[nova] Running multinode resize tests in the gate
mriedemos at gmail.com
Fri Apr 19 20:39:17 UTC 2019
On 4/18/2019 1:04 PM, Artom Lifshitz wrote:
> Yo Novans,
> In trying to understand why my partial fix to bug 1813789  -
> specifically the part where reverting a resize fails because we get
> the network-vif-plugged event from Neutron before we start listening
> for it - I noticed that we don't test multinode resize (and revert) in
> the gate. As far as I can tell, the only place where the revert tests
> are run are in tempest-full-py3, and that's on an allinone single
> In order to test a theory of mine about revert resize on the same node
> vs source to dest and back again, I do have a DNM patch up  to run
> the resize tests in the nova-live-migration job - I think that's the
> only multinode tempest job we have, right? I think that job also runs
> evacuation tests , correct? So, should we just rename it
> nova-multinode and add resize to the list of tests it runs?
We used to run tempest-multinode-full until recently .
That's probably what you're looking for, but I don't really want to add
that job back for the reasons mentioned in the commit message and
associated mailing list thread.
We could make nova-next multinode but then we start having overlap with
nova-live-migration (minus the evacuate testing) and tempest-slow. The
tempest-slow job is also multinode but apparently there aren't any
resize tests marked slow so those don't get run in that job.
So...enabling resize tests in nova-live-migration is probably the best
option at this point, and yeah likely means renaming it as well (since
it's already confusingly named nova-live-migration but also runs
evacuate tests too via a post-test hook). The only thing I worry about
with that one is needing to keep a regex of tempest tests we run, but
technically we're already doing that for the live migration tests .
Maybe it's time to bite the bullet and use a real rc file like we have
for some other jobs, e.g. .
More information about the openstack-discuss