We had a customer case where lots of OCP clusters are deployed on top of OpenStack. An OSP upgrade attempt triggered OCP VIP failover that loaded ovn-northd so much that other OCP clusters started their failovers too causing a snowball effect.

The dataplane has never recovered fully and OCPs kept sending gratuitous ARPs periodically from all their nodes, causing lot of virtual parent ports updates in the southbound DB.

This ticket serves as a discussion starter how to make OVN more resilient to gARPs coming from multiple sources at the same time.