Paul Jones created YARN-10073:
---------------------------------
Summary: Intraqueue preemption doesn't work across partitions
Key: YARN-10073
URL: https://issues.apache.org/jira/browse/YARN-10073
Project: Hadoop YARN
Issue Type: Bug
Components: capacity scheduler, capacityscheduler, scheduler
preemption
Affects Versions: 2.8.5
Reporter: Paul Jones
Cluster:
1 Node with label "A"
yarn.scheduler.capacity.root.accessible-node-labels=*
yarn.resourcemanager.monitor.capacity.preemption.intra-queue-preemption.enabled=true
yarn.scheduler.capacity.root.default.minimum-user-limit-percent=50
User 1: Submit job Y require 10x cluster resources to queue, default, using
label ""
User 2: (after job Y starts) submit job Z to queue, default using label ""
What we see: Job Z doesn't start until job Y releases resources. This happens
because the pending requests for job Y and Z are in partition "". However,
queue default is using resources in partition "A". Pending requests in
partition "" don't cause intra queue preemptions in partition "A".
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
