[
https://issues.apache.org/jira/browse/YARN-7258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16191701#comment-16191701
]
Arun Suresh commented on YARN-7258:
-----------------------------------
Thanks for the patch [~kartheek]
Some comments:
* It looks generally good. We might need to add some more tests - to
differentiate between when we have requests with numcontainers > 1 and those
with numcontainers == 1.
* Also, looks like we might hit a Concurrent modification exception, when we
remove the scheduler keys from the outstanding opportunistic requests.
> Add Node and Rack Hints to Opportunistic Scheduler
> --------------------------------------------------
>
> Key: YARN-7258
> URL: https://issues.apache.org/jira/browse/YARN-7258
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Arun Suresh
> Assignee: kartheek muthyala
> Attachments: YARN-7258.001.patch
>
>
> Currently, the Opportunistic Scheduler ignores the node and rack information
> and allocates strictly on the least loaded node (based on queue length) at
> the time it received the request. This JIRA is to track changes needed to
> allow the OpportunisticContainerAllocator to take the node/rack name as hints.
> The flow would be:
> # If requested node found in the top K leastLoaded nodes, allocate on that
> node
> # Else, allocate on least loaded node on the same rack from the top K least
> Loaded nodes.
> # Else, allocate on least loaded node.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
