An illustrative recovery approach for stateful interaction failures of orchestrated processes

During a stateful interaction, a partner service may become unavailable because of a server crash or a temporary network failure. Once the failed service becomes available again, the interaction partners do not have any knowledge about each other's state, possibly resulting in errors or deadlocks. This paper proposes an approach to the recovery of stateful interactions based on service interaction patterns and process transformations. Our recovery approach works without a central management node and without additional communication protocols. We also minimize the changes to the description of the service supported by the recovery-enabled process. Our approach allows one partner process to be modified in order to support failures in a way that interaction with the other (unchanged) processes is still possible.