Comment 2 for bug 1491276

No, it's not Active-Active.
Shared storage means the volumes are available on all nodes but active only on one. We use our own cinder driver to enable this, so, for example, we can do live-migration without having to copy the volume over from one node to another.

We try to do the same thing with evacuate, since the volume is available on any node it should be easy to just rebuild the instance and start it.

PS. Looking closer at the chain of events, this looks more like an error in nova (n-cpu called c-api to terminate_connection, although the evacuate command states - Evacuate server from failed host. - so it shouldn't try to connect to the failed host). Not sure though which component is to blame here.