hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arun Suresh (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (YARN-5978) ContainerScheduler and Container state machine changes to support ExecType update
Date Thu, 10 Aug 2017 15:22:02 GMT

    [ https://issues.apache.org/jira/browse/YARN-5978?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16121781#comment-16121781
] 

Arun Suresh edited comment on YARN-5978 at 8/10/17 3:21 PM:
------------------------------------------------------------

Thanks for the patch [~kartheek].

Couple of nits:
* I think you can rollback the changes to hadoop-yarn-server-common/pom.xml
* In {{TestAMRMClient}}, some of the changes like the one in line 994, when the timeout is
commented can be rolled-back.

Also, in {{ContainerShceduler::onUpdateContainer()}} method, you should add {{killOpportunisticContainers(updateEvent.getContainer())}}
before line 203 to ensure that any running opportunistic containers are killed to make room
for this promoted container. Which reminds me:
I think we need 1 more test-case. probably in {{TestContainerSchedulerQueuing}} - to test
the above.
Essentially, we should have a situation where we have a full NM with a bunch of Opportunistic
containers running and some Opp containers queued. Then the testcase should promote a queued
Opp container, we should verify that it starts running - and one/more of the running opportunistic
containers is/are killed to make room.


was (Author: asuresh):
Thanks for the patch [~kartheek].

Couple of nits:
* I think you can rollback the changes to hadoop-yarn-server-common/pom.xml
* In {{TestAMRMClient}}, some of the changes like the one in line 994, when the timeout is
commented can be rolled-back.

Also, in {{ContainerShceduler::onUpdateContainer()}} method, you should add {{killOpportunisticContainers()}}
before line 203 to ensure that any running opportunistic containers are killed to make room
for this promoted container. Which reminds me:
I think we need 1 more test-case. probably in {{TestContainerSchedulerQueuing}} - to test
the above.
Essentially, we should have a situation where we have a full NM with a bunch of Opportunistic
containers running and some Opp containers queued. Then the testcase should promote a queued
Opp container, we should verify that it starts running - and one/more of the running opportunistic
containers is/are killed to make room.

> ContainerScheduler and Container state machine changes to support ExecType update
> ---------------------------------------------------------------------------------
>
>                 Key: YARN-5978
>                 URL: https://issues.apache.org/jira/browse/YARN-5978
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Arun Suresh
>            Assignee: kartheek muthyala
>         Attachments: YARN-5978.001.patch, YARN-5978.002.patch
>
>
> ContainerScheduler should support updateContainer API for
> - Container Resource update
> - ExecType update that can change an opportunistic to guaranteed and vice-versa
> Adding a new ContainerState event, UpdateContainerStateEvent to support UPDATE_CONTAINER
call from RM.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org


Mime
View raw message