This is a clone of issue OCPBUGS-4370. The following is the description of the original issue:
—
When we set the  k8s.ovn.org/node-primary-ifaddr annotation on the node, we simply take the first valid IP address we find on the node gateway. We exclude link-local addresses and those in internally reserved subnets (https://github.com/openshift/ovn-kubernetes/pull/1386). 

Now, we might have more than one "valid" IP address on the gateway, as observed in:
 https://bugzilla.redhat.com/show_bug.cgi?id=2081390#c11 , https://bugzilla.redhat.com/show_bug.cgi?id=2081390#c14

For instance, taken from a different cluster than in the linked BZ:

sh-4.4# ip a show br-ex
7: br-ex: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
link/ether 00:52:12:af:f3:53 brd ff:ff:ff:ff:ff:ff
inet6 fd69::2/125 scope global dadfailed tentative <---- masquerade IP, excluded
valid_lft forever preferred_lft forever
inet6 fd2e:6f44:5dd8:c956::4/128 scope global nodad deprecated <--- real node IP, included
valid_lft forever preferred_lft 0sec
inet6 fd2e:6f44:5dd8:c956::17/128 scope global dynamic noprefixroute <---added by keepalive, INCLUDED!!
valid_lft 3017sec preferred_lft 3017sec
inet6 fe80::252:12ff:feaf:f353/64 scope link noprefixroute <--- link local, excluded
valid_lft forever preferred_lft forever

Above we have fd2e:6f44:5dd8:c956::4/128 which is the LB VIP of ingress added by keepalive.

We don't currently distinguish in the code between the node IP as in node.spec.IP and other IPs that might be added to br-ex by other components. 

Would it be a good idea to just set the node primary address annotation to match node.spec.IP?