Generative AI Test Generation and Intelligent Defect Attribution Method for Large Scale Distributed Systems

Abstract

To address the increasingly complex demands of large-scale distributed systems in test generation and defect localization, this study proposes a strategy grounded in a generative AI-driven test generation and intelligent attribution framework. The test environment is first constructed through the formal definition of system architecture, business semantics, and interaction constraints, ensuring that the generated tests accurately reflect real operational logic. Then, leveraging the reasoning and combinatorial capabilities of large-scale generative models, the test environment is expanded into a diversified set of execution paths, thereby enhancing the system's ability to cope with highly dynamic and uncertain runtime conditions and improving test coverage in complex operational states. Furthermore, multi-source observability data-including logs, traces, metrics, and dependency metadata-is integrated and modeled to produce a structured representation for abnormal correlation analysis. Based on this representation, causal reasoning is applied to derive dependency relationships and event propagation paths among system modules, enabling effective and efficient root cause diagnosis across distributed nodes. This approach significantly reduces the diagnostic search space and improves the interpretability of system anomalies. Experimental validation using a model test prototype demonstrates that the proposed method outperforms traditional testing and fault attribution approaches in terms of test diversity, fault excitability, and accuracy of causal determination. These results indicate that the framework offers robust support for the construction of AI-based assurance mechanisms in large-scale distributed systems, contributing to improved system reliability, fault tolerance, and automated quality assurance.