Unsupervised Microservice Log Anomaly Detection Method Based on Graph Neural Network

The widespread adoption of microservice architecture within the industrial internet domain has led to a substantial increase in system complexity and failure risk. Microservice logs records service and user behavior information in detail. Consequently, microservice logs anomaly detection is crucial for identifying potential issues and maintaining system stability and reliability. In view of the problems of data labeling and difficulty in dealing with timing cross data, this paper proposes an unsupervised microservice logs anomaly detection method based on graph neural network. Based on the optimization of the log parsing process, the proposed method effectively captures the intricate dependencies and temporal relationships in microservice logs using graph structures. Moreover, by learning the graph representation of log sequences in normal mode, this method establishes judgment boundaries and ultimately achieves unsupervised anomaly detection. The experimental results show that the proposed method can effectively extract the features that distinguish log sequences and performs well in accuracy and recall metrics compared with DeepLog, LogAnomaly and DeepTraLog algorithms.