[
https://issues.apache.org/jira/browse/FLINK-40553?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18112263#comment-18112263
]
Matthias Pohl commented on FLINK-40553:
---------------------------------------
Reiterating over the problem space let me re-evaluate the issue: Corrupted or
missing Flink-owned data points to some unknown cause that should be
investigated and mitigated rather then narrowed down to job scope because it
could have caused inconsistencies in other places. Keeping the cluster-wide
failure makes sense in that case.
Closing the ticket(s) as {{Won't Do}}.
> Corrupted job-local HA data causes cluster-wide failover
> --------------------------------------------------------
>
> Key: FLINK-40553
> URL: https://issues.apache.org/jira/browse/FLINK-40553
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
> Affects Versions: 2.3.0, 2.2.1, 2.4.0
> Reporter: Matthias Pohl
> Assignee: Matthias Pohl
> Priority: Major
>
> There are two code paths that can cause a cluster-wide failover if the HA
> data for a single job is corrupted in some way.
> One example was where the ExecutionPlan files got deleted causing a
> {{FileNotFoundException}} when recovering the job. That triggered a
> JobManager failover.
> A similar issue can appear when reading the JobResultStore entry files. The
> parsing happens in a single command for all files. One file failing to be
> retrieved results in the entire JobManager to fail over.
> The two issues are covered in separate subtasks and hidden behind a
> configuration parameter.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
