[jira] [Resolved] (FLINK-40446) UnalignedCheckpointRescaleWithMixedExchangesITCase is unstable: @TempDir cleanup races with post-cancel async checkpoint

["Martijn Visser (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/FLINK-40446?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Martijn Visser resolved FLINK-40446.
------------------------------------
    Resolution: Fixed

> UnalignedCheckpointRescaleWithMixedExchangesITCase is unstable: @TempDir cleanup races with post-cancel async checkpoint
> ------------------------------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-40446
>                 URL: https://issues.apache.org/jira/browse/FLINK-40446
>             Project: Flink
>          Issue Type: Bug
>    Affects Versions: 2.3.0, 2.2.1, 1.20.5
>            Reporter: Rui Fan
>            Assignee: Rui Fan
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 2.2.2, 1.20.6, 2.3.1, 2.4.0
>
>
> UnalignedCheckpointRescaleWithMixedExchangesITCase.testRescaleFromUnalignedCheckpoint is occasionally unstable. The test body passes, but teardown fails:
> {code:java}
>   java.io.IOException: Failed to delete temp directory /tmp/junit-...
>     Suppressed: java.nio.file.DirectoryNotEmptyException: /tmp/junit-...{code}
> Root cause: the test ends each job with `jobClient.cancel().get()`. cancel() is
> asynchronous, so .get() only means cancellation was acknowledged, not that the job
> reached a terminal state. The test then returns and JUnit deletes the @TempDir (which
> also serves as the checkpoint dir) while the source subtasks' async snapshot of the
> just-completed checkpoint is still writing SourceReaderState into the file-merging
> "taskowned" directory. The post-order delete walk then hits DirectoryNotEmptyException.
> It only reproduces with file-merging enabled (execution.checkpointing.file-merging.enabled
> = true), which is why it shows up on a fraction of runs.
> Fix: wait for the job to reach the CANCELED terminal state before proceeding, instead of
> relying on cancel().get() alone. (CommonTestUtils.terminateJob now waits for CANCELED,
> and the UC rescale ITCases go through it.)



--
This message was sent by Atlassian Jira
(v8.20.10#820010)