-
Bug
-
Resolution: Done
-
Undefined
-
None
-
None
When the descheduler tries to evict a VM and the migration fails, it may select a different VM for eviction on next cycle. Meanwhile, virt-controller has logic to keep retrying failed migrations. This can lead to both components independently attempting to evacuate VMs and as result - some of migrations in Pending state because of the KubeVirt migration limits:
NAME PHASE VMI kubevirt-evacuation-5t7nh Failed vm-vm-descheduler-test-3-1750179204-5846608 kubevirt-evacuation-7sj8b Pending vm-vm-descheduler-test-1-1750179196-4016204 kubevirt-evacuation-8kdw4 Pending vm-vm-descheduler-test-3-1750179204-5846608 kubevirt-evacuation-f4pz2 Running vm-vm-descheduler-test-0-1750179192-571093 kubevirt-evacuation-ffprv Failed vm-vm-descheduler-test-5-1750179212-7171865 kubevirt-evacuation-snmp6 Failed vm-vm-descheduler-test-0-1750179192-571093 kubevirt-evacuation-tz7mv Running vm-vm-descheduler-test-5-1750179212-7171865
It may be worth exploring whether descheduler-initiated migrations should be marked differently or skipped from automatic retries by virt-controller to avoid such conflicts.
- duplicates
-
-
- Verified
-
