Fedora Account System
Red Hat Associate
Red Hat Customer
Description of problem: If DVM controller rsync daemon pods are not cleaned up properly, rollback will not delete them due to missing correlation labels, but it _will_ wait for the Pods to be terminated (which will never happen). Version-Release number of selected component (if applicable): 1.4.0 How reproducible: Saw this once. Steps to Reproduce: 1. Kill DVM in middle of migration (in my case, was triggered by an out of date operator not providing the "RSYNC_PRIVILEGED" env var 2. Migration will fail with DVM errors 3. Subsequent rollback should stall Expected results: DVM pods have correlation labels attached and will get removed by rollback
I believe the Target Release should 1.4.z.
Just in case you need another way to reproduce the issue. I hit this issue and I could reproduce it forcing an OutOfcpu failure like this: Steps to Reproduce: 1. Configure limits/requests for the rsync pod in the source cluster high enough to trigger an OutOfcpu error. $ oc edit migrationcontroller -n openshift-migration ..... spec: ..... source_rsync_pod_cpu_limits: 10 source_rsync_pod_cpu_requests: 9 2. Migrate any application using a PVC. Use Direct Volume Migration 3. The migration will report a warning saying that it could not migrate the PVCs 4. Rollback the migration
*** This bug has been marked as a duplicate of bug 1920559 ***
Fix attached to duplicate https://bugzilla.redhat.com/show_bug.cgi?id=1920559