Agentic Structured Graph Traversal for Root Cause Analysis of Code-related Incidents in Cloud Applications

2025-12-26

AI Summary

To address the challenge of root-cause localization in cloud-native applications caused by tight coupling between code and configuration, this paper proposes an LLM-driven dual-graph collaborative root-cause analysis (RCA) framework. It integrates a Service Dependency Graph (SDG) with a fine-grained Hammock-block Program Dependence Graph (PDG) to establish a structured, cross-service-and-code-level graph traversal paradigm. We introduce a novel ReAct-style agent workflow augmented with graph-topology-aware reinforcement learning to jointly optimize diagnostic paths and underlying graph structure. Evaluated on 30 real-world cloud failure cases, our method achieves a 3.1× improvement in root-cause localization accuracy over the state-of-the-art ReAct baseline, reduces token consumption by 3.8×, and enables the construction of the first open-source RCA benchmark.

Technology Category

• Reasoning under Uncertainty: Causality
• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Cloud incidents pose major operational challenges in production, with unresolved production cloud incidents cost on average over $2M per hour. Prior research identifies code- and configuration-related issues as the predominant category of root causes in cloud incidents. This paper introduces PRAXIS, an orchestrator that manages and deploys an agentic workflow for diagnosing code- and configuration-caused cloud incidents. PRAXIS employs an LLM-driven structured traversal over two types of graph: (1) a service dependency graph (SDG) that captures microservice-level dependencies; and (2) a hammock-block program dependence graph (PDG) that captures code-level dependencies for each microservice. Together, these graphs encode microservice- and code-level dependencies and the LLM acts as a traversal policy over these graphs, moving between services and code dependencies to localize and explain failures. Compared to state-of-the-art ReAct baselines, PRAXIS improves RCA accuracy by up to 3.1x while reducing token consumption by 3.8x. PRAXIS is demonstrated on a set of 30 comprehensive real-world incidents that is being compiled into an RCA benchmark.

Problem

• Diagnoses code- and configuration-caused cloud incidents using agentic workflow
• Traverses service and code dependency graphs to localize and explain failures
• Improves root cause analysis accuracy while reducing computational token usage

Innovation

• LLM-driven structured traversal over service and code dependency graphs
• Orchestrates agentic workflow for diagnosing code-related cloud incidents
• Improves root cause analysis accuracy while reducing token consumption