New Framework Tackles Diagnosing Failures in Cloud Microservices Without Manual Setup

Microservice architectures power modern cloud applications, but their distributed nature makes troubleshooting failures a major operational challenge. When systems falter, users suffer degraded experiences and companies face real financial consequences. A new research effort addresses fundamental gaps in how the industry automates anomaly detection and root cause analysis (RCA) for these complex environments.

The Problem with Current Approaches#

Existing techniques for diagnosing microservice failures rely on a patchwork of incomplete methods. Most treat anomaly detection and root cause identification as separate steps, which breaks down when detectors make mistakes due to noisy data or processing delays. Current systems lean heavily on metrics, logs, and traces—overlooking event-level signals like API calls and configuration changes that often hold diagnostic clues.

A critical limitation is the requirement for a pre-existing service call graph. Many approaches cannot function without knowing exactly how services communicate with each other beforehand, making them impractical for dynamic or unknown architectures. Beyond methodology, the field lacks standardized evaluation datasets and comparison frameworks, making it nearly impossible to benchmark progress fairly. Causal inference has become the dominant RCA paradigm, yet researchers still question its real-world effectiveness, speed, and reliability under adverse conditions.

Multi-Modal Solutions and Unified Frameworks#

The research introduces three key contributions. BARO combines anomaly detection and root cause analysis as an integrated end-to-end pipeline for metric-based diagnosis. EventADL extends this approach to event data—a richer but underexplored signal source in microservice systems. TORAI advances the field further by performing multimodal RCA without requiring operators to provide service call graphs upfront, removing a major deployment barrier.

Testing on real production microservice systems confirms these methods work reliably across diverse failure scenarios and architectural configurations.

Benchmarks to Accelerate Research#

The work also establishes RCAEval, a standardized benchmark suite offering ready-to-use datasets and reproducible baseline implementations. This shared foundation enables researchers to compare methods objectively and build on each other's progress rather than starting from scratch.