zhihai xu commented on YARN-3491:

I uploaded a new patch YARN-3491.003 patch for review.
I think the new patch YARN-3491.003 is better than previous solutions.
It can solve the race condition "if a directory goes from bad to good." in 
[~wilfreds]'s comment.
I add a small feature in DirectoryCollection.java:{{DirsChangeListener}}.
{{ResourceLocalizationService}} can register {{DirsChangeListener}} for 
localDirs and logDirs.
Once DirectoryCollection#localDirs is changed in {{checkDirs}}, 
{{ResourceLocalizationService}} will get callback 
DirsChangeListener#onDirsChanged, which will call {{getInitializedLocalDirs}} 
and {{getInitializedLogDirs}}

> PublicLocalizer#addResource is too slow.
> ----------------------------------------
>                 Key: YARN-3491
>                 URL: https://issues.apache.org/jira/browse/YARN-3491
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: nodemanager
>    Affects Versions: 2.7.0
>            Reporter: zhihai xu
>            Assignee: zhihai xu
>            Priority: Critical
>         Attachments: YARN-3491.000.patch, YARN-3491.001.patch, 
> YARN-3491.002.patch, YARN-3491.003.patch
> Based on the profiling, The bottleneck in PublicLocalizer#addResource is 
> getInitializedLocalDirs. getInitializedLocalDirs call checkLocalDir.
> checkLocalDir is very slow which takes about 10+ ms.
> The total delay will be approximately number of local dirs * 10+ ms.
> This delay will be added for each public resource localization.
> Because PublicLocalizer#addResource is slow, the thread pool can't be fully 
> utilized. Instead of doing public resource localization in 
> parallel(multithreading), public resource localization is serialized most of 
> the time.
> And also PublicLocalizer#addResource is running in Dispatcher thread, 
> So the Dispatcher thread will be blocked by PublicLocalizer#addResource for 
> long time.

This message was sent by Atlassian JIRA

Reply via email to