[ 
https://issues.apache.org/jira/browse/YARN-4512?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15091339#comment-15091339
 ] 

Karthik Kambatla commented on YARN-4512:
----------------------------------------

bq. Regarding the overallocation vs oversubscription, I'm more concerned about 
consistency, an example is {{Context}} in the NM:
Fair point. Will fix them up in the next iteration.

bq. I'd say the RM should know what the NM is doing, otherwise, after a 
container preemption, the RM may try to keep doing the same and the NM 
preempting again.
When the NM preempts a container, it needs to provide that information to the 
RM anyway so the container is allocated elsewhere. The threshold itself is 
inconsequential here, no? The RM might be able to use the preemptionThreshold 
as a lookahead, but I wonder if it defeats the purpose of having two configs in 
the first place. 

> Provide a knob to turn on over-allocation
> -----------------------------------------
>
>                 Key: YARN-4512
>                 URL: https://issues.apache.org/jira/browse/YARN-4512
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: nodemanager
>            Reporter: Karthik Kambatla
>            Assignee: Karthik Kambatla
>         Attachments: YARN-4512-YARN-1011.001.patch, 
> yarn-4512-yarn-1011.002.patch
>
>




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

Reply via email to