Sometimes when Compute Server is rebooted suddenly or in another similar situation, we want to evacuate instances to another host we get this error from Cinder.
Our Cinder's backed is Ceph
{'code': 400, 'created': '2022-09-26T15:34:22Z', 'message': 'Unable to update attachment.(Invalid volume: duplicate connectors detected on volume ....). (HTTP 500) (Request-ID:req-c....)'}
I want to migrate or evacuate the instance but we get this error. How can we solve this?
This behaviour is a safeguard against data corruption by mounting the same volume in 2 instances, as explained here:
https://opendev.org/openstack/cinder/commit/eadff0df7efbe662bf107f77e1b2ac96c912e4e0
If multiple attachments are requested simultaneously, a volume can be attached to an instance more than once.
This fixes this problem by detecting that a race has occurred in _attachment_reserve, and backing out one of the invalid attachment records.
This means that if the attachment API is called repeatedly and quickly for the same volume, some requests may fail, but this is much better than incorrectly creating multiple attachments.
I would try to dettach and re-attach the volume to the server to workaround this (openstack server remove volume and openstack server remove volume)
