Dealing with observability in interaction-based Offline Runtime Verification of Distributed Systems

TL;DR

This paper presents an offline Runtime Verification algorithm for distributed systems that handles partial observability by dynamically restricting reference interactions, ensuring correctness despite incomplete monitoring data.

Contribution

It introduces a novel offline RV algorithm that manages partial observations in distributed systems by on-the-fly restriction of reference interactions, with proven correctness and performance evaluation.

Findings

The algorithm correctly handles partial observations in distributed systems.

Implementation demonstrates acceptable performance in practical scenarios.

The approach improves robustness of offline RV under incomplete monitoring conditions.

Abstract

Interactions are formal models describing asynchronous communications within a Distributed System (DS). They can be drawn in the fashion of sequence diagrams and executed thanks to an operational semantics akin to that of process algebras. Executions of DS can be characterized by tuples of local traces (one per subsystem) called multi-traces. For a given execution, those local traces can be collected via monitoring and the resulting multi-trace can be analysed using offline Runtime Verification (RV). To that end, interactions may serve as formal references. In practice, however, not all subsystems may be observed and, without synchronising the end of monitoring on different subsystems, some events may not be observed, e.g. the reception of a message may be observed but not the corresponding emission. So as to be able to consider all such cases of partial observation, we propose an offline RV algorithm which uses removal operations to restrict the reference interaction on-the-fly, disregarding the parts concerning no longer observed subsystems. We prove the correctness of the algorithm and assess the performance of an implementation.