Instance Migration Validity for Dynamic Evolution of Data-Aware Processes

Abstract

Likely more than many other software artifacts, business processes constantly evolve to adapt to ever changing application requirements. To enable dynamic process evolution, where changes are applied to in-flight processes, running process instances have to be migrated. On the one hand, as many instances as possible should be migrated to the changed process. On the other hand, the validity to migrate an instance should be guaranteed to avoid introducing dynamic change bugs after migration. As our theoretical results show, when the state of variables is taken into account, migration validity of data-aware process instances is undecidable. Based on the trace of an instance, existing approaches leverage trace replaying to check migration validity. However, they err on the side of caution, not identifying many instances as potentially safe to migrate. We present a more relaxed migration validity checking approach based on the dependence graph of a trace. We evaluate effectiveness and efficiency of our approach experimentally showing that it allows for more instances to safely migrate than for existing approaches and that it scales in the number of instances checked.