You are viewing a plain text version of this content. The canonical link for it is here.

Posted to mapreduce-issues@hadoop.apache.org by "Jinghui Wang (JIRA)" <ji...@apache.org> on 2014/02/18 18:52:20 UTC

[jira] [Commented] (MAPREDUCE-5490) MapReduce doesn't set the environment variable for children processes

    [ https://issues.apache.org/jira/browse/MAPREDUCE-5490?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13904308#comment-13904308 ] 

Jinghui Wang commented on MAPREDUCE-5490:
-----------------------------------------

Added patch based on trunk. Making exporting the classpath as a configurable option in case some people are still used to the seeing it under ps -ef. Also have checked the outputted script that starts the Child manually and looks fine. An unit test was also added here.

> MapReduce doesn't set the environment variable for children processes
> ---------------------------------------------------------------------
>
>                 Key: MAPREDUCE-5490
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5490
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>    Affects Versions: 1.2.1
>            Reporter: Owen O'Malley
>            Assignee: Owen O'Malley
>         Attachments: MAPREDUCE-5490.patch, mr-5490.patch, mr-5490.patch
>
>
> Currently, MapReduce uses the command line argument to pass the classpath to the child. This breaks if the process forks a child that needs the same classpath. Such a case happens in Hive when it uses map-side joins. I propose that we make MapReduce in branch-1 use the CLASSPATH environment variable like YARN does.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)