We need to ensure there is no conflict between the two deployed versions of OVN when they are run together in a OCP cluster. 

OCP will use OVN-k8s as the default CNI moving forward. Accordingly, it runs all of the various components required for OVN within the cluster. It's essential that we don't conflict with OVN-k8s when we deploy our own version of OVN for podified. This includes any conflict that may arise from the configuration of OVS.

This is a blocking issue, as the risk here is very high. In this case, the OVS pods coming up interfere with the Geneve interfaces on the OCP Hosts which takes down all Geneve tunnels. Subsequently, all ingress is broken to all applications running in the cluster. We can't risk any impact to existing OCP workloads.

My best guess is that OvS tries to interact with the same Datapath on the host kernel openvswitch API that ovn-k8s is using, and this causes the Geneve interfaces to be removed.

See the investigation below