Env:
3 node cluster

Cluster Version: 4.16.38(unverified)
Desired Version: 4.16.38
Channel: stable-4.16
Previous Version(s):

Infrastructure
--------------
Platform: BareMetal
Control Plane Topology: HighlyAvailable
apiServerInternalIP: xxx.xxx.xxx.xxx
apiServerInternalIPs: xxx.xxx.xxx.xxx
ingressIP: xxx.xxx.xxx.xxx
ingressIPs: xxx.xxx.xxx.xxx
loadBalancer: None
machineNetworks: xxx.xxx.xxx.xxx/26
Install Type: agent-installer

Network
-------
Network Type: OVNKubernetes

Issue
When the cluster is shutdown and the restarted the cluster is unstable.

Observations:
The ingress vip will not be assigned to any nodes
The api vip is assigned to a node.
The NNCP's will be stuck at provisioning.

To get the cluster back is difficult.
To get the ingress vip assigned to a node will move the keepalived static manifest on each node to a temp location, wait for the keeapalived pods to terminate then move the manifests back.
This will need to be tried a couple of times and results in the api vip getting assigned to multiple nodes.

For the NNCP's they have to be deleted and then recreated whci also does not work the 1st time sometimes.