Reproduce “Eadro: An End-to-End Troubleshooting Framework for Microservices on Multi-source Data” on OpenResearch

About this paper

Cheryl Lee, Tianyi Yang, Zhuangbin Chen, Yuxin Su, Michael R. Lyu

Researchers from The Chinese University of Hong Kong and Sun Yat-sen University developed Eadro, an end-to-end framework that jointly performs anomaly detection and root cause localization in microservice systems by integrating multi-source data (logs, KPIs, and traces). Eadro achieved F1-scores up to 0.989 for anomaly detection and Hit Rate @1 up to 0.982 for root cause localization, substantially outperforming existing methods on benchmark systems.

Key insights

• Integrating multi-source data (logs, KPIs, traces) provides a more holistic and accurate understanding of microservice system health, enabling the detection and localization of diverse anomaly types.
• Jointly training anomaly detection and root cause localization in an end-to-end framework allows for shared knowledge and improves both tasks' accuracy and efficiency compared to disconnected pipelines.
• Modeling inter-service dependencies with graph neural networks (e.g., GAT) is crucial for tracing anomaly propagation and precisely pinpointing root causes, complementing intra-service insights from individual data sources.

Summary by alphaXiv.