hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sunil G (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (YARN-2113) Add cross-user preemption within CapacityScheduler's leaf-queue
Date Tue, 16 May 2017 19:07:04 GMT

     [ https://issues.apache.org/jira/browse/YARN-2113?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Sunil G updated YARN-2113:
    Attachment: YARN-2113.0018.patch

Thanks [~eepayne] for investigating this.
{{when(cs.getResourceCalculator()).thenReturn(rc);}} is needed to be added. However simply
adding this still passes all my tests.

I think you have made some more changes in the tests such as increasing cluster resource to
a higher vcores. This has impacted the preemptionLimit value (usually 10% or 20% of cluster
resource) to give more dominance over vcores.  I could see that preemption is happening till
vcores are satisfied (thought memory is 0). However this is not needed.

As mentioned earlier, this issue is know in YARN-6538. I was trying to fix as part of that
to add a new api in {{Resources}} class. Usually we will have a resource object and will deduct
resources from it in every loop. At the start of loop, {{Resources.lessThanOrEqual}} will
be checked to see whether resource has gone under {{Resources.none}}

For eg:

    // assign all cluster resources until no more demand, or no resources are
    // left
    while (!orderedByNeed.isEmpty() && Resources.greaterThan(rc, totGuarant,
        unassigned, Resources.none())) {
Here {{unassigned}} could be like *0 memory and 10 vcores*. However {{Resources.greaterThan}}
does work with dominance and loop will continue.

I have added a new api named {{Resources.isAnyResourceZero}} to ensure that loop will not
continue if any resource is zero in DRF. I ll extend same in YARN-6538 as well.

> Add cross-user preemption within CapacityScheduler's leaf-queue
> ---------------------------------------------------------------
>                 Key: YARN-2113
>                 URL: https://issues.apache.org/jira/browse/YARN-2113
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: scheduler
>            Reporter: Vinod Kumar Vavilapalli
>            Assignee: Sunil G
>         Attachments: IntraQueue Preemption-Impact Analysis.pdf, TestNoIntraQueuePreemptionIfBelowUserLimitAndDifferentPrioritiesWithExtraUsers.txt,
YARN-2113.0001.patch, YARN-2113.0002.patch, YARN-2113.0003.patch, YARN-2113.0004.patch, YARN-2113.0005.patch,
YARN-2113.0006.patch, YARN-2113.0007.patch, YARN-2113.0008.patch, YARN-2113.0009.patch, YARN-2113.0010.patch,
YARN-2113.0011.patch, YARN-2113.0012.patch, YARN-2113.0013.patch, YARN-2113.0014.patch, YARN-2113.0015.patch,
YARN-2113.0016.patch, YARN-2113.0017.patch, YARN-2113.0018.patch, YARN-2113.apply.onto.0012.ericp.patch,
YARN-2113 Intra-QueuePreemption Behavior.pdf, YARN-2113.v0.patch
> Preemption today only works across queues and moves around resources across queues per
demand and usage. We should also have user-level preemption within a queue, to balance capacity
across users in a predictable manner.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org

View raw message