Description of problem: After CSI migration tests finish, there are some leftover PVs in the cluster. See https://testgrid.k8s.io/redhat-openshift-ocp-release-4.8-broken#periodic-ci-openshift-release-master-nightly-4.8-e2e-aws-csi-migration From https://prow.ci.openshift.org/view/gs/origin-ci-test/logs/periodic-ci-openshift-release-master-nightly-4.8-e2e-aws-csi-migration/1391694972192821248: +persistentvolume/local-pv9zb5m +persistentvolume/local-pvpt5mt persistentvolume/pvc-e3b4f183-4c75-4d9c-9385-18b088abcd1b persistentvolume/pvc-f15067db-d6d5-4c94-9079-d28bc37ed925 +persistentvolume/pvc-f9c9f7d0-51b8-402a-80f8-4bbc1982a6d0 ERROR: Timed out waiting for PVs to get deleted. ERROR: It seems that some test left some PVs behind. ERROR: Check the diff between expected and existing PVs above. So we have three PVs left behind and not deleted. Two of them are local, the third one looks like mock: name: pvc-f9c9f7d0-51b8-402a-80f8-4bbc1982a6d0 ... csi: driver: csi-mock-e2e-csi-mock-volumes-7168 ... phase: Released (from https://gcsweb-ci.apps.ci.l2s4.p1.openshiftapps.com/gcs/origin-ci-test/logs/periodic-ci-openshift-release-master-nightly-4.8-e2e-aws-csi-migration/1391694972192821248/artifacts/e2e-aws-csi-migration/storage-pv-check/artifacts/pvs-29.yaml) How reproducible: Always Steps to Reproduce: Run openshift conformance/parallel tests against OCP on AWS. CSI migration is enabled in the linked CI jobs, IMO it should not affect local or mock volumes, but one never knows. We do not check for leftover PVs in other e2e jobs, so we don't know if there are some on GCE or without migration. Actual results: At least two local PVs left behind (this is 100% reproducible) Sometimes, a CSI mock PV left behind (started appearing in https://prow.ci.openshift.org/view/gs/origin-ci-test/logs/periodic-ci-openshift-release-master-nightly-4.8-e2e-aws-csi-migration/1391332579759624192 ???) Expected results: No PVs left behind
Just a minor update on this: both the local volumes have "persistentVolumeReclaimPolicy: Retain". The mock volume is being used by a snapshot test. Logs from the snapshot controller: createSnapshotContent: Creating content for snapshot e2e-csi-mock-volumes-7168/snapshot-j2zf8 through the plugin ... Added protection finalizer to persistent volume claim e2e-csi-mock-volumes-7168/snapshot-test-pvc Keeping PVC e2e-csi-mock-volumes-7168/snapshot-test-pvc, it is used by snapshot e2e-csi-mock-volumes-7168/snapshot-j2zf8 ... checkandRemovePVCFinalizer[snapshot-j2zf8]: Remove Finalizer for PVC snapshot-test-pvc as it is not used by snapshots in creation ... cannot get claim from snapshot [snapshot-9jkcn]: [failed to retrieve PVC snapshot-test-pvc from the lister: "persistentvolumeclaim \"snapshot-test-pvc\" not found"] Claim may be deleted already I don't know yet how is the snapshot creation and PV protection interacting with the CSI migration.
The local volumes come from these two tests: https://github.com/kubernetes/kubernetes/blob/master/test/e2e/storage/persistent_volumes-local.go#L375-L381. They look to create two local volumes in the ginkgo.BeforeEach that get cleaned up in ginkgo.AfterEach, but then the test funciton itself (https://github.com/kubernetes/kubernetes/blob/7229ea343dd649f9a6c20fa1fd6b13e602f3f082/test/e2e/storage/persistent_volumes-local.go#L719) creates also a volume and that one seems to be never removed.
Local volumes leak upstream fix: https://github.com/kubernetes/kubernetes/pull/102292
*** Bug 1948474 has been marked as a duplicate of this bug. ***
This got merged to 4.9 with rebase to 1.22.