[jira] Commented: (HADOOP-4018) limit memory usage in jobtracker

["Amar Kamat (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-4018?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12627918#action_12627918 ] 

Amar Kamat commented on HADOOP-4018:
------------------------------------

bq. Can you pl explain which potion of code you are referring to here?
Look at how the job is inited. If the init fails then there is a cleanup process associated with it. So simply throwing an exception would work and there is no need to explicitly set the job state and finish-time.

bq. This API is used by JobInProgress.initTasks. This method computes the number of tasks that is needed by this job.
Oops! I missed that. But its still flawed as I have mentioned in comment #3. Plz check.

bq. Regarding 3 and 4 i agree with you that it is better if I can check these limits in the constructor of JobInProgress.  ....
I just checked and it seems that the job client never overwrites the number of maps to be spawned. Since the num-maps passed by the user is just a hint to the jobclient while calculating the splits, this information is of no use to the jobtracker and hence the job-client can overwrite the num-maps parameter before uploading the {{job.xml}} on the dfs. With this the job that should fail will fail fast (i.e in the constructor itself) and the user will be informed as to why the job failed. 

Comment #3 just states that {{totalNumTasks()}} will also count tasks from non-running (i.e killed/completed/failed) jobs. So {{totalNumTasks()}} should only take {{RUNNING}} jobs into consideration which calculating total tasks.

> limit memory usage in jobtracker
> --------------------------------
>
>                 Key: HADOOP-4018
>                 URL: https://issues.apache.org/jira/browse/HADOOP-4018
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: mapred
>            Reporter: dhruba borthakur
>            Assignee: dhruba borthakur
>         Attachments: maxSplits.patch, maxSplits2.patch, maxSplits3.patch, maxSplits4.patch, maxSplits5.patch
>
>
> We have seen instances when a user submitted a job with many thousands of mappers. The JobTracker was running with 3GB heap, but it was still not enough to prevent memory trashing from Garbage collection; effectively the Job Tracker was not able to serve jobs and had to be restarted.
> One simple proposal would be to limit the maximum number of tasks per job. This can be a configurable parameter. Is there other things that eat huge globs of memory in job Tracker?

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