[
https://issues.apache.org/jira/browse/YARN-624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13732188#comment-13732188
]
Steve Loughran commented on YARN-624:
-------------------------------------
[~bikassaha] has pointed out that gang scheduling can be implemented in an AM
today: it can just hang on to assigned nodes until a minimum has been
allocated, at which point it can bring up its service.
This would move all scheduling decisions into the AM, which is free to
implement it own policy.
If we start this way, then once a few apps have done this we can look at
commonality and decide what features -if any- need to go into YARN.
The most likely - again, credit to Bikas - is for YARN to recognise that an AM
has been given some containers but not, after a time period, deployed anything
to them. YARN could then cancel the lease unless the AM specifically indicates
it wants to retain it.
> Support gang scheduling in the AM RM protocol
> ---------------------------------------------
>
> Key: YARN-624
> URL: https://issues.apache.org/jira/browse/YARN-624
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: api, scheduler
> Affects Versions: 2.0.4-alpha
> Reporter: Sandy Ryza
> Assignee: Sandy Ryza
>
> Per discussion on YARN-392 and elsewhere, gang scheduling, in which a
> scheduler runs a set of tasks when they can all be run at the same time,
> would be a useful feature for YARN schedulers to support.
> Currently, AMs can approximate this by holding on to containers until they
> get all the ones they need. However, this lends itself to deadlocks when
> different AMs are waiting on the same containers.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
