TraceNet: Operation Aware Root Cause Localization of Microservice System Anomalies
Jingjing Yang, Yuchun Guo, Yishuai Chen, Yongxiang Zhao · 2023
Microservice architecture has been widely-adopted to enable scalable, flexible, and resilient applications. However, numerous components and complex dependencies pose daunting challenges for anomaly root cause localization. Existing localization methods rely on supervision or coarse-grained analysis. In this paper, we propose TraceNet, an algorithm to locate root causes, based on a fine-grained construction of microservice dependency at the operation level. Firstly, to mitigate the interference from anomaly propagation, TraceNet builds the microservice dependency graph by distinguishing different operations, which depicts the complex microservice dependency at the operation dimension. Then, TraceNet classifies abnormality into inner-abnormality and outer-abnormality, and quantifies the abnormality of microservices by an edge-operation-microservice aggregation, Finally, TraceNet locates root causes based on the largest anomaly subgraph and candidates' different impacts, without human interference. Our experimental evaluation on an open dataset shows that TraceNet can effectively locate root causes, with 90% mean average precision, outperforming state-of-the-art methods.