Graph-based Anomaly Detection and Root Cause Analysis for Microservices in Cloud-Native Platform
Xinwei Wang, Xu Liu, Peng Xu, Haoran Du · 2024
Due to the excellent characteristics of microservices architecture, such as high scalability, high fault tolerance, and ease of maintenance, and the flexible, efficient, and scalable deployment environment provided by cloud-native platforms, an increasing number of applications are being constructed based on microservices architecture. In such scenarios, the complexity of anomaly detection rises sharply due to the involvement of numerous interrelated microservices components and the complex invocation relationships among these components in each microservices application. Existing anomaly detection methods struggle to maintain sufficient accuracy in such complex environments. To address this challenge, this paper proposes the Graph-based Anomaly Detection and Root Cause Analysis (GADRCA) method for Microservices in Cloud-Native Platform. This method leverages graph structures to comprehensively considers the fundamental distribution characteristics of multiple resource consumption metrics involved in the runtime of microservices applications, such as response time, CPU usage, and memory usage. It also preserves the structural information of trace data within microservices to achieve more comprehensive anomaly detection. Upon identifying anomalous traces, the method determines the direction of anomaly propagation using the traces root cause analysis approach based on multiple metric anomaly types. Finally, a topological graph approach is employed to accurately pinpoint the root cause service. The proposed method’s effectiveness in enhancing anomaly detection and root cause analysis has been demonstrated through a series of experiments.