Comment 3 for bug 1526989

From logstash over the last 7 days:

http://logstash.openstack.org/#dashboard/file/logstash.json?query=message%3A%5C%22_create_domain%5C%22%20AND%20message%3A%5C%22libvirtError%3A%20internal%20error%3A%20early%20end%20of%20file%20from%20monitor%3A%20possible%20problem%5C%22%20AND%20message%3A%5C%22qemu%3A%20terminating%20on%20signal%2015%20from%20pid%5C%22%20AND%20tags%3A%5C%22screen-n-cpu.txt%5C%22&from=7d

This only fails in multinode jobs where we set an exact cpu model match between the guest and host. I'm not sure why those tests aren't failing and the jobs are reporting success though.

But I noticed this from one of the failing snapshots:

http://logs.openstack.org/34/283134/6/check/gate-tempest-dsvm-neutron-dvr-multinode-full/a1ab6f2/logs/subnode-2/screen-n-cpu.txt.gz?level=TRACE#_2016-03-05_11_39_37_887

<cpu mode='custom' match='exact'>
    <model fallback='allow'>gate64</model>
    <topology sockets='1' cores='1' threads='1'/>
  </cpu>

I think the multinode jobs are the only ones where we require a specific cpu model and that's for live migration to work, i.e.:

http://logs.openstack.org/34/283134/6/check/gate-tempest-dsvm-neutron-dvr-multinode-full/a1ab6f2/logs/subnode-2/etc/nova/nova.conf.txt.gz

[libvirt]
cpu_model = gate64
inject_partition = -2
live_migration_uri = qemu+ssh://stack@%s/system
use_usb_tablet = False
cpu_mode = custom
virt_type = kvm

I was wondering about cpu modes between the host and guest because I saw this related forum:

https://bbs.archlinux.org/viewtopic.php?id=159279

We should get Daniel Berrange (danpb in #openstack-nova in freenode IRC) to take a look at this.