[
https://issues.apache.org/jira/browse/YARN-5532?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15435799#comment-15435799
]
Vinod Kumar Vavilapalli commented on YARN-5532:
-----------------------------------------------
Tx for your comments, [~asuresh].
bq. If you do make Resource Localization independent of containers, against who
would the resources (mem, iobandwidth, storage etc) be accounted against ?
Today we only have ref-counting and no resource-usage accounting, but the
principles will be the same.
I think {{ResourceLocalizationService}} should completely abstract away the
requester and just manage everything via a RequesterID - resource usage per
requester, notifying a requester when a resource is successfully localized,
fails etc.
bq. Will this be a service exposed to any container running in the NM?
That would be a use-case and API discussion. Today, the two use-cases we have -
YARN-1503 and YARN-4876 / YARN-4726 - don't need any container-> NM
interaction. The only interaction is with AMs.
If new use-cases come up which need container->NM interaction, we can design as
part of implementing those use-cases.
> Make ResourceLocalizationService inside NodeManager independent of containers
> -----------------------------------------------------------------------------
>
> Key: YARN-5532
> URL: https://issues.apache.org/jira/browse/YARN-5532
> Project: Hadoop YARN
> Issue Type: New Feature
> Reporter: Jian He
> Assignee: Jian He
>
> The goal is to enable NodeManagers to localize resources while container is
> running. Today, resource-localization is always the first step before
> starting a container. It will be useful if YARN can localize the resources
> continuously even while container is running.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
