[ 
https://issues.apache.org/jira/browse/MAPREDUCE-1486?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12835514#action_12835514
 ] 

Aaron Kimball commented on MAPREDUCE-1486:
------------------------------------------

I looked at the failing testcases reported by Hudson. None of them reproduced 
locally:

* TestNodeRefresh seems to be bit by the "Hudson doesn't get the whole 
classpath" issue that's been coming up recently. The NoClassDefFoundErrors are 
really bizarre; does anyone have any insight into why this has hit Hudson in 
the last couple weeks?
* TestMapReduceLazyOutput seems to have trouble initializing a MiniDFSCluster 
and getting the port it wants; this is unrelated to the patch, as mini-dfs 
initialization happens outside any MapContext.
* TestMultipleLevelCaching has a broken test; I filed MAPREDUCE-1508 to address 
this. Unrelated failure (looking at the code at 
TestMultipleLevelCaching.java:122, it's probably a MiniDFS init problem like 
TestMapReduceLazyOutput)
* TestJobOutputCommitter also passes locally.

> Configuration data should be preserved within the same MapTask
> --------------------------------------------------------------
>
>                 Key: MAPREDUCE-1486
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-1486
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: task
>            Reporter: Aaron Kimball
>            Assignee: Aaron Kimball
>         Attachments: MAPREDUCE-1486.patch
>
>
> Map tasks involve a number of Contexts -- at least a TaskAttemptContext and a 
> MapContext. These context objects contain a Configuration each; when one 
> context is initialized, it initializes its own Configuration by deep-copying 
> a previous Configuration.
> If one Context instance is used entirely prior to a second, more specific 
> Context then the second Context should contain the configuration data 
> initialized in the previous Context. This specifically affects the 
> interaction between an InputFormat and its RecordReader instance(s).

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

Reply via email to