[jira] Commented: (HADOOP-1128) Missing progress information in map tasks

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

    [ https://issues.apache.org/jira/browse/HADOOP-1128?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12481705 ] 

Owen O'Malley commented on HADOOP-1128:
---------------------------------------

This is a good catch. A couple of other points:
  1. start should be set to the position in the stream after the sync, so that the "real" start point is used.
  2. the result of the division should be compared to 1.0 to make sure the getProgress never returns numbers bigger than 1.0f. This can happen because of the way that splits are chosen blindly and then adjusted to the sync boundaries. (Both the start and end boundaries are pushed back to the next sync boundary.)

> Missing progress information in map tasks
> -----------------------------------------
>
>                 Key: HADOOP-1128
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1128
>             Project: Hadoop
>          Issue Type: Bug
>          Components: mapred
>    Affects Versions: 0.12.1
>            Reporter: Andrzej Bialecki 
>         Assigned To: Andrzej Bialecki 
>             Fix For: 0.12.1
>
>         Attachments: progress.patch
>
>
> Long-running map tasks don't update properly their progress - the propgress percentage stays at 0% only to jump suddenly at the end of the task to 100%. The reason, discovered by Espen Amble Kolstad, is that there's a missing cast to float in SequenceFileRecordReader and in LineRecordReader.

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