Recently a bug/oversight in the L2 announcements logic surfaced. Currently L2 announcements just looks at policies, all services selected by a policy will get leases. It is only after a node becomes lease hold that we look at the IPs inside the service. This means that we also maintain leases for internal services with only ClusterIPs but no ExternalIP or LoadBalancerIP. This is extremely wasteful in terms of performance and reduces the max scalability of the feature.

The fix is simple, we should check, before "selecting" services if they have any useful IPs