[jira] Commented: (MAPREDUCE-144) TaskMemoryManager should log process-tree's status while killing tasks.

["Hemanth Yamijala (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAPREDUCE-144?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12750840#action_12750840 ] 

Hemanth Yamijala commented on MAPREDUCE-144:
--------------------------------------------

Some comments:
- Firstly, it appears cmdline could be empty if the whole process is swapped out. I suppose this is possible. Is the method we use reliable then ?
- I think getCmdLine can return "Unknown" rather than empty when it can't find a command line.
- Also, the replacement of '\0' with space should happen in getCmdLine, that too, only when we read the line, rather than getProcessTreeDump(), because it is more relevant only there.
- I think it will be useful to also show session id and process group id, so that we can catch for mysterious changes to these if tasks are doing them.
- Regarding the test case, it seems we can get away with a unit test for this, just like testVirtualMemoryForProcessTree(), which would be more controlled and reliable rather than on the relying on system behavior of procfs. That way, even if there is a problem, we can differentiate between system incompatibilities and bugs in code. I am fine with the existing test, but would like to treat that as a smoke test of the feature, but have a unit test for fast runs.

> TaskMemoryManager should log process-tree's status while killing tasks.
> -----------------------------------------------------------------------
>
>                 Key: MAPREDUCE-144
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-144
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>            Reporter: Vinod K V
>            Assignee: Vinod K V
>         Attachments: HADOOP-5570-1.patch, HADOOP-5570-20090614.txt, HADOOP-5570-20090615.txt, HADOOP-5570-20090619.txt, HADOOP-5570-20090622.txt, MAPREDUCE-144-20090902.txt
>
>
> This helps a lot in debugging why a particular task has gone beyond memory limits.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.