Yi Zhang created FLINK-37697:
--------------------------------
Summary: ExecutionGraphInfo may not be archived when cluster
shutdown in application mode
Key: FLINK-37697
URL: https://issues.apache.org/jira/browse/FLINK-37697
Project: Flink
Issue Type: Bug
Components: Runtime / Coordination
Reporter: Yi Zhang
In application mode, the ExecutionGraphInfo archiving may not be completed
before the cluster is shut down, which prevents users from directly obtaining
the final job results and requires checking the logs instead. FLINK-24491
raised the same issue, and the solution was to wait for the archiving to finish
before performing job cleanup. The MiniDispatcher would then wait for job
cleanup to finish before completing the shutdown future, thereby resolving the
issue in per-job mode. However, this problem still exists in application mode.
To resolve this issue in application mode, the Dispatcher needs to wait for the
archiving to finish before completing the shutdown future. FLINK-26772 raised a
related issue, and considering now the Dispatcher already waits for the
archiving to complete before performing job cleanup, further modifications can
be made to ensure that the shutdown future is completed only after the job
cleanup are finished, thus also fixing FLINK-26772.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
