[ 
https://issues.apache.org/jira/browse/MAPREDUCE-3426?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13167894#comment-13167894
 ] 

Hitesh Shah commented on MAPREDUCE-3426:
----------------------------------------

Summary of changes: 

LocalContainerLauncher changes:
  - in runSubTask(), set the correct config values for task specific 
information as well as the local dirs setting so that the output is written to 
the correct dir ( which the process has write permissions to )
  - cleaned up commented out code. 

Added flags to be able to profile map and reduce tasks with separate profiling 
params. 

MRAppMaster changes: 
  - fixed disabling of speculative execution if job is uberized.
  - Moved uber decision into new function
    - hardcoded max reduce count supported to 1 as multiple reduces are not 
supported. 
    - using MR AM vm size as memory bound check for uber decision 
  
RMCommunicator change:
  - for MR-3421 

JobStatus/JobReport changes:
  - Print out uber choice on client side to inform user whether running job in 
uber or non-uber mode

Comments changes:
   - Removed comments on race condition in local launcher for "MapTask can't 
send TA_UPDATE before TA_CONTAINER_LAUNCHED" 
     - the state machine has a write lock, also the launched event is sent 
first followed by any updates if any from the task  itself. 
   - renameMapOutputForReduce is still required. Output files are generated 
into top level dir and not a task specific dir. 
                
> uber-jobs tried to write outputs into wrong dir
> -----------------------------------------------
>
>                 Key: MAPREDUCE-3426
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-3426
>             Project: Hadoop Map/Reduce
>          Issue Type: Sub-task
>          Components: mrv2
>    Affects Versions: 0.23.0, 0.24.0
>            Reporter: Hitesh Shah
>             Fix For: 0.23.1
>
>         Attachments: MR-3426.1.patch, MR-3426.2.patch
>
>
> Incorrect setup of the uber tasks causes tasks to try to write intermidiate 
> outputs into dirs that the user does not have permissions to write to on a 
> secure cluster. 

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to