Multi-Agent AI Outperforms Baselines in Microservice Root Cause Analysis.

Identifying the root causes of failures in complex microservice environments, especially with large-scale, multimodal telemetry data, remains a significant challenge for both classical and current LLM-based approaches. A recent analysis using the OpenRCA dataset, which features diverse telemetry without extensive domain knowledge, confirms these difficulties, with existing methods yielding consistently low accuracy. Researchers developed a Structured Multi-Agent RCA pipeline that substantially outperforms previous LLM-based and classical baselines. This pipeline can operate with or without explicit domain knowledge. A key diagnostic tool, a reverse reasoning agent, was introduced to determine if failures stemmed from a "Reasoning Gap" (evidence present but unused) or "Data Ambiguity" (evidence truly absent). The analysis revealed that in most cases, the necessary evidence was available, indicating that the primary limitation is the agent's ability to correctly reason over the data, rather than a lack of data itself. The study also proposes an automated rule mining pipeline to extract discrimination rules, reducing the need for manual knowledge curation. Ultimately, the research highlights that stronger models with embedded domain expertise are crucial, and simply improving data pipelines or scaffolding alone will not close the reasoning gap.