[
https://issues.apache.org/jira/browse/YARN-1927?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13967706#comment-13967706
]
Wangda Tan commented on YARN-1927:
----------------------------------
Hi [~chris.douglas], thanks for your jumping in for this JIRA!
I actually have no doubt in current way to make decision about preempt a
container. My point is (please correct me if I was wrong), we don't need send a
preemption message to scheduler twice or more on the same container. When an
application master get a message about a container will be preempted, it should
do some works like checkpointing and waiting (or just kill itself) the
container killed by NM. IMHO, Multiple preemption message for same container
aren't meaningful to AM.
And another question in my head is, have you ever considered adding a
"preemption cancel" event? Because in following cases: numerous of containers
completed in queue-A, and some other containers in queue-A were marked
"preemption" before, such original marked "preemption" containers may not need
to be actually killed, scheduler may need to tell AM -- "You can go on using
these containers!". Do you agree with me?
> Preemption message shouldn’t be created multiple times for same container-id
> in ProportionalCapacityPreemptionPolicy
> --------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-1927
> URL: https://issues.apache.org/jira/browse/YARN-1927
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacityscheduler
> Affects Versions: 2.4.0
> Reporter: Wangda Tan
> Assignee: Wangda Tan
> Priority: Minor
> Attachments: YARN-1927.patch
>
>
> Currently, after each editSchedule() called, preemption message will be
> created and sent to scheduler. ProportionalCapacityPreemptionPolicy should
> only send preemption message once for each container.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
