[
https://issues.apache.org/jira/browse/MAPREDUCE-2218?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12974206#action_12974206
]
Lianhui Wang commented on MAPREDUCE-2218:
-----------------------------------------
that is a good idea.but there are many updated things on the JT and TT.at the
JT,after the following:
// Check for tasks to be killed
List<TaskTrackerAction> killTasksList = getTasksToKill(trackerName);
if (killTasksList != null) {
actions.addAll(killTasksList);
}
// Check for jobs to be killed/cleanedup
List<TaskTrackerAction> killJobsList = getJobsForCleanup(trackerName);
if (killJobsList != null) {
actions.addAll(killJobsList);
}
we continuely use the job scheduer to assign the new tasks,the number is the
same the killed number.
at the TT,it must wait the task to be killed.but the kill task is in the daemon
thread。it must monitor the thread kill the task,after it,that can run the task.
> schedule additional tasks when killactions are being dispatched
> ---------------------------------------------------------------
>
> Key: MAPREDUCE-2218
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-2218
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: jobtracker
> Reporter: Joydeep Sen Sarma
>
> we see a very high turnaround time for preemption (from the time preemption
> happens to the time other jobs are scheduled). part of the reason is that we
> do not dispatch new tasks until the TT confirms termination of killed tasks.
> The high turnaround time causes all kinds of issues (we keep issuing
> preemptions repeatedly. other jobs of higher priority arrive and get
> scheduled).
> One remediation is that one can dispatch additional tasks equal to the number
> of tasks killed whenever we send killaction to a TT. that would reduce the
> turnaround time somewhat.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
