[
https://issues.apache.org/jira/browse/MAPREDUCE-2801?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13105023#comment-13105023
]
Arun C Murthy commented on MAPREDUCE-2801:
------------------------------------------
Bobby, the more I think about this approach the less I like it.
Essentially relying on NM to supply critical env for the task is a broken
concept - this ensures that you can't run the same task with different env.
Here 'env' can be jars, shared-objects etc.
I'm thinking we shud just bite the bullet, go the whole 9 yards and fix the
classpath/ld_library_path etc. in the *correct* way i.e. throw out our reliance
on bin/mapred and do everything via the DistributedCache i.e. MAPREDUCE-2880.
Thoughts?
Apologies for dragging this. IAC, this won't fix the original issue - the NM
doesn't have the right java.library.path for setting up compression libs
correctly.
Arguably we can fix this, but I'd rather do it the *right* way i.e.
MAPREDUCE-2880.
> Include the native libs in java.library.path
> ---------------------------------------------
>
> Key: MAPREDUCE-2801
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-2801
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mrv2
> Affects Versions: 0.23.0
> Reporter: Ramya Sunil
> Assignee: Robert Joseph Evans
> Fix For: 0.23.0
>
> Attachments: MR-2801.patch, MR-2801.patch
>
>
> For the child tasks in mrv2, java.library.path is set to just $PWD and the
> native libs are not included. Whereas in 0.20.x, java.library.path for child
> tasks was set to <path to native libs>:$PWD
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira