[
https://issues.apache.org/jira/browse/YARN-8473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16531852#comment-16531852
]
Jason Lowe commented on YARN-8473:
----------------------------------
bq. In which case, container can come to this case?
It comes to this case whenever a container tries to launch when the application
has been asked to tear down, just as described in this JIRA's description. For
example, consider an AM that continues to launch containers even after the RM
has decided to kill the application. There's a window where the RM has told an
NM that the application should be killed before the AM has actually been
killed. During that window the NM can receive container launch events after
the ApplicationImpl has transitioned beyond the RUNNING state.
I think this could also theoretically happen with an unmanaged AM that
continues to launch containers even after unregistering with the RM.
I'll update the patch to add a warning message when we decide to kill a
container because the app is beyond the RUNNING state.
> Containers being launched as app tears down can leave containers in NEW state
> -----------------------------------------------------------------------------
>
> Key: YARN-8473
> URL: https://issues.apache.org/jira/browse/YARN-8473
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Affects Versions: 2.8.4
> Reporter: Jason Lowe
> Assignee: Jason Lowe
> Priority: Major
> Attachments: YARN-8473.001.patch, YARN-8473.002.patch
>
>
> I saw a case where containers were stuck on a nodemanager in the NEW state
> because they tried to launch just as an application was tearing down. The
> container sent an INIT_CONTAINER event to the ApplicationImpl which then
> executed an invalid transition since that event is not handled/expected when
> the application is in the process of tearing down.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
