[
https://issues.apache.org/jira/browse/FLINK-37588?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18016353#comment-18016353
]
chaoran.su commented on FLINK-37588:
------------------------------------
I think this issue need more information, like which flink version is using.
When flink version lower than 1.14 and we manually change task state to finish
/ fail, the cluster will exit immediately, since config
execution.shutdown-on-application-finish is not exist. The config
jm-deployment.shutdown-ttl will not take effect.
When flink version higher than 1.14, flink operator always set
execution.shutdown-on-application-finish to false, so cluster will remain after
job is finished / failed. Application clusters should not exit immediately and
we can observe them. The config jm-deployment.shutdown-ttl will take effect,
since the status and updateTime are changed in the same time.
I think the version of flink operator higher than 1.7, which support flink
higher than 1.15, the config jm-deployment.shutdown-ttl will take effect.
> Flink Kubernetes Operator Parameter `jm-deployment.shutdown-ttl` does not
> take effect
> -------------------------------------------------------------------------------------
>
> Key: FLINK-37588
> URL: https://issues.apache.org/jira/browse/FLINK-37588
> Project: Flink
> Issue Type: Bug
> Components: Kubernetes Operator
> Reporter: yuanfenghu
> Priority: Critical
>
> When I configured jm-deployment.shutdown-ttl, it did not take effect because
> its judgment was based on the JobStatus#updateTime, but updateTime will only
> change if state changes are observed in two previous observations, and
> observations are triggered periodically. Therefore, if the time difference
> between the last state change and the current time after we stop the task
> exceeds jm-deployment.shutdown-ttl, the cluster will be destroyed
> immediately. This is not expected, so we should also update updateTime when
> the task state is manually triggered to change to FINISHED.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
