AI-Powered Root Cause Analysis System Cuts Microservice Diagnosis Time by 77%

A new root cause analysis system demonstrates how agentic AI can tackle the challenge of diagnosing failures across massive, interconnected microservice platforms. Deployed by video platform Kuaishou for over six months, the system—called KRCA—has cut average diagnosis time by more than three-quarters, a substantial improvement for site reliability teams managing production infrastructure.

The Challenge at Scale#

Modern internet companies rely on hyper-scale microservice architectures comprising hundreds or thousands of loosely coupled services that evolve continuously. When failures occur in such systems, pinpointing which service caused the problem becomes exponentially harder as the system grows. Traditional root cause analysis methods struggle under this complexity, leaving engineers to sift through sprawling data manually.

How KRCA Works#

The system employs a multi-stage pipeline designed to navigate the massive search space efficiently. It begins by analyzing API-level traffic patterns to narrow down potentially problematic services. Next, it constructs a skeleton-based causal graph from anomalous metrics, establishing a high-confidence structural foundation. Finally, a memory-augmented multi-agent framework verifies the relationships between services and generates a comprehensive failure report.

By layering structured causal constraints with multi-agent reasoning, KRCA balances diagnostic precision with the speed needed for real-world production environments.

Real-World Performance#

Testing shows KRCA achieved accuracy scores of 88% and 79% when identifying the root cause service and classifying the failure type, respectively—outperforming existing approaches by at least 31% in absolute gains. In production, the system reduced average diagnosis time by 77.3%, freeing engineers to focus on remediation rather than investigation.

Why It Matters#

As microservice architectures grow more prevalent, the ability to rapidly diagnose failures becomes a competitive advantage. KRCA's success suggests that agentic AI systems capable of reasoning across causality and dynamically adapting to system changes could reshape how teams maintain infrastructure reliability.