Prediction and localization of potential failures in microservices through learning link data

With the widespread adoption of microservice architectures in distributed systems, efficiently detecting and localizing microservice failures has become a critical challenge in ensuring system stability and reliability. Traditional fault detection methods often struggle to address the dynamic nature and complexity of microservice systems. To overcome this challenge, this paper proposes a microservice fault prediction and localization approach based on link data learning. The proposed method combines fault injection and performance analysis to design a microservice fault detection and classification model. Specifically, historical fault data is first generated using fault injection techniques, upon which a fault detection model is constructed to predict faults. Subsequently, a fault classification model is applied to accurately localize the fault types. Experimental validation is performed on two open-source microservice systems. The results demonstrate that, compared to traditional methods, the proposed model exhibits a lower false positive rate in fault type prediction and achieves significant improvements in both recall and precision. Furthermore, this paper explores the application of the prediction model in real-world scenarios. By collecting real-time microservice link data and extracting relevant features, the system is capable of dynamically predicting potential faults. Experimental results further show that the system can automatically trigger scaling mechanisms when potential faults are detected, thereby effectively ensuring the stable operation of microservice systems.