Any sane design is going to have exactly one ingress/egress for end-user traffic per application. Even an application with a few hundred servers is only going t

Any sane design is going to have exactly one ingress/egress for end-user traffic per application. Even an application with a few hundred servers is only going to be available to end-users at 1 IP address (maybe several for load balancing/redundancy). It wouldn't be hard to add a global reverse-proxy/load-balancer HA cluster at the actual network edge forwarding traffic to individual applications' own load balancers.