A fault diagnosis framework for microservice systems in cloud-native networks

The complexity and dynamics of cloud-native data center networks impose increased demands on multisource data-driven fault detection and fine-grained analysis. Effective fault detection and detailed fault correlation and analysis are highly important for improving resilience in cloud-native networks. However, the existing methods lack correlation analysis between faults and multisource data features and face difficulties in generating fault reports related to network components or instances. In this study, we propose a fault-diagnosis method for microservice systems in cloud-native networks to improve network resilience via fault detection, correlation, and analysis. In the proposed method, the correlation among three types of faults and multisource data is analyzed, and feature sequences are constructed by combining key metrics, span duration, and offset features. Additionally, a dynamic multilayer microservice architecture graph employing traces and network resource interfaces is constructed using the fault detection results obtained from a deep learning model. Ultimately, the proposed method generates detailed fault analysis reports. The experimental results show that the fault detection performance of our method is superior to that of baseline approaches and provides engineers with fault correlation and analysis reports related to cloud-native networks, thus improving network resilience.