Multinode jobs failing on libvirt issues

[Ben Nemec]

Opening a bug to track this since it's causing a large number of failures in the multinode jobs and I don't see an existing bug for it.

There are two issues I'm aware of here, and I'm not sure whether they're related:
1) An error message about an unsupported "arat" flag in the nova-compute logs.
2) A libvirt segfault (search for "segfault" in /var/log/messages on subnode-2 to check for this)

These may present as an error during the ping test where the cinder volume is in-use instead of available. I suspect it has to do with Nova retrying the failed vm but the volume not being detached first. In any case, the volume error appears to be a symptom, not the cause.

I've seen multiple failures caused by both over the past couple of days and it's basically blocking everything from merging because these jobs are gating.

[Ben Nemec (bnemec)]

This may be fixed by https://review.openstack.org/#/c/410359/ We'll have to keep an eye on the jobs once that merges.