Hi Michal,
We had similar issue in our kolla-ansible deployment. In our case we were not able to attach volume to the VM in 1 from 5 cases and it sometimes caused also inconsistencies in DB tables. We are also running Victoria release with Ceph backend. We found out that there were many disconnects/reconnects between nova and rabbitmq because of missing heartbeats which happened every 3 minutes. If attachment of volume happened in the time of reconnection, attachment was unsuccessful. What helped in our case was adding following lines to nova.conf:
[oslo_messaging_rabbit]
heartbeat_timeout_threshold = 0
In your case I would check cinder logs files for messaging errors whether you do not see regular reconnects to rabbitmq.
Kamil Madac