MicroEGRCL: An Edge-Attention-Based Graph Neural Network Approach for Root Cause Localization in Microservice Systems

Microservices architecture has become the latest trend in building modern applications due to its flexibility, scalability, and agility. However, due to the complex interdependencies between microservices, an anomaly in any one service in a microservice system has the potential to propagate along service dependencies and affect multiple services. Therefore, accurate and efficient root cause localization is a significant challenge for current microservice system operation and maintenance. Focusing on this challenge and leveraging the dynamically constructed service call graph, we propose MicroEGRCL, a root cause localization approach based on graph neural networks with an attention mechanism that includes edge feature enhancement. We conducted an experimental evaluation by injecting various types of service anomalies into two microservice benchmarks running in a Kubernetes cluster. The experimental results demonstrate that MicroEGRCL can achieve an average top1 localization accuracy of 87%, exceeding the state-of-the-art baseline approaches.