the primary shard when replication events complete.
Recovery will manage this initial call.

Signed-off-by: Marc Handalian <handalm@amazon.com>

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Segment Replication - Add additional unit tests for update & delete ops. (#4237)

* Segment Replication - Add additional unit tests for update & delete operations.

Signed-off-by: Marc Handalian <handalm@amazon.com>

* Fix spotless.

Signed-off-by: Marc Handalian <handalm@amazon.com>

Signed-off-by: Marc Handalian <handalm@amazon.com>

Signed-off-by: Rishikesh1159 <rishireddy1159@gmail.com>
Signed-off-by: Marc Handalian <handalm@amazon.com>
Signed-off-by: Suraj Singh <surajrider@gmail.com>
Signed-off-by: Poojita Raj <poojiraj@amazon.com>
Co-authored-by: Marc Handalian <handalm@amazon.com>
Co-authored-by: Suraj Singh <surajrider@gmail.com>
Co-authored-by: Poojita Raj <poojiraj@amazon.com>
@mch2 mch2 deleted the 3834 branch January 11, 2023 18:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

[BUG] [Segment replication] Search requests to index hang indefinitely
4 participants