[YUNIKORN-1070] Potential scheduler memory leak

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Duplicate
• Target Version: 1.0.0
• Assignee: Unassigned
• Reporter: Weiwei Yang
• Created: 07/Feb/22 22:14
• Updated: 13/Sep/22 14:33
• Resolved: 07/Feb/22 22:16

Description

Ben mentioned this in the slack, he runs 0.12.2 on EKS and runs into periodic OOM cases for the scheduler in EKS after a few days. Currently, the scheduler is configured for 10GB of memory and eventually always seems to run out of memory. In my environment, I have a lot of nodes coming in and out of the cluster due to autoscaling. Wondering if this could be a possible reason or if you guys have any other ideas. Let me know what kind of troubleshooting information might be useful here, but there is just a continuous growth of memory consumption that ends with OOMKilled.

Issue Links

• duplicates: YUNIKORN-876 Optimize scheduler memory consumption