[YUNIKORN-946] Accounting resources for deleted executor pods

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.11
• Fix Version/s: 0.12.1
• Component/s: core - scheduler
• Target Version: 1.0.0
• Assignee: Anuraag Nalluri
• Reporter: Ashutosh Singh
• Created: 17/Nov/21 07:18
• Updated: 17/Feb/22 15:29
• Resolved: 17/Feb/22 09:15

Description

Even when executors are deleted, YK UI shows that resources are consumed by the pod (the one which is already deleted). kubectl get pods  does not show the executor but YK UI shows the information of a deleted pod consuming resources even after few hours. 

It results into leaking cluster resources.

Steps:

1. Run a spark application using k8s spark operator
2. Wait for executors to be in running state.
3. Delete the application using `kubectl delete sparkapplications <appName>` OR `kubectl delete {}f <yaml{-}file>`
4. All the driver and executor pods would be deleted. check `kubectl get pods`
5. However, YK UI still shows some of the executors running and consuming resources. It leads to leak of the resource as they are considered as used and could not be used by pending pods.

More details: 

https://yunikornworkspace.slack.com/archives/CLNUW68MU/p1637126093006900

 

Issue Links

• is caused by: YUNIKORN-766 driver/executor get hang occasionally if the applicationId is defined