[
https://issues.apache.org/jira/browse/YUNIKORN-404?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17194549#comment-17194549
]
Weiwei Yang commented on YUNIKORN-404:
--------------------------------------
I have spent some time to debug this issue, this is indeed a bug, nice catch
[~kmarton].
This is actually due to the queue was named as upper case "X", internally we
are converting all upper case letters in the queue to lower cases, see code
[here|https://github.com/apache/incubator-yunikorn-core/blob/c2699e725c8a0f89fbcd7618908ad4cf712227e9/pkg/cache/queue_info.go#L70].
When the scheduler tries to update the queues, it got confused because it
thinks it visited "root.x", but not "root.X". So it marks the queue "root.X"
for removal.
> Config update breaks when there is upper case queue names
> ---------------------------------------------------------
>
> Key: YUNIKORN-404
> URL: https://issues.apache.org/jira/browse/YUNIKORN-404
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - common
> Reporter: Kinga Marton
> Assignee: Weiwei Yang
> Priority: Major
>
> While updating the configuration there are cases when queues are removed
> without any reason.
> Steps to reproduce: start YK with the following configuration:
> {code:yaml}
> configuration: |
> partitions:
> -
> name: default
> queues:
> -
> name: root
> submitacl: '*'
> queues:
> -
> name: someQueue
> resources:
> guaranteed:
> memory: 50000
> vcore: 5000
> max:
> memory: 80000
> vcore: 8000{code}
> Update the configuration to the following one:
> {code:yaml}
> partitions:
> -
> name: default
> queues:
> -
> name: root
> properties:
> application.sort.policy: stateaware
> submitacl: '*'
> queues:
> - name: a
> queues:
> - name: b
> queues:
> - name: c
> - name: X
> queues:
> - name: riskteam
> resources:
> guaranteed:
> gpu: "0"
> memory: "256"
> vcore: "1"
> max:
> gpu: "0"
> memory: "512"
> vcore: "2"
> placementrules:
> - name: tag
> create: true
> value: namespace{code}
> *Expected result*: someQueue should be deleted, a.b.c. and X.riskteam should
> be created
> *Actual result:* someQueue is deleted, a.b.c is created, X.riskteam is
> created and right after then is is marked for removal.
> *Note:* is happening both through API and with direct configmap update
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
