[
https://issues.apache.org/jira/browse/YARN-1317?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13798175#comment-13798175
]
Vinod Kumar Vavilapalli commented on YARN-1317:
-----------------------------------------------
Here are the specific details:
- Today we force each scheduler to update QueueMetrics. Schedulers only deal
with AppAttempts today (YARN-1311), so QueueMetrics sometimes aren't updated
correctly (YARN-1166).
- Every scheduler does the same ACL checks when we don't need to. There are
inconsistencies in the way QueueACLs are handled (YARN-1288) which can be
completely done away with once we have ACLs checked by a top level
QueueACLsManager.
- Queue Configuration itself is wildly scheduler specific. We reconciled that
during 0.21 via MAPREDUCE-861.
-- This is a much larger effort with compatibility implications, but this is
something that we need to think about may be separately
-- One other side benefit of MAPREDUCE-861 is that hierarchical queues are so
much easier to configure in a hierarchical conf file.
- Managing apps by queues(YARN-807). Every scheduler essentially manages the
same detail. A top level view manipulated by individual schedules eases this
pain.
> Make Queue, QueueACLs and QueueMetrics first class citizens in YARN
> -------------------------------------------------------------------
>
> Key: YARN-1317
> URL: https://issues.apache.org/jira/browse/YARN-1317
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: Vinod Kumar Vavilapalli
> Assignee: Vinod Kumar Vavilapalli
>
> Today, we are duplicating the exact same code in all the schedulers. Queue is
> a top class concept - clientService, web-services etc already recognize queue
> as a top level concept.
> We need to move Queue, QueueMetrics and QueueACLs to be top level.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
