First-Order Dynamic Logic for Compensable Processes

Abstract

Abstract. Compensable programs offer a convenient paradigm to deal with longrunning transactions, because they offer a structured and modular approach to the composition of distributed transactional activities, like services. The basic idea is that each activity has its own compensation and that the compensable program fixes the order of execution of such activities. The main problem is how to guarantee that if one or even many faults occur then the compensations are properly executed so to reach a consistent configuration of the system. We propose a formal model for such problems based on a concurrent extension of dynamic logic that allows us to distill the hypothesis under which the correctness of compensable programs can be ensured. The main result establishes that if basic activities have a correct compensation we can show the correctness of any compound compensable program. Moreover, we can use dynamic logic to reason about behavioural and transactional properties of programs.

LTS Semantics for Compensation-Based Processes

“…Actually, under the assumption that compensation cannot fail, only successful traces are present in S (as well as in S i for any i ∈ [1,5]). This is not necessarily the case for the last extension in § 7.…”

“…In the end we would like to integrate the specified extensions as well as the option to choose which compensation policy should be used together with a high-level dynamic logic for validation and verification. Some promising steps in this direction are described in [5]. Related work.…”

Data Abstraction in Coordination Constraints

“…Using constraints to describe how data flows in a connector has been investigated, for example, for the BIP [3,4] and the Reo [1,7,15] coordination languages. Constraints have also been used to describe desirable properties of process algebras, such as Bruni's et al's compensable processes [5]. In order to keep the problem of producing and executing connectors tractable, only properties that bear no computation are captured by the constraints.…”