[
https://issues.apache.org/jira/browse/HADOOP-4653?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12647263#action_12647263
]
Hemanth Yamijala commented on HADOOP-4653:
------------------------------------------
Since the operation of changing the priority for a completed job does not make
sense, I propose this be stopped in the jobtracker itself, rather than having
all listeners handle this case as required.
We could:
- change the priority, but not notify if the job is completed
- not change the priority at all, and treat it as a no-op. The problem with
this is that the job client could treat this operation as successful, and lead
to misleading results.
- throw an exception to indicate the operation is invalid.
- make setJobPriority return a boolean indicating Success/Failure.
Unfortunately, would this be a backwards incompatible change ?
Also, we need to make the web UI handle this.
Probably throwing an exception is the most feasible, yet correct solution.
Thoughts ?
> Changing priority of a completed job causes problems in JobInProgressListeners
> ------------------------------------------------------------------------------
>
> Key: HADOOP-4653
> URL: https://issues.apache.org/jira/browse/HADOOP-4653
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Reporter: Hemanth Yamijala
>
> If the priority of a completed job is changed, a {{JobChangeEvent}} is raised
> and the {{JobInProgressListeneners}}, like the capacity scheduler, are
> notified. Most implementations handle the event by re-sorting their data
> structures, and thus could end up re-inserting the completed job into their
> lists.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
