Hedger Lai created YUNIKORN-3458:
------------------------------------
Summary: Premature victim loop termination in
calculateVictimsByNode starves preemption
Key: YUNIKORN-3458
URL: https://issues.apache.org/jira/browse/YUNIKORN-3458
Project: Apache YuniKorn
Issue Type: Bug
Components: core - scheduler
Reporter: Hedger Lai
Assignee: Hedger Lai
Fix For: 1.6.0
In {{calculateVictimsByNode()}} ({{preemption.go}}), when a candidate victim
exceeds {{askQueueRemaining}}, the loop executes {{break}}:
{code:go}
if askQueueRemaining != nil &&
askQueueRemaining.FitInActual(victim.GetAllocatedResource()) {
askQueue.AddAllocation(victim.GetAllocatedResource())
} else {
queueSnapshot.AddAllocation(victim.GetAllocatedResource())
break
}
{code}
This prematurely aborts victim evaluation for the node, ignoring subsequent
viable victims and causing preemption to fail.
h3. Proposed Fix
Change {{break}} to {{continue}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
