Microservice Anomaly Diagnosis with Graph Convolution Network Based on Implicit Microservice Dependency
Hao Tang, Yuchun Guo, Jingjing Yang, Yishuai Chen · 2023
Recently, microservice architecture has become the mainstream choice for enterprise business system design due to flexibility and scalability. However, numerous components and complex dependencies make it difficult to diagnose microservice anomalies. Microservice dependency changes by anomaly types and components, which has not gained enough attention. In this paper, we propose an anomaly diagnosis method, named ID-GCN, based on a graph convolution network. Firstly, we customize the feature vectors of microservices via multi-aspects characteristic extraction. Secondly, an implicit dependency graph of microservices is constructed dynamically to capture the complex and variable microservice relationship. Finally, we use a graph convolution network to classify the normal and anomalous microservices, with the consideration of the real-time microservice dependency and microservice states. Based on an open dataset with 389 cases, our experimental evaluation shows that ID-GCN can effectively diagnose anomalies, with 94% mean average precision, outperforming the baseline method by 19%.