Comment 3 for bug 1873265

@Rodolfo Alonso This is a 2019/09/01 Rocky, we used openstack-helm to manage neutron-ovs-agent as a k8s pod.

In my environment, rabbitmq entered split brain state because of physical network loop, once rabbitmq cluster is in split brain state, every rabbitmq node will be out of message sync with one another, so sometimes neutron-ovs-agent can't read full neutron network data, so it can't install full flows into ovs bridges, the result is VMs can't be accessed normally because correct flows aren't there. neutron-ovs-agent also is restarted by k8s because liveness probe failed.

I don't understand why neutron-ovs-agent restart will trigger all the flows will be completely re-installed, this obviously will interrupt ovs bridge data plane forwarding, ideally it should only install new flows or modified flows.

In our environment, now ovs bridges only installed default flows, many flows aren't installed.