[
https://issues.apache.org/jira/browse/MAPREDUCE-3?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12729916#action_12729916
]
Aaron Kimball commented on MAPREDUCE-3:
---------------------------------------
A caution here is that {{mapred.child.ulimit}} needs to account for the memory
overhead of the JVM itself. Merely setting {{mapred.child.ulimit}} to the same
value as the {{-Xmx}}_hhhh_{{m}} in {{mapred.child.java.opts}} will fail to
launch child tasks. You'll need some overhead room; I don't know exactly how
much.
> Set mapred.child.ulimit automatically to the value of the RAM limits for a
> job, if they are set
> -----------------------------------------------------------------------------------------------
>
> Key: MAPREDUCE-3
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-3
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Reporter: Hemanth Yamijala
>
> Memory based monitoring and scheduling allow users to set memory limits for
> the tasks of their jobs. This parameter is the total memory taken by the
> task, and any children it may launch (for e.g. in the case of streaming). A
> related parameter is mapred.child.ulimit which is a hard limit on the memory
> used by a single process of the entire task tree. For user convenience, it
> would be sensible for the system to set the ulimit to atleast the memory
> required by the task, if the user has specified the latter.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
