System Under Oath

Abstract

Stateful learning techniques are crucial for the verification and validation of complex software systems. As the world becomes increasingly reliant on software, the need for secure and reliable applications continues to grow. Previous work on stateful learning has primarily focused on monolithic systems, leaving the application of these techniques to microservice architectures largely unexplored. This is despite their widespread use in modern software development.

This thesis investigates the challenges associated with learning systems based on microservice architectures. First, we present a framework for generating valid API requests across interacting services. Next, we address system flakiness during learning through the introduction of Two-Phase Equivalence Querying with adaptive sampling. Finally, to explain the non-deterministic behavior observed in microservice systems, we introduce the concept of State Drift, which captures state changes caused by internal communication between services.

To conclude this thesis, we apply the techniques developed throughout this work to a real-world software system. The system learned is an implementation of OAuth provided by Ubiqu. This case study demonstrates how our proposed techniques can be used in industry and highlights the practical challenges that remain when applying stateful learning to microservice architectures.