[
https://issues.apache.org/jira/browse/YARN-6210?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15877010#comment-15877010
]
Daniel Templeton commented on YARN-6210:
----------------------------------------
Thanks for the patch, [~kasha]. A few comments:
* {{ResourceCalculator.compare()}}: {{singleType}} needs to be explained
better. I still don't know what it means.
* {{TestFailSchedulerPreemption.writeAllocFile()}}: should _preemptable_ maybe
be _preemptable-1_?
* {{FSAppAttempt.isStarvedForFairShare()}}: No likey the "0 > ..." There's no
reason for the weird inverted order, and _less than x_ is easier to logic about.
* {{FSAppAttempt.isStarved()}}: looks like it can be private...
* {{FSAppAttempt.fairShareStarvation()}}: the meaning of _starved_ changed. It
would be nice to document in a comment why it's OK that _starved_ doesn't
account for the threshold whereas the return value does.
* {{TestFairScheduler.testReservationWhileMultiplePriorities()}}: The comment
{{Create one application and take up all resources}} isn't actually true, is it?
* {{TestFairScheduler.testReservationWhileMultiplePriorities()}}: Looks like
you took out the tests that the scheduler resources are as expected at the
various steps and the test that the reservation is still for the lower priority
request. Not critical, but I would leave them in if it were me.
* {{TestFairScheduler.testReservationsStrictLocality}}: I like messages in my
asserts.
> FS: Node reservations can interfere with preemption
> ---------------------------------------------------
>
> Key: YARN-6210
> URL: https://issues.apache.org/jira/browse/YARN-6210
> Project: Hadoop YARN
> Issue Type: Bug
> Components: fairscheduler
> Affects Versions: 2.9.0
> Reporter: Karthik Kambatla
> Assignee: Karthik Kambatla
> Attachments: YARN-6210.1.patch, YARN-6210.2.patch
>
>
> Today, on a saturated cluster, apps with pending demand reserve nodes. A new
> app might not be able to preempt resources because these nodes are already
> reserved. This can be reproduced by the example in YARN-6151.
> Since node reservations are to prevent starvation of apps requesting large
> containers, triggering these reservations only on starved applications would
> avoid this situation.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
