[
https://issues.apache.org/jira/browse/FLINK-40655?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18115376#comment-18115376
]
Qiu Yanjun commented on FLINK-40655:
------------------------------------
I investigated the flakiness in
ExecutionVertexTest.testResetForNewExecutionReleasesPartitions on the latest
master. The test calls execution.markFinished() and then immediately asserts
that the partition-release future is not completed. Partition release can be
triggered asynchronously by the scheduler before resetForNewExecution(), so
this intermediate timing assertion is race-prone. The final assertion already
verifies that the expected ResultPartitionID is released after
resetForNewExecution().
I reproduced the issue context locally and prepared a minimal test-only fix by
removing the unstable intermediate isNotDone() assertion while retaining the
final partition-ID verification. After the change, the targeted test passed in
10 consecutive runs with:
./mvnw -pl flink-runtime
-Dtest=ExecutionVertexTest#testResetForNewExecutionReleasesPartitions test
The change is currently local and uncommitted. I would like to work on this
issue; could a committer please assign it to me?
> ExecutionVertexTest.testResetForNewExecutionReleasesPartitions is unstable
> --------------------------------------------------------------------------
>
> Key: FLINK-40655
> URL: https://issues.apache.org/jira/browse/FLINK-40655
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination, Tests
> Reporter: Martijn Visser
> Priority: Major
>
> {code}
> [ERROR]
> org.apache.flink.runtime.executiongraph.ExecutionVertexTest.testResetForNewExecutionReleasesPartitions
> -- Time elapsed: 0.070 s <<< FAILURE!
> java.lang.AssertionError:
> Expecting
> <CompletableFuture[Completed:
> [c7b11773d4337c893ef2d96dab905a8a#0@206983cc7f2453fa4bc230fce7b00509_50dd1e4e2f60188233f1f72f88eecc93_0_0]]>
> not to be done.
> {code}
> https://github.com/apache/flink/actions/runs/34732018337
> https://github.com/apache/flink/actions/runs/34775554131
> Both on 2026-09-13 on master in the core module, one nightly and one push run
> on a different commit.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
