[
https://issues.apache.org/jira/browse/FLINK-33121?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17767214#comment-17767214
]
Chesnay Schepler commented on FLINK-33121:
------------------------------------------
Had an offline chat; our suspicion is that something is calling the failure
handling logic of the scheduler without running this call in the main thread.
We'll try to confirm this theory.
> Failed precondition in JobExceptionsHandler due to concurrent global failures
> -----------------------------------------------------------------------------
>
> Key: FLINK-33121
> URL: https://issues.apache.org/jira/browse/FLINK-33121
> Project: Flink
> Issue Type: Bug
> Reporter: Panagiotis Garefalakis
> Priority: Major
>
> {{JobExceptionsHandler#createRootExceptionInfo}} *only* allows concurrent
> exceptions that are local failures *--* otherwise throws an assertion as part
> of {{{}asserLocalExceptionInfo{}}}.
> However, there are rare cases where multiple concurrent global failures are
> triggered and added to the failureCollection, before transitioning the job
> state to Failed e.g., through {{StateWithExecutionGraph#handleGlobalFailure}}
> of the AdaptiveScheduler.
> In this case the last added will be the root and the next one will trigger
> the assertion
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
