[
https://issues.apache.org/jira/browse/YARN-4902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15410270#comment-15410270
]
Konstantinos Karanasos commented on YARN-4902:
----------------------------------------------
bq. LRA planning looks like an implementation.
LRA planning is much more than an implementation. Think of it as planning
multiple applications at once. This is something that the scheduler cannot do,
no matter what its implementation is.
Please give a look at YARN-1051 to see a similar use case for
planning/admission control but in a constraint-free context.
I can give more details as I update the document. In any case, that does not
block any of the changes that are required in the scheduler per se to support
constraints.
bq. For cardinality, could you share a more detailed use case for that?
As you mention, an example would be to limit the number of hbase-masters in a
node/rack or even the number of AMs in a node.
You could do it with resource isolation, but especially network isolation is
really hard to get right, so until we reach that point, I think it would be
great for applications to be able to express such constraints.
bq. It seems to me that cardinality is a special case of anti-affinity.
I would say that it is the other way around: affinity and anti-affinity is a
special case of cardinality. If you say there is cardinality 1 for that node,
it means you have anti-affinity for that node.
I agree that you can currently express it with your proposal, so we are just
suggesting an alternative way that would be more succinct and we will not need
to have different types of constraints, but just a single one.
> [Umbrella] Generalized and unified scheduling-strategies in YARN
> ----------------------------------------------------------------
>
> Key: YARN-4902
> URL: https://issues.apache.org/jira/browse/YARN-4902
> Project: Hadoop YARN
> Issue Type: New Feature
> Reporter: Vinod Kumar Vavilapalli
> Assignee: Wangda Tan
> Attachments: Generalized and unified scheduling-strategies in YARN
> -v0.pdf, LRA-scheduling-design.v0.pdf, YARN-5468.prototype.patch
>
>
> Apache Hadoop YARN's ResourceRequest mechanism is the core part of the YARN's
> scheduling API for applications to use. The ResourceRequest mechanism is a
> powerful API for applications (specifically ApplicationMasters) to indicate
> to YARN what size of containers are needed, and where in the cluster etc.
> However a host of new feature requirements are making the API increasingly
> more and more complex and difficult to understand by users and making it very
> complicated to implement within the code-base.
> This JIRA aims to generalize and unify all such scheduling-strategies in YARN.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
