Comment 8 for bug 1888665

Yes, I have haproxy, but i think that's not the point. Devs shouldn't assume that cinder api is aware of failure. I had disaster last week when something hapeened wrong with network and openstack had bad contact with Ceph. During that failure cinder-volume was flapping (ceph volume service upping and downing continously) and compute nodes also were not able to proper handling attach/detach requests. Attaching - minor problem, but detaching - it was disaster, because some detaches went ok, some not, almost every detaching stucked in detaching state, regardless of actual status of the task. There is Openshift on the Openstack, so many attach/detach requests occuring every time. So after that I had inconsistent state of volumes, which I had to solve since today. It won't be happend if detaching error handling would be hendled in proper way.