Ensuring consistency in long running transactions

Abstract

Flow composition languages permit the construction of long-running transactions from collections of independent atomic services. Due to environmental limitations, such transactions usually cannot be made to conform to standard ACID semantics. We propose set consistency, a powerful, yet intuitive, notion of consistency for long-running transactions. Set consistency considers the collection of permanent (non-intermittent) changes made by a process, when viewed at the end of its execution. Consistency requirements for such collections of changes are specified as predicates over the atomic actions of a process. Set consistency generalizes self-cancellation, a standard consistency requirement for long-running transactions, where failed processes are responsible for undoing any partially completed work. Set consistency can also express strictly stronger requirements, such as mutual exclusion or dependency.