Efficient interaction-based offline runtime verification of distributed systems with lifeline removal

Abstract

Runtime Verification (RV) refers to a family of techniques in which system executions are observed and confronted to formal specifications, with the aim of identifying faults. In offline RV, observation and verification are done in two separate and successive steps. In this paper, we define an approach to offline RV of Distributed Systems (DS) against interactions. Interactions are formal models describing communications within a DS. A DS is composed of subsystems deployed on different machines and interacting via message passing to achieve common goals. Therefore, observing executions of a DS entails logging a collection of local execution traces, one for each subsystem, collected on its host machine. We call multi-trace such observational artifacts. A major challenge in analyzing multi-traces is that there are no practical means to synchronize the ends of observations of all the local traces. We address this via an operation called lifeline removal, which we apply on-the-fly to the specification during the verification of a multi-trace once a local trace has been entirely analyzed. This operation removes from the interaction the specification of actions occurring on the subsystem that is no longer observed. This may allow further execution of the specification by removing potential deadlock. We prove the correctness of the resulting RV algorithm and introduce two optimization techniques, which we also prove correct. We implement a Partial Order Reduction (POR) technique by selecting a one-unambiguous action (as a unique first step to a linearization) whose existence is determined via the lifeline removal operator. Additionally, Local Analyses (LOC), i.e., the verification of local traces, can be leveraged during the global multi-trace analysis to prove failure more quickly. Experiments illustrate the application of our RV approach and the benefits of our optimizations.