[ https://issues.apache.org/jira/browse/HADOOP-3130?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12584213#action_12584213 ] 

Devaraj Das commented on HADOOP-3130:
-------------------------------------

bq. At least, we have to differentiate between the case of failing to fetch one map output numerous times and the case of failing to
fetch a lot of different map outputs. In the first case, it is better to re-execute the map. 
In the second case, maybe it makes sense to consider to fail the reducer.

That's how the model works in case of maps - if too many reducers complain about fetch failure for a particular map, the map is killed. The change here should be to also consider the host the map ran on otherwise we run into issues like HADOOP-2175 (https://issues.apache.org/jira/browse/HADOOP-2175?focusedCommentId=12582425#action_12582425). The problem in the reducers case is this that the numbers are hardcoded and the decision to kill is totally local. So if a reducer fails to fetch 5 unique map outputs it kills itself. This should be augmented with your suggestion on accounting for the shuffle progress.

> Shuffling takes too long to get the last map output.
> ----------------------------------------------------
>
>                 Key: HADOOP-3130
>                 URL: https://issues.apache.org/jira/browse/HADOOP-3130
>             Project: Hadoop Core
>          Issue Type: Bug
>            Reporter: Runping Qi
>         Attachments: HADOOP-3130.patch, shuffling.log
>
>
> I noticed that towards the end of shufflling, the map output fetcher of the reducer backs off too aggressively.
> I attach a fraction of one reduce log of my job.
> Noticed that the last map output was not fetched in 2 minutes.

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