[
https://issues.apache.org/jira/browse/YARN-5216?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15738525#comment-15738525
]
Arun Suresh commented on YARN-5216:
-----------------------------------
Given the fact that the actual container executor executable binary that is
deployed on the cluster can be decided by the cluster admin and the admin can
choose to deploy an executor + runtime combo that supports pausing/freezing or
not, I am wondering if a conf/knob to determine if the scheduler should
pause/kill a container might seem superfluous in that case.
bq. What we are looking for is preemption of opp. container to schedule a
guaranteed one and I feel that's best captured in the scheduler state.
I agree, and im not debating if we should capture the paused state of the
container in the scheduler and the NM container state (we definitely should). I
am just saying, when the scheduler decides that there are no resources to run a
guaranteed container AND if there are currently running opportunistic
containers... instead of checking a conf to see if it can kill/pause, it should
just ask the executor to preempt; then based on the capability of the executor
and target system, the container can be either killed or paused and the
Scheduler will be notified via either a CONTAINER_KILLED or CONTAINER_PAUSED.
if the executor was able to pause the container, the Scheduler puts it in a
separate queue which will be polled when there are enough resources to resume.
Would like to hear [~vvasudev] / [[email protected]] / [~kasha] /
[~jianhe] 's thoughts on this too.
> 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: Sub-task
> Components: distributed-scheduling
> Reporter: Arun Suresh
> Assignee: Hitesh Sharma
> Labels: oct16-hard
> Attachments: YARN-5216-YARN-5972.001.patch, 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]
