[
https://issues.apache.org/jira/browse/HADOOP-4576?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sreekanth Ramakrishnan updated HADOOP-4576:
-------------------------------------------
Attachment: HADOOP-4576-4.patch
Attaching latest patch incorporating Amar's comments.
With respect to the test cases:
Following is the order we are testing the scheduling information:
- Submit 5 jobs to a queue.
- Check the waiting jobs count, it should be 5.
- Then run initializationPoller(), this initializes first two jobs in the queue
but does not raise the job status changed event.
- Check once again the waiting queue, it should be 5 jobs again.
- Then raise status change events.
- Assign one task to a task tracker.
- Check waiting job count, it should be 4 now.
- Then pick an initialized job but not scheduled job and fail it.
- Run the poller, since poller is responsible for removing, failed jobs and
scheduled jobs from job queue maintained by job queue manager, there is no
requirement for raising status changed event as the job is cleared without it
by the poller once job fails before it gets scheduled. With respect to removal
of the jobs which are scheduled and failing is taken care by the
JobQueueManager.
- Check the waiting job count should now be 3.
- Now fail a job which has not been initialized at all.
- Run the poller, so that it can clean up the job queue.
- Check the count, the waiting job count should be 2.
I hope this answers the doubt with regarding test case, I have also mentioned
the steps as inline comment in the test case.
> Modify pending tasks count in the UI to pending jobs count in the UI
> --------------------------------------------------------------------
>
> Key: HADOOP-4576
> URL: https://issues.apache.org/jira/browse/HADOOP-4576
> Project: Hadoop Core
> Issue Type: Bug
> Components: contrib/capacity-sched
> Affects Versions: 0.19.0
> Reporter: Hemanth Yamijala
> Assignee: Sreekanth Ramakrishnan
> Attachments: HADOOP-4576-1.patch, HADOOP-4576-2.patch,
> HADOOP-4576-3.patch, HADOOP-4576-4.patch
>
>
> The UI for capacity scheduler displays 'pending tasks' counts. However the
> capacity scheduler does not update these counts to be the actual values for
> optimization purposes, for e.g. to avoid walking all pending jobs on all
> heartbeats. Hence this information is not very accurate.
> Also, while 'running tasks' counts are useful to compare against capacities
> and limits, 'pending tasks' counts do not add too much user value. A better
> count to display would be the number of running and pending jobs.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
