[
https://issues.apache.org/jira/browse/YARN-4831?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15197895#comment-15197895
]
Siqi Li commented on YARN-4831:
-------------------------------
When NM does a stateful restart, the ContainerManagerImpl will try to recover
applications, and containers, and then send out ApplicationFinishEvent to apps
that in appsState.getFinishedApplications().
The ApplicationFinishEvent could result in newly recovered containers to
transit from NEW to DONE with a KillOnNewTransition.
We could add an additional check in KillOnNewTransition to avoid killing
completed containers.
> Recovered containers will be killed after NM stateful restart
> --------------------------------------------------------------
>
> Key: YARN-4831
> URL: https://issues.apache.org/jira/browse/YARN-4831
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Siqi Li
>
> {code}
> 2016-03-04 19:43:48,130 INFO
> org.apache.hadoop.yarn.server.nodemanager.containermanager.container.Container:
> Container container_1456335621285_0040_01_000066 transitioned from NEW to
> DONE
> 2016-03-04 19:43:48,130 INFO
> org.apache.hadoop.yarn.server.nodemanager.NMAuditLogger: USER=henkins-service
> OPERATION=Container Finished - Killed TARGET=ContainerImpl
> RESULT=SUCCESS APPID=application_1456335621285_0040
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
