Bumping my question. Any hints would be highly appreciated! My colleague also created https://github.com/kubernetes/cloud-provider-openstack/issues/2553 but there hasn't been any activity yet.

Zitat von Eugen Block eblock@nde.ag:

Hi,

we have a couple of k8s clusters in our openstack environment (Victoria) and are facing some problems when pods get moved to different worker nodes. The PVCs get stuck in "reserved" state, and they only get attached after resetting the volumes to "available" state. First I thought it was due to the missing multiattach volume type which seems to work fine in our customer's cluster. I also found the "duplicate connector" error messages in the nova-compute.log:

...

Unable to update attachment.(Invalid volume: duplicate connectors detected on volume

But after adding the multiattach volume type and retyping the affected volumes, the PVCs still get stuck in "reserved" state while I don't see any error messages on the hypervisors anymore. So it seems that something has changed/improved, but now I don't have a trace where to look next. Could this be a cinder-csi driver issue or something? Our cinder backend is ceph (Pacific).

Not sure which information could help so please ask for anything I can provide. Any pointers are highly appreciated!

Thanks! Eugen