on clusters with a large number of services with externalIPs or services from type loadBalancer the ovnkube-node initialization can take up to 50 min

The problem is after a node reboot done by MCO the unschedule taint is removed from the node so the api allocates pods to that node that get stuck on ContrainerCreating and other nodes continue to go down for reboot making the workloads unavailable. (if no PDB exists for the workload to protect it)