Root Cause Analysis Framework for Microservice Systems

This research article presents a root cause analysis framework specifically designed for microservice systems that utilize multimodal data. The framework addresses the challenges associated with traditional root cause analysis methods, which often rely on a single modality of data and struggle to classify unknown failure categories. The proposed method employs a masked graph autoencoder (GAE) for feature extraction and dimensionality reduction, enhancing the quality of data analysis. Additionally, it incorporates an online clustering approach that utilizes expert input to improve failure classification accuracy. The article details the experimental evaluation of this framework on two public datasets, comparing its performance against two baseline methods. The results indicate significant advantages in identifying failure categories, thereby providing targeted troubleshooting guidance for operations personnel. This framework aims to enhance operational efficiency and reduce the workload associated with expert feedback annotations, ultimately contributing to more effective management of microservice systems.