Description of problem:

Customer has quite special usage model for his RHOSP 16.2 deployment with ML2/OVN: numerous VMs are created and deleted daily. We have another bugs reported for his deployments:
- bug #2203811
- bug #2203848

After applying hotfix for bug #2203848 (it may be unrelated, but leaving a note just in case) customer reported another problem: 10-20 %% of VMs are not able to start because of timeout while waiting for network-vif-plugged event.

It looks like this event should come from Neutron Server. We have collected logs for these events and kindly ask for a help to figure out what went wrong there.


Version-Release number of selected component (if applicable):
Red Hat OpenStack Platform release 16.2.4 (Train)


How reproducible:

#!/bin/bash
COUNT=${1}
NET=${2}
for i in {1..10}; do
  echo "round ${i}"
  openstack server create --image IMAGE_UUID --use-config-drive --min ${COUNT} --max ${COUNT} --flavor m1.tiny --network ${NET} launch-test
  sleep 30
  openstack server delete $(openstack server list --status ACTIVE -c ID -f value | xargs)
done


Actual results:
10-20%% of servers fail because of network-vif-plugged timeout

Expected results:
All servers started properly