[
https://issues.apache.org/jira/browse/HADOOP-5420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12698183#action_12698183
]
Hemanth Yamijala commented on HADOOP-5420:
------------------------------------------
bq. Further, in the current code, this parameter is job configurable ...
Sigh. I didn't notice this. In that case, I am not comfortable with this patch.
Configuring a value at two places is inconvenient, but does not violate
correctness. However, a per job value being overridden by a cluster wide
configuration just in the case of LinuxTaskController does not seem right.
Further, with the kind of plans mentioned in the comment above (though I saw
nothing in HADOOP-5614 - having the thoughts recorded there will help), I think
we should relook at how this part is designed.
For the short term, it is probably still OK to not have code corrresponding to
the SIGKILL code at all in the task-controller. Then the issue of how to handle
SIGKILL can be addressed as a follow-up preferably after HADOOP-4490 is
committed.
> Support killing of process groups in LinuxTaskController binary
> ---------------------------------------------------------------
>
> Key: HADOOP-5420
> URL: https://issues.apache.org/jira/browse/HADOOP-5420
> Project: Hadoop Core
> Issue Type: Bug
> Reporter: Sreekanth Ramakrishnan
> Assignee: Sreekanth Ramakrishnan
> Attachments: hadoop-5420-1.patch, hadoop-5420.patch
>
>
> Support setsid based kill in LinuxTaskController.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
