Multi-Agent Systems for Root Cause Analysis in Microservices

Root cause analysis (RCA) in microservice systems is notoriously difficult due to distributed logs and metrics. A new paper from researchers at the University of Oulu proposes LATS-RCA, a multi-agent framework that leverages large language models to automate RCA. Unlike prior linear approaches, LATS-RCA formulates RCA as a reflection-guided tree-structured search using a Language Agent Tree Search algorithm. Multiple LLM-driven agents each analyze execution logs and performance metrics for a specific microservice, collecting evidence. Reflection scores derived from intermediate diagnostic states guide the search toward the most probable root cause, enabling a more thorough exploration of fault paths.

The framework was evaluated on two environments: Light-OAuth2 (LO2), a small-team Java system with homogeneous tech stack, and a production environment (Prod) from a case company with higher complexity. Results on LO2 show high diagnostic accuracy, while Prod shows lower accuracy and higher computational costs. The production deployment highlights real-world challenges such as polyglot technology stacks, inconsistent logging practices, and multi-factor root causes. The study demonstrates LATS-RCA's practical viability and sets a benchmark for future LLM-driven RCA tools in complex microservice architectures.