Comment 4 for bug 1807723

I think this is where that cinder attachment delete fails:

https://github.com/openstack/cinder/blob/9d4fd16bcd8eca930910798cc519cb5bc5846c59/cinder/volume/manager.py#L4528

Notice the FIXME in there:

https://github.com/openstack/cinder/blob/9d4fd16bcd8eca930910798cc519cb5bc5846c59/cinder/volume/manager.py#L4530

# FIXME(jdg): Obviously our volume object is going to need some
# changes to deal with multi-attach and figuring out how to
# represent a single failed attach out of multiple attachments

Because of that exception we don't call this:

            self.db.volume_detached(context.elevated(), vref.id,
attachment.get('id'))

Which would delete the volume attachment record in the Cinder DB, but I'm not sure how much that matters since on the nova side we update the volume1 BDM to point at a new attachment_id (769b75cc-4283-4b46-a975-243faae6e263).