A robust web service composition model with decentralized data dependency analysis and rule-based failure recovery capability
Le Gao · ThinkTech (Texas Tech University) · 2012
Past work on transactional workflows is inadequate for service composition since most techniques that support relaxed isolation do not actively address the impact that the failure and recovery of one process can have on other data dependent processes. This research has developed a robust web service composition model with decentralized data dependency analysis and rule-based failure recovery capability to dynamically respond to an execution failure of a process. By introducing the concept of assurance point (AP) into the web service composition, pre and post conditions of the critical operations can be checked during the process execution, which provides multiple protections against service execution failure. In this research, the AP model has been fully developed in the context of programming control structures for conditionals, iteration, and parallelism. In addition, a process execution engine that supports the AP model has been developed. The execution and recovery semantics of the AP model have been formalized by using Petri Nets and YAWL. The formalization verifies the correctness of the AP model. An architecture of a process execution agent (PEXA) has been designed by integrating the decentralized data dependency analysis with the AP model. The PEXA includes several components that can execute processes in the AP model, build local data dependency lists, determine the action for responding to a data dependency event, and also send and receive notifications among PEXAs. Unlike past work that performs a total rollback or checks a single rule to determine the action, the PEXA uses the combination of event-driven and rule-based techniques to dynamically respond to a data dependency event, which minimizes the impact caused by data dependencies between a failed process and other concurrently executing processes.