Excessive warnings about snapshotting a non-running instance

[Matt Riedemann]

We see this warning from the compute manager in CI runs:

https://github.com/openstack/nova/blob/7b96206699ac28f807676bd08c6dee7a89bcb77c/nova/compute/manager.py#L3340

http://logs.openstack.org/20/541420/2/check/tempest-full/a416830/controller/logs/screen-n-cpu.txt.gz?level=WARNING#_Mar_22_16_45_28_061143

Mar 22 16:45:28.061143 ubuntu-xenial-ovh-bhs1-0003133028 nova-compute[13877]: WARNING nova.compute.manager [None req-ee41217e-bf1a-4622-aed5-f872fc772d5f tempest-ImagesTestJSON-641968735 tempest-ImagesTestJSON-641968735] [instance: 6a3e1cb2-63ff-4514-aa32-c5c4c73f84d8] trying to snapshot a non-running instance: (state: 3 expected: 1)

state=3 is PAUSED and 1=RUNNING.

And this:

Mar 22 16:45:51.800064 ubuntu-xenial-ovh-bhs1-0003133028 nova-compute[13877]: WARNING nova.compute.manager [None req-d9c5694c-899c-4a57-8318-043df137d564 tempest-ImagesTestJSON-641968735 tempest-ImagesTestJSON-641968735] [instance: 3b7589a5-ce3b-4f82-a43a-a48497de9382] trying to snapshot a non-running instance: (state: 4 expected: 1)

state=4 is SHUTDOWN.

Maybe this is related to bug 1741667 which for older versions of libvirt, trying to snapshot a PAUSED instance would hang:

https://review.openstack.org/#/c/532214/

If you look at that patch, it's specifically about doing *live* snapshots with the libvirt driver on a SHUTDOWN or PAUSED instance. Live snapshot is controlled in the libvirt driver via a config option:

[workarounds]/disable_libvirt_livesnapshot

That now defaults to False so we always attempt a live snapshot with the libvirt driver, at least in CI runs.

Given the guest state of the instance during the snapshot is only a concern for the underlying virt driver and if it's doing a live snapshot or not, (which should probably be a capability trait on the compute node via the driver btw), and the API allows users to snapshot paused and stopped instances:

https://github.com/openstack/nova/blob/7b96206699ac28f807676bd08c6dee7a89bcb77c/nova/compute/api.py#L2717

We should either downgrade the warning to DEBUG level or remove it completely from the compute manager since it's really virt-driver specific.

[Matt Riedemann (mriedem)]

Another thing we should cleanup is the documentation around the "disable_libvirt_livesnapshot" config option because it talks about libvirt 1.2.2 but the minimum required libvirt version at this point is 1.2.9, so that's confusing. Per bug 1741667, we should probably mention that live snapshot of a non-running instance isn't known to work until at least libvirt 3.6.0 per our test infrastructure. Anything between 1.2.9 and 1.3.6 would have to be tested for snapshotting a non-running instance.