[
https://issues.apache.org/jira/browse/YARN-1725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13900551#comment-13900551
]
Bikas Saha commented on YARN-1725:
----------------------------------
The only way to do it on AMRMClient would be to know which user request this
container would match and then submit that user request again to the RM. There
is no general way to do that correctly. I guess the problem is similar on the
RM side since once its decremented the *, rack and node counters for requests,
it can undo the * counter but does not know what to undo on the rack and node
counters.
> RM should provide an easier way for the app to reject a bad allocation
> ----------------------------------------------------------------------
>
> Key: YARN-1725
> URL: https://issues.apache.org/jira/browse/YARN-1725
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Bikas Saha
>
> Currently, if the app gets a bad allocation then it can release the
> container. However, the app now needs to request those resources again or
> else the RM will not give it a new container in lieu of the one just
> rejected. This makes the app writers life hard.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
