We consider the problem of developing reliable services to be deployed in partitionable asynchronous distributed systems. What makes this task difficult is guaranteeing the consistency of shared state despite failures and recoveries. While process groups and view-synchronous communication can significantly simplify reasoning about failures in such systems, they are insufficient for dealing with recoveries and merging of partitions after repairs. We define shared state reconciliation as the abstract characterization of recovery and show how view-synchronous communication can be enriched in order to solve it simply and systematically.
