[
https://issues.apache.org/jira/browse/MAPREDUCE-5351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13698228#comment-13698228
]
Sandy Ryza commented on MAPREDUCE-5351:
---------------------------------------
In JIP.cleanupJob:
{code}
Path tempDir = jobtracker.getSystemDirectoryForJob(getJobID());
CleanupQueue.getInstance().addToQueue(
new PathDeletionContext(tempDir, conf));
// delete the staging area for the job and cancel delegation token
String jobTempDir = conf.get("mapreduce.job.dir");
if (jobTempDir != null && conf.getKeepTaskFilesPattern() == null &&
!conf.getKeepFailedTaskFiles()) {
Path jobTempDirPath = new Path(jobTempDir);
tempDirFs = jobTempDirPath.getFileSystem(conf);
CleanupQueue.getInstance().addToQueue(
new PathDeletionContext(jobTempDirPath, conf, userUGI, jobId));
}
{code}
The CleanupQueue is used twice, once with the UGI set and once without.
> JobTracker memory leak caused by CleanupQueue reopening FileSystem
> ------------------------------------------------------------------
>
> Key: MAPREDUCE-5351
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5351
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: jobtracker
> Affects Versions: 1.1.2
> Reporter: Sandy Ryza
> Assignee: Sandy Ryza
> Priority: Critical
> Fix For: 1.2.1
>
> Attachments: MAPREDUCE-5351-1.patch, MAPREDUCE-5351-2.patch,
> MAPREDUCE-5351-addendum.patch, MAPREDUCE-5351.patch
>
>
> When a job is completed, closeAllForUGI is called to close all the cached
> FileSystems in the FileSystem cache. However, the CleanupQueue may run after
> this occurs and call FileSystem.get() to delete the staging directory, adding
> a FileSystem to the cache that will never be closed.
> People on the user-list have reported this causing their JobTrackers to OOME
> every two weeks.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
