[
https://issues.apache.org/jira/browse/MAPREDUCE-794?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12734553#action_12734553
]
Amar Kamat commented on MAPREDUCE-794:
--------------------------------------
Job killed in PREP state will cause the job to invoke job.garbageCollect()
which does
{code}
jobtracker.getInstrumentation().decWaitingMaps(getJobID(), pendingMaps());
jobtracker.getInstrumentation().decWaitingReduces(getJobID(),
pendingReduces());
{code}
which goes ahead and blindly deletes _pendingMaps()=total_maps_ from
_numWaitingMaps_. But pendingMaps() is added to the JobTrackerInstrumentation
only upon init.
{code}
@Override
public synchronized void decWaitingMaps(JobID id, int task) {
numWaitingMaps -= task;
}
{code}
{code}
initTasks() {
..
..
jobtracker.getInstrumentation().addWaitingMaps(getJobID(), numMapTasks);
jobtracker.getInstrumentation().addWaitingReduces(getJobID(), numReduceTasks);
..
..
}
{code}
> JobTrackerInstrumentation data might be grabled
> ------------------------------------------------
>
> Key: MAPREDUCE-794
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-794
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: jobtracker
> Reporter: Amar Kamat
>
> Here is the sequence of events
> 1) submit a job
> 2) kill it in prep state
> This should result into -ve values of pending maps in
> JobTrackerInstrumentation.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
