[MAPREDUCE-5508] JobTracker memory leak caused by unreleased FileSystem objects in JobInProgress#cleanupJob

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1-win, 1.2.1
• Fix Version/s: 1-win, 1.3.0
• Component/s: jobtracker
• Target Version/s: 1-win, 1.3.0
• Hadoop Flags: Reviewed
• Assignee: Douma Fang
• Reporter: Douma Fang
• Created: 13/Sep/13 23:52
• Updated: 07/Nov/13 18:34
• Resolved: 24/Sep/13 05:35

Description

MAPREDUCE-5351 fixed a memory leak problem but introducing another filesystem object (see "tempDirFs") that is not properly released.

 JobInProgress#cleanupJob()

  void cleanupJob() {
...
          tempDirFs = jobTempDirPath.getFileSystem(conf);
          CleanupQueue.getInstance().addToQueue(
              new PathDeletionContext(jobTempDirPath, conf, userUGI, jobId));
...
 if (tempDirFs != fs) {
      try {
        fs.close();
      } catch (IOException ie) {
...
}

Issue Links

• is related to: HDFS-5211 Race condition between DistributedFileSystem#close and FileSystem#close can cause return of a closed DistributedFileSystem instance from the FileSystem cache.
• is related to: MAPREDUCE-5351 JobTracker memory leak caused by CleanupQueue reopening FileSystem
• is related to: HADOOP-9993 FileSystem#close JavaDoc should state that the implementation must be idempotent.