[
https://issues.apache.org/jira/browse/YARN-9846?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16934761#comment-16934761
]
Jim Brennan commented on YARN-9846:
-----------------------------------
[~belugabehr] thanks for the patch, but can you provide some background on what
motivated this change? It's not clear to me that the new approach is
actually better in this case. In the handle() and cleanupPrivLocalizers()
methods, you are now acquiring two locks instead of one. And in
processHeartbeat() we are no longer holding the privLocalizers lock while
calling the localizer.processHeartbeat() - I'm not sure if that will break
anything, but the localization code is pretty fragile so I'd be careful.
I personally find the refactoring of LocalizerTracker.handle() to be less
readable than the original, but that may just be a style issue.
> Use Finer-Grain Synchronization in ResourceLocalizationService
> --------------------------------------------------------------
>
> Key: YARN-9846
> URL: https://issues.apache.org/jira/browse/YARN-9846
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: nodemanager
> Affects Versions: 3.2.0
> Reporter: David Mollitor
> Assignee: David Mollitor
> Priority: Minor
> Attachments: YARN-9846.1.patch, YARN-9846.2.patch, YARN-9846.3.patch
>
>
> https://github.com/apache/hadoop/blob/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/containermanager/localizer/ResourceLocalizationService.java#L788
> # Remove these synchronization blocks
> # Ensure {{recentlyCleanedLocalizers}} is thread safe
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
