Repository navigation
[CELEBORN-2483] Fix flaky RemoteShuffleMasterSuiteJ shuffleCount assertions racing with the initial app heartbeat - #3875
Open
afterincomparableyum wants to merge 1 commit into
Conversation
…rtions racing with the initial app heartbeat ApplicationHeartbeater schedules its first heartbeat with an initial delay of 0, and each heartbeat drains LifecycleManager#shuffleCount and the fallback counts via sumThenReset. The LifecycleManager is created in registerJob, so that first heartbeat runs concurrently with the test's registerPartitionWithProducer calls and can reset the counter between an increment and its assertion (expected:<1> but was:<0>, expected:<3> but was:<1>). CELEBORN-2431 raised the heartbeat interval to 1h, which only removed the periodic heartbeats, not the initial one. Add the internal conf celeborn.test.client.application.heartbeat.enabled (default true) and disable the app heartbeat in RemoteShuffleMasterSuiteJ, which has no master to heartbeat to. The production initial delay is unchanged.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this pull request?
celeborn.test.client.application.heartbeat.enabled(defaulttrue). When it isfalse,ApplicationHeartbeaterdoes not schedule the app heartbeat.RemoteShuffleMasterSuiteJ, replacing theheartbeatInterval=1hworkaround from CELEBORN-2431.Why are the changes needed?
RemoteShuffleMasterSuiteJ.testRegisterPartitionWithProduceris still flaky after CELEBORN-2431:ApplicationHeartbeaterschedules its first heartbeat with an initial delay of 0. Each heartbeat drainsLifecycleManager#shuffleCountand the fallback counts viasumThenReset. TheLifecycleManageris created inregisterJob, so the first heartbeat runs at the same time as the test'sregisterPartitionWithProducercalls and can reset the counter between an increment and its assertion. Raising the interval to 1h only removed the periodic heartbeats, not the initial one. The suite has no master, so no test in it needs the heartbeat. The production initial delay is unchanged.Example failures: flink1 (11, 1.18), flink2 (17, 2.2).
Does this PR resolve a correctness bug?
Does this PR introduce any user-facing change?
How was this patch tested?
RemoteShuffleMasterSuiteJV120: 7 tests passed. TheAskSync HeartbeatFromApplication failederrors, which appeared once per test in CI, should no longer appear.