Graph-based Root Cause Analysis for Service-Oriented and Microservice Architectures

Abstract Service-oriented architectures and microservices define two ways of designing software with the aim of dividing an application into loosely-coupled services that communicate among each other. This translates into rapid development, where each service is developed and deployed by small teams, enabling continuous shipping of new features and fast-evolving applications. However, the underlying complexity of this type of architecture can hinder observability and maintenance by the user. In particular, identifying the root cause of an anomaly detected in the application can be a difficult and time-consuming task, considering the numerous services and connections to be examined. In this work, we present a root cause analysis framework, based on graph representations of these architectures. The graphs can be used to compare any anomalous situation that happens in the system with a library of anomalous graphs that serves as a knowledge base for the user troubleshooting those anomalies. We use the Grid’5000 testbed to deploy three different architectures and inject a set of anomalies. The results show how our graph-based approach is 19.41% more effective than a machine learning method that does not take into account the relationship between elements.