[
https://issues.apache.org/jira/browse/YARN-7373?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16213026#comment-16213026
]
Haibo Chen commented on YARN-7373:
----------------------------------
[~asuresh] Can you please provide some background and details of container
update?
The atomicity is not clear to us in term of how it is guaranteed. Our concern
is that
another container allocation may come in between the two containerUpdated() call
and there is not enough resource available for the allocation.
> The atomicity of container update in RM is not clear
> ----------------------------------------------------
>
> Key: YARN-7373
> URL: https://issues.apache.org/jira/browse/YARN-7373
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: resourcemanager
> Reporter: Haibo Chen
> Assignee: Haibo Chen
>
> While reviewing YARN-4511, Miklos noticed that
> {code:java}
> 342 // notify schedulerNode of the update to correct resource accounting
> 343 node.containerUpdated(existingRMContainer, existingContainer);
> 344
> 345
> ((RMContainerImpl)tempRMContainer).setContainer(updatedTempContainer);
> 346 // notify SchedulerNode of the update to correct resource accounting
> 347 node.containerUpdated(tempRMContainer, tempContainer);
> 348
> {code}
> bq. I think that it would be nicer to lock around these two calls to become
> atomic.
> Container update, and thus container swap as part of that, is atomic
> according to [~asuresh].
> It'd be nice to discuss this in more details to see if we want to be more
> conservative.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
