Both clients and servers should be aware of service clusters. Clients should round-robin requests to identical services, and services should be able to proxy connections to other services transparently if they are experiencing failure or in the process of shutting down.