Comment 7 for bug 1311533

Confirmed, I've been seeing the same, https://review.openstack.org/#/c/86638/ appears to have caused a regression where OS::Cinder::VolumeAttachment appears to work, but the volume is not really visible to the instance at all.

Using nova volume-attach on the CLI works fine, so reverting to the nova API seems like the best approach, provided we can avoid reintroducing the gate race which moving to the cinder calls was supposedly fixing ref bug #1298350

Can anyone provide any further details on why the cinder volume attach/detach API doesn't actually work?

I noticed it's not documented and only partially tested:

https://bugs.launchpad.net/cinder/+bug/1311733

FYI I'm currently working on a tempest test which when merged will cause us to fail the gate for a regression like this in future (I'll link the review here when I post it).