[
https://issues.apache.org/jira/browse/YARN-5117?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15299094#comment-15299094
]
Konstantinos Karanasos commented on YARN-5117:
----------------------------------------------
[~kasha], can you please give a look at this patch (it is super small)?
I changed the way I am checking for available CPU before starting a container.
Two questions in particular:
(1) do you agree with the current check in the patch?
(2) given that currently we are not really checking in the RM either whether
there is available CPU in a node (if I am not wrong), should I also disable it
(or add a boolean to control if the check should happen) in the NM as well?
> QueuingContainerManager does not start GUARANTEED Container even if Resources
> are available
> -------------------------------------------------------------------------------------------
>
> Key: YARN-5117
> URL: https://issues.apache.org/jira/browse/YARN-5117
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Arun Suresh
> Assignee: Konstantinos Karanasos
> Attachments: YARN-5117.001.patch
>
>
> When NM Queuing is turned on, it looks like GUARANTEED containers do not
> start even when there are no containers running on the NM. The following is
> seen in the logs :
> {noformat}
> .....
> 2016-05-19 22:34:12,711 INFO [IPC Server handler 0 on 49809]
> queuing.QueuingContainerManagerImpl
> (QueuingContainerManagerImpl.java:pickOpportunisticContainersToKill(351)) -
> There are no sufficient resources to start guaranteed
> container_1463711648301_0001_01_000001 even after attempting to kill any
> running opportunistic containers.
> .....
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
