[
https://issues.apache.org/jira/browse/HADOOP-5516?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vinod K V updated HADOOP-5516:
------------------------------
Attachment: HADOOP-5516-1.txt
This seems to happen when ProcfsBasedProcessTree.getProcessTree() when the root
process is no longer alive. Attaching a patch for branch-0.20. This does the
following:
- When the root process is not found, getProcessTree() method returns an empty
process-tree.
- Added another null check in the same method when we cannot read process
information from /proc vfs.
- Surrounded the memory management code of a single task with a try-catch
block so that MemoryManager thread continues to function on other tasks even in
the event of occurrence of uncaught exceptions on one task.
- Removed a spurious HashMap.put() inside an iterator on the same map
processTreeInfoMap . This was pointed out by Hemanth.
- Modified test-case to verify that method calls on ProcfsBasedProcessTree are
sane even if the process-tree is not alive or is gone.
> TaskMemoryManagerThread crashes in a corner case
> ------------------------------------------------
>
> Key: HADOOP-5516
> URL: https://issues.apache.org/jira/browse/HADOOP-5516
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.20.0
> Reporter: Vinod K V
> Assignee: Vinod K V
> Priority: Blocker
> Attachments: HADOOP-5516-1.txt
>
>
> TT's stdout says.
> {code}
> Exception in thread "org.apache.hadoop.mapred.TaskMemoryManagerThread"
> java.lang.NullPointerException
> at
> org.apache.hadoop.util.ProcfsBasedProcessTree.getProcessTree(ProcfsBasedProcessTree.java:126)
> at
> org.apache.hadoop.mapred.TaskMemoryManagerThread.run(TaskMemoryManagerThread.java:200)
> {code}
> TaskMemoryManager crashes and no further memory management is done.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
