[
https://issues.apache.org/jira/browse/YARN-4552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15091308#comment-15091308
]
Xuan Gong commented on YARN-4552:
---------------------------------
Thanks, Junping for working on this. The patch looks good overall.
* Can we add some testcases to verify it, please ?
* instead of using
{code}
try {
lfs.listStatus(new Path(localDir));
return;
} catch (…) {
} catch (…)
{code}
Can we directly using
{code}
lfs.util().exist(somepath)
{code}
to check whether the path exists or not ?
> NM ResourceLocalizationService should check and initialize local filecache
> dir (and log dir) even if NM recover is enabled.
> ---------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-4552
> URL: https://issues.apache.org/jira/browse/YARN-4552
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Reporter: Junping Du
> Assignee: Junping Du
> Priority: Critical
> Attachments: YARN-4552.patch
>
>
> In some cases, user are cleanup localized file cache for debugging/trouble
> shooting purpose during NM down time. However, after bring back NM (with
> recovery enabled), the job submission could be failed for exception like
> below:
> {noformat}
> Diagnostics: java.io.FileNotFoundException: File
> /disk/12/yarn/local/filecache does not exist.
> {noformat}
> This is due to we only create filecache dir when recover is not enabled
> during ResourceLocalizationService get initialized/started.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
