-
Bug
-
Resolution: Duplicate
-
Undefined
-
None
-
2.6.2
-
None
-
None
-
False
-
None
-
False
-
-
-
Moderate
Components:
- OCP 4.15.11
- OCV 4.15.2
- MTV Operator 2.6.2
While migrating a VM from RHV to OpenShift, scratch pvcs are left bound and in use as the importer pods are not cleaned up. Additionally one pvc is left in Lost state, as it is in https://issues.redhat.com/browse/MTV-1238 We note that the importer pod cleanup does not happen.
After migration:
$oc get pods NAMESPACE NAME READY STATUS RESTARTS AGE default importer-prime-9c4ca6df-b651-4ab6-8a11-96a2a7f656e9-checkpoint-4f5bcb18-7f69-4e2c-915c-d112ac0dd203 0/1 Completed 0 1h default importer-prime-9c4ca6df-b651-4ab6-8a11-96a2a7f656e9-checkpoint-8df7d2c4-4290-4d98-b5b8-a32577b0f55d 0/1 Completed 0 1h default virt-launcher-ovirt-testvm2-rkzdg 1/1 Running 0 1h
$ oc get pvc -n default
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE
prime-9c4ca6df-b651-4ab6-8a11-96a2a7f656e9 Lost pvc-6cfe9aa7-7c8d-4bc0-8f07-4117cd408838 0 ocs-storagecluster-ceph-rbd 1h
prime-9c4ca6df-b651-4ab6-8a11-96a2a7f656e9-scratch Bound pvc-dbedc23a-a8aa-4247-9952-e20cff859655 284058683598 RWO ocs-storagecluster-ceph-rbd 1h
testwarm-july2-84e9514e-bd7d-495f-8b72-4a67f0ea02e1-xwsrv Bound pvc-6cfe9aa7-7c8d-4bc0-8f07-4117cd408838 284058683598 RWX ocs-storagecluster-ceph-rbd 1h
This is seen on Netapp trident storage (customer), as well as ocs storage(in our lab). However it is not reproducible using nfs-subdir-external-provisioner(unsupported, but used in testing)
When the migration is done using the nfs provisioner, the importer pods are removed, the prime-XXX-scratch pvc is removed, and everything looks clean. In this case we see something like the following in the forklift-controller logs:
{"level":"info","ts":"2024-07-02 20:50:57.861","logger":"plan|xnhwb","msg":"Deleted importer pod.","plan":{"name":"shaggymigrated","namespace":"openshift-mtv"},"migration":"openshift-mtv/shaggymigrated-tfthd","pod":"openshift-mtv/
When warm migrating to Netapp or ODF storage on OpenShift, we don't see the "Deleted importer pod" messages.
While we know there is a manual work around to delete the importer pods manually, that's a lot of extra work especially when migrating many VMs at once. Additionally it appears to strand pvcs which consume space on the backend storage.
- duplicates
-
-
- Closed
-
