Inductive Learning-Based Qualitative Fault Diagnosis in Distributed Systems

Abstract

The growing complexity of microservice systems poses significant challenges in diagnosing faulty systems. Traditional monitoring techniques often fall short due to the distributed and dynamic nature of these systems. This paper presents a novel model-based diagnostics framework that uses multimodal observability data for accurate fault detection and localization in microservice environments. The diagnostic process uses Answer Set Programming (ASP), a declarative programming language that leverages logic reasoning over a qualitative multimodal data model to provide insights into the system's state. The presented approach introduces an inductive learning solution for extracting the diagnostic rules, utilizing Inductive Learning of Answer Set Programs (ILASP) to derive explainable diagnostic rules from labeled historical datasets automatically. The approach was evaluated on a benchmark microservice application dataset with promising results compared to existing fault detection and diagnostic solutions.