[
https://issues.apache.org/jira/browse/HADOOP-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Enis Soztutar updated HADOOP-1351:
----------------------------------
Attachment: killTask_v1.5.patch
At last, here is the final(i hope) version of the patch.
As per Doug's and Arun's comments i have enabled both killing and failing
tasks.
{{JobSubmissionProtocol}} includes
{code}
public void killTask(String taskId, boolean shouldFail) throws IOException;
{code}
the shouldFail argument is used to indicate failing or killing the task.
and we have command line options
{noformat}
bin/hadoop job -kill-task <task_id>
bin/hadoop job -fail-task <task_id>
{noformat}
I have observed some issues with failing the task directly, such as that we
discard task output, but until the kill action has been send to TT, tasks will
output data. So i have sticked with the first design.
> Want to kill a particular task or attempt
> -----------------------------------------
>
> Key: HADOOP-1351
> URL: https://issues.apache.org/jira/browse/HADOOP-1351
> Project: Hadoop
> Issue Type: New Feature
> Components: mapred
> Affects Versions: 0.15.0
> Reporter: Owen O'Malley
> Assignee: Enis Soztutar
> Fix For: 0.15.0
>
> Attachments: killTask_v1.0.1.patch, killTask_v1.2.patch,
> killTask_v1.3.patch, killTask_v1.4.patch, killTask_v1.5.patch
>
>
> It would be convenient to be able to kill a particular task or attempt from
> the command line. It would look like:
> bin/hadoop job -kill-task tip_0001_m_000000
> bin/hadoop job -kill-attempt task_0001_m_000000_0
> This would allow the user to tell the system to stop a particular task or
> attempt without having to restart a task tracker.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
