With the pervasive adoption of cloud computing, modern software systems have evolved toward highly distributed, elastic, and microservice-based architectures, which significantly increase the difficulty of effective fault localization. Spectrum- Based Fault Localization (SBFL) techniques are widely used to assist automated program repair and manual debugging; however, their practical effectiveness remains limited. Specifically, existing SBFL methods largely overlook the static semantic characteristics of program statements and fail to fully exploit the abundant execution data and scalable computational resources available in cloud computing environments. To address these limitations, this paper proposes a lightweight fault localization approach based on learning to rank, explicitly designed for cloud computing scenarios. The proposed method employs a linear ranking Support Vector Machine (SVM) that jointly integrates traditional SBFL suspiciousness scores with static statement-level features, including variables, operators, and statement categories, to construct a more discriminative fault localization model. Furthermore, to better leverage resource coordination and large-scale data processing capabilities in cloud environments, a cross-project training strategy is adopted, and distributed cloud resources are utilized to enable efficient model training and validation. The proposed approach is evaluated on large-scale datasets comprising 19 Java, 19 C, and 2 C++ projects. Experimental results demonstrate that, under the EXAM metric with the worst-case evaluation strategy, the proposed method reduces the number of statements requiring inspection by 26.1% compared to the best-performing SBFL technique. These findings indicate that integrating static program features with cloud-enabled learning and resource coordination can substantially improve fault localization effectiveness in complex cloud-based software systems.
Loading....