[ https://issues.apache.org/jira/browse/YARN-2113?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15989599#comment-15989599 ]
Wangda Tan commented on YARN-2113: ---------------------------------- [~eepayne]/[~sunilg] here's my understanding of the two policies: 1) Priority first: App1 can preempt App2 if: When App1.priority > App2.priority - User of app1 not reached user-limit When App1.priority = App2.priority - app1.user != app2.user - User of app1 not reached user-limit - User of app2 beyond user-limit. 2) User limit first: App1 can preempt app2 if: - app1.user != app2.user - User of app1 not reached user-limit - User of app2 beyond user-limit. For both: - No matter if {{app1.user}} is outside of {{#100/MULP}} or not, no change to preemption behavior. (To the example in https://issues.apache.org/jira/browse/YARN-2113?focusedCommentId=15951222&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15951222, if policy is PRIORITY_FIRST, preemption is allowed but if policy is USERLIMIT_FIRST, preemption is not allowed). If I understand Jason's comment correctly: https://issues.apache.org/jira/browse/YARN-2113?focusedCommentId=15951538&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15951538, above behavior should support both scenarios via different policies: USERLIMIT_FIRST don't allow any preemption to an user who is below user limit; PRIORITY_FIRST allows preemption happen if one app has higher priority than another. > 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.apply.onto.0012.ericp.patch, > 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 (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