[
https://issues.apache.org/jira/browse/FLINK-29308?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17611349#comment-17611349
]
Zhu Zhu commented on FLINK-29308:
---------------------------------
It's possible that the released resource cannot fulfill the new slot request.
Therefore Flink use this mechanism to fail fast, otherwise users may find a job
pending for quite sometime with no progress. This may be helpful when the
resource cluster is problematic or there is mis-configuration in the job(wrong
resource spec, wrong resource queue, etc).
One potential problem to ignore the NoResourceAvailableException is that a job
may wait indefinitely until it can obtain a required slot. You can do it if
it's acceptable in your case.
> NoResourceAvailableException fails the batch job
> ------------------------------------------------
>
> Key: FLINK-29308
> URL: https://issues.apache.org/jira/browse/FLINK-29308
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Reporter: Aitozi
> Priority: Major
>
> When running batch job configured with the following restart strategy
> {code:java}
> restart-strategy: fixed-delay
> restart-strategy.fixed-delay.delay: 15 s
> restart-strategy.fixed-delay.attempts: 10 {code}
> If the cluster resource is not enough to run the single stage, it can run
> partial of the stage, but it still will fail after the 10 times
> {{{}NoResourceAvailableException{}}}. IMO, for batch job the
> {{NoResourceAvailableException}} do not necessary to trigger the job to fail.
> Or at least this failure reason is not suitable to share the same restart
> strategy with other failure reasons
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
