Model-based Testing of Practical Distributed Systems in Actor Model

TL;DR

This paper presents a method for model-based testing of distributed systems in the actor model, generating exhaustive test suites without modifying the system, demonstrated on a replication algorithm.

Contribution

It introduces an efficient approach to generate comprehensive test suites for actor-based distributed systems without code modifications.

Findings

Successfully verified a real-world replication algorithm

Generated exhaustive test suites covering all states and transitions

No modifications needed for the distributed system code

Abstract

Designing and implementing distributed systems correctly can be quite challenging. Although these systems are often accompanied by formal specifications that are verified using model-checking techniques, a gap still exists between the implementation and its formal specification: there is no guarantee that the implementation is free of bugs. To bridge this gap, we can use model-based testing. Specifically, if the model of the system can be interpreted as a finite-state automaton, we can generate an exhaustive test suite for the implementation that covers all possible states and transitions. In this paper, we discuss how to efficiently generate such a test suite for distributed systems written in the actor model. Importantly, our approach does not require any modifications to the code or interfering with the distributed system execution environment. As an example, we verified an implementation of a replication algorithm based on Viewstamped Replication, which is used in a real-world system.