Aditya Maheshwari created YUNIKORN-3284:
-------------------------------------------
Summary: Quota Preemption check quota preemption incorrect victim
selection
Key: YUNIKORN-3284
URL: https://issues.apache.org/jira/browse/YUNIKORN-3284
Project: Apache YuniKorn
Issue Type: Task
Reporter: Aditya Maheshwari
Assignee: Aditya Maheshwari
Test steps - # Create a queue {{queue-a}} with max resource limit of 300 and
submit an application {{app-a}} with 3 pods to this queue. All pods should come
up fine.
# Update quota of queue-a and set max resource to 250 now (from 300 earlier)
with delay of 30 seconds and quotaPreemption enabled.
# Before the delay expires, submit another application {{app-b}} with resource
50 to {{queue-a}} . As expected {{app-b}} becomes pending as the queue is full.
# After delay expires, preemption triggers, but instead of pod of
{{{}app-a{}}}, pod of {{app-b}} is chosen as victim, which is already in
pending state. I think this is done because {{app-b}}
is of 50 and the resource went down by 50 (300->250) in step 2.
So step 4 is not working as expected.
In the end all 3 pods of app-a are running consuming 300, and in configs max
resource limit is still 250 set from step 2. This is inconsistent step.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
