Vsphere seed: Rebooting nodes requires manual detaching of volumes

[alvaroaleman]

When using a vsphere seed and rebooting a node that runs one or more pods with attached PVs, that node can not be started anymore because the cloud provider does not remove the volume binding from the old node even thought the pod will get rescheduled to another node.

This then means that the vsphere instance can not be started anymore until an operator manually removes the binding to the node inside vsphere.

Upstream issue: kubernetes/kubernetes#63577

A possible quickfix would be to implement a systemd unit that taints and drains nodes prior to rebooting them and removes the taint on startup.

[mrIncompetent]

Looking at kubernetes/kubernetes#63413 (comment) we might just simply wait for 1.12 to land?

[alvaroaleman]

Yes, we should potentially communicate this for the time being to Kubermativ + Vsphere users and leave this issue open as internal tracking issue but not act on it.

[mrIncompetent]

Closed in favor of kubermatic/docs#44