We describe a collection of communication primitives integrated with a mechanism for handling process failure and recovery. These primitives facilitate the implementation of fault-tolerant process groups, which can be used to provide distributed services in an environment subject to non-malicious crash failures.