[
https://issues.apache.org/jira/browse/YARN-5216?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15363983#comment-15363983
]
Konstantinos Karanasos commented on YARN-5216:
----------------------------------------------
Thanks for the patch, [~hrsharma].
Some first comments:
It seems there is a lot of refactoring that is not really needed for
introducing pluggable preemption policies. Essentially for the preemption
policies, you simply need to override the way you are stopping a container (and
possibly the way you are starting it too?).
For instance, why do we need to pull the queues of opportunistic containers to
a new class? After all, the {{QueuingContainersManagerImpl}} extends the
{{ContainersManagerImpl}} just to add these containers queues.
Another drawback of the refactoring is that in various points we now have to
treat the queued/allocated opportunistic differently from the queued/allocated
guaranteed by adding many explicit if statements.
More concretely, I suggest to add a {{ContainerPreemptionManager}} that
provides different ways to do the stopping (and starting?) of containers,
instead of pulling the data structures from the
{{QueuingContainersManagerImpl}} and having to override many more methods.
Minor:
# Let's give a more descriptive name to the first implementation of the
{{ContainerPreemptionManager}} to show that it preempts containers by killing
them.
# After the parameter you added in {{YarnConfiguration}}, please add an extra
constant that determines its default value (check the other parameters for
reference).
> Expose configurable preemption policy for OPPORTUNISTIC containers running on
> the NM
> ------------------------------------------------------------------------------------
>
> Key: YARN-5216
> URL: https://issues.apache.org/jira/browse/YARN-5216
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Arun Suresh
> Assignee: Hitesh Sharma
> Attachments: YARN5216.001.patch, yarn5216.002.patch
>
>
> Currently, the default action taken by the QueuingContainerManager,
> introduced in YARN-2883, when a GUARANTEED Container is scheduled on an NM
> with OPPORTUNISTIC containers using up resources, is to KILL the running
> OPPORTUNISTIC containers.
> This JIRA proposes to expose a configurable hook to allow the NM to take a
> different action.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
