COCA is a novel approach that leverages code knowledge and LLMs to automatically identify root causes of distributed system failures using only user-reported issues.

• Integrates code context knowledge with generative AI to diagnose failures without comprehensive monitoring data
• Works effectively with limited information from Github or JIRA issue reports
• Significantly outperforms existing techniques in failure diagnosis accuracy
• Provides actionable insights for rapid problem resolution in complex distributed systems

This research addresses a critical engineering challenge: reducing mean time to resolution for distributed system failures, potentially saving companies thousands of development hours and reducing system downtime.

COCA: Generative Root Cause Analysis for Distributed Systems with Code Knowledge