[ 
https://issues.apache.org/jira/browse/YARN-433?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14644877#comment-14644877
 ] 

zhihai xu commented on YARN-433:
--------------------------------

Thanks for the new patch [~xgong]!
There is a check style issue at 
./hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestRMNodeTransitions.java:221:
 
But looking at the change, this check style issue is already there without the 
change. So it is not related to your patch.
+1 for the latest patch.

> When RM is catching up with node updates then it should not expire acquired 
> containers
> --------------------------------------------------------------------------------------
>
>                 Key: YARN-433
>                 URL: https://issues.apache.org/jira/browse/YARN-433
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Bikas Saha
>            Assignee: Xuan Gong
>         Attachments: YARN-433.1.patch, YARN-433.2.patch, YARN-433.3.patch, 
> YARN-433.4.patch
>
>
> RM expires containers that are not launched within some time of being 
> allocated. The default is 10mins. When an RM is not keeping up with node 
> updates then it may not be aware of new launched containers. If the expire 
> thread fires for such containers then the RM can expire them even though they 
> may have launched.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to