What happened:

When kube-controller-manager process in terminated on a leader node, a new leader election starts. This will trigger DetachVolume.Detach for all persistent volumes in the cluster. In our case, we use Cinder CSI driver, the volumes are detached in cinder, but the volumeattachmens are not deleted/updated, so this requires manual intervention to clean up the statefulsets/volumeattachments.

What you expected to happen:

The volumes should not be detached.

How to reproduce it (as minimally and precisely as possible):

Terminate kube-controller-manager on the leader node.

kill -s SIGHUP $(pidof kube-controller-manager)

Anything else we need to know?:

Cinder CSI driver enabled.

Environment:

• Kubernetes version: 1.19.9
• Cloud provider or hardware configuration: Openstack Cinder
• OS (e.g: cat /etc/os-release): Ubuntu 18.04.4 LTS (Bionic Beaver)"