[
https://issues.apache.org/jira/browse/YARN-1775?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13941120#comment-13941120
]
Chris Nauroth commented on YARN-1775:
-------------------------------------
[~rajesh.balamohan], thank you for explaining the testing. These results sound
very promising! Also interesting would be confirming that containers still get
killed for exceeding the limit with private/non-shared pages.
I wonder then if counting RSS still has some potential advantages in certain
deployments, or if the PSS approach is always superior. Your testing so far
seems to indicate that PSS is always superior. Therefore, should this just be
combined right into the current code? (This echoes Vinod's prior comment about
folding the logic back into {{ProcfsBasedProcessTree}}.) A conservative
approach is to introduce a config flag, try to get some experience running it
in real-world clusters, and then we can flip the default in a later release if
it goes well.
> Create SMAPBasedProcessTree to get PSS information
> --------------------------------------------------
>
> Key: YARN-1775
> URL: https://issues.apache.org/jira/browse/YARN-1775
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: nodemanager
> Reporter: Rajesh Balamohan
> Assignee: Rajesh Balamohan
> Attachments: yarn-1775-2.4.0.patch
>
>
> Create SMAPBasedProcessTree (by extending ProcfsBasedProcessTree), which will
> make use of PSS for computing the memory usage.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
