[
https://issues.apache.org/jira/browse/YARN-3136?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14305891#comment-14305891
]
Jason Lowe commented on YARN-3136:
----------------------------------
It appears getTransferredContainers is grabbing the lock because it's not sure
it's safe to lookup the SchedulerApplication from the applications map, yet in
practice it's always a ConcurrentHashMap. Similarly the lookup of the RMApp
is also from a concurrent hash map and does not require a lock. After that
we're simply walking the containers of the SchedulerApplication which at best
should only be locking the app and not the entire scheduler. Or am I missing a
critical point where we really need the scheduler lock?
> getTransferredContainers can be a bottleneck during AM registration
> -------------------------------------------------------------------
>
> Key: YARN-3136
> URL: https://issues.apache.org/jira/browse/YARN-3136
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: scheduler
> Affects Versions: 2.6.0
> Reporter: Jason Lowe
>
> While examining RM stack traces on a busy cluster I noticed a pattern of AMs
> stuck waiting for the scheduler lock trying to call getTransferredContainers.
> The scheduler lock is highly contended, especially on a large cluster with
> many nodes heartbeating, and it would be nice if we could find a way to
> eliminate the need to grab this lock during this call. We've already done
> similar work during AM allocate calls to make sure they don't needlessly grab
> the scheduler lock, and it would be good to do so here as well, if possible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
