[
https://issues.apache.org/jira/browse/MAPREDUCE-5570?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13933477#comment-13933477
]
Jason Lowe commented on MAPREDUCE-5570:
---------------------------------------
Looks good overall, just some cleanup needed on the test:
- The test should verify the task is in the expected state before and after
sending the fetch failure event and the task has a non-zero finish time,
otherwise we could be passing the test for the wrong reasons (e.g.: finish time
is 0 because task does not think it's finished or finish time doesn't change
because we ignored the fetch failure event completely because task is in the
wrong state).
- Nit: The test doesn't need to sleep for a full second, probably only 1 to 10
msec. Just enough time to make sure the finish timestamp would expect to be
different if it was updated. A second doesn't seem like much, but it adds up
across all the tests if they sleep longer than necessary.
- Nit: There's some unnecessary whitespace changes in the patch at the top of
the TestTaskAttempt class
> Map task attempt with fetch failure has incorrect attempt finish time
> ---------------------------------------------------------------------
>
> Key: MAPREDUCE-5570
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5570
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: mr-am, mrv2
> Affects Versions: 0.23.9, 2.1.1-beta
> Reporter: Jason Lowe
> Assignee: Rushabh S Shah
> Attachments: patch-MapReduce-5570.patch
>
>
> If a map task attempt is retroactively failed due to excessive fetch failures
> reported by reducers then the attempt's finish time is set to the time the
> task was retroactively failed rather than when the task attempt completed.
> This causes the map task attempt to appear to have run for much longer than
> it actually did.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
