You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Prabhu Joseph (JIRA)" <ji...@apache.org> on 2019/05/20 11:02:00 UTC
[jira] [Comment Edited] (YARN-8224) LogAggregation status TIME_OUT
for absent container misleading
[ https://issues.apache.org/jira/browse/YARN-8224?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16843863#comment-16843863 ]
Prabhu Joseph edited comment on YARN-8224 at 5/20/19 11:01 AM:
---------------------------------------------------------------
[~snemeth] Have missed to debug when the issue happened. But as per my understanding, {{RMAppImpl}} adds the {{NodeId}} into Map {{logAggregationStatus}} when container is Acquired in {{AppRunningOnNodeTransition}}. But when the container is Killed after Acquired, the NM looks does not have any information about this container. {{RMAppImpl#getLogAggregationReportsForApp}} will set TIME_OUT since the Node does not send any {{LogAggregationReport}} for this container.
I think this issue can happen by killing a acquired container or stopping NM after container is acquired.
was (Author: prabhu joseph):
[~snemeth] Have missed to debug when the issue happened. But as per my understanding, {{RMAppImpl}} adds the {{NodeId}} into Map {{logAggregationStatus}} when container is Acquired in AppRunningOnNodeTransition. But when the container is killed after Acquired, the NM looks does not have any information about this container. RMAppImpl#getLogAggregationReportsForApp will set TIME_OUT since the Node does not send any LogAggregationReport for this container.
I think this issue can happen by killing a acquired container or stopping NM after container is acquired.
> LogAggregation status TIME_OUT for absent container misleading
> --------------------------------------------------------------
>
> Key: YARN-8224
> URL: https://issues.apache.org/jira/browse/YARN-8224
> Project: Hadoop YARN
> Issue Type: Bug
> Components: log-aggregation
> Affects Versions: 2.7.3
> Reporter: Prabhu Joseph
> Assignee: Prabhu Joseph
> Priority: Major
>
> When a container is not launched on NM and it is absent, RM still tries to get the Log Aggregation Status and reports the status as TIME_OUT in RM UI.
> {code}
> 2018-04-26 12:47:38,403 WARN containermanager.ContainerManagerImpl (ContainerManagerImpl.java:handle(1070)) - Event EventType: KILL_CONTAINER sent to absent container container_e361_1524687599273_2110_01_000770
> 2018-04-26 12:49:31,743 WARN containermanager.ContainerManagerImpl (ContainerManagerImpl.java:handle(1086)) - Event EventType: FINISH_APPLICATION sent to absent application application_1524687599273_2110
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org