An Efficient Hierarchical Failure Recovery Algorithm Ensuring Semantic Atomicity for Workflow Applications

Failure recovery is essential for transactional workflow management systems. When a failure occurs, compensation is usually used to rollback corresponding processes and ensures semantic atomicity. Previously, many mechanisms have been proposed while performance problem is rarely considered. This is a limitation since compensating activities are rather expensive. In this paper, we propose an efficient hierarchical failure recovery algorithm, HFR. In stead of treating the compensation sphere as flat flow graphs, it generates compensation graph in a hierarchical manner from lower layers to higher ones according to nested structure of workflow execution history. The end compensation point is dynamically generated such that the compensation sphere is confined to a layer as low as possible and the number of compensations can be reduced. HFR can guarantee semantic atomicity of a workflow instance in case of a failure. The correctness of it is proved. Theoretical performance analysis shows that HFR has preferable efficiency.