[jira] Commented: (HADOOP-4251) The reducer output is not created if the mapper input is empty

["Owen O'Malley (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-4251?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12633939#action_12633939 ] 

Owen O'Malley commented on HADOOP-4251:
---------------------------------------

I believe that check was put in because it used to fail if the number of maps was 0. Obviously, if you take out the check, you had better add a test case of both reduces != 0 and reduces = 0 when maps = 0.

That said, it is probably better to run the reduces with no input than the current behavior, which tends to surprise users.

> The reducer output is not created if the mapper input is empty
> --------------------------------------------------------------
>
>                 Key: HADOOP-4251
>                 URL: https://issues.apache.org/jira/browse/HADOOP-4251
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: mapred
>            Reporter: Namit Jain
>            Assignee: Namit Jain
>
> For some Hive stuff, I ran into the following scenario:
> For a given map-reduce job, the input was empty. Because of that no mappers and reducers were created. It would have been helpful if an empty output for the reducer would have been created.
> After browsing though the code, it seems that in initTasks() in JobInProgress, no mappers and reducers are initialized if input is empty.
> I was thinking of putting a fix there. If the input is empty, before returning, create the output directory (as specified by the reducer) if needed.Any comments/suggestions....

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