[ 
https://issues.apache.org/jira/browse/MAPREDUCE-6434?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14707248#comment-14707248
 ] 

Augusto Souza commented on MAPREDUCE-6434:
------------------------------------------

Thank you very much [~chris.douglas]! 

I tested the version I submitted before with a map-only job, and I think the 
IdentityReducer is used in cases with num reduce tasks setted to zero and no 
setting for the reducer class, so the previous patch doesn't crash. Am I right 
in this assumption? Is there another way of defining jobs in which I force the  
{{context.getReducerClass()}} to get undefined?

But, I think your feedback is valid, so I am adding a another statement to the 
expression to make sure only the PartialFileOutputCommiter can only be 
instantiated for reduce tasks. If there is a way of making 
{{context.getReducerClass()}} undefined, I can try to make better tests for the 
patch too.


> Add support for PartialFileOutputCommiter when checkpointing is an option 
> during preemption
> -------------------------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-6434
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-6434
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>            Reporter: Augusto Souza
>            Assignee: Augusto Souza
>         Attachments: MAPREDUCE-6434.001.patch, MAPREDUCE-6434.002.patch, 
> MAPREDUCE-6434.003.patch, MAPREDUCE-6434.004.patch, MAPREDUCE-6434.005.patch
>
>
> Finish up some renaming work related to the annotation @Preemptable (it 
> should be @Checkpointable now) and help in the splitting of patch in 
> MAPREDUCE-5269 that is too large for being reviewed or accepted by Jenkins CI 
> scripts.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to