DyTrace: Capturing Dynamic Call Dependencies in Microservice Anomaly Using the Spatial-Temperal Attention Mechanism

Microservice architecture is widely used in the Internet of Things (IoT) and cloud-native domains. As the scale of micro services systems increases and the inter-service call dependencies become more complex, anomaly detection in microservices has become a technical challenge for data centers. Most existing work relies on metric and trace for anomaly detection. However, these approaches do not account for the dynamic changes in the call dependencies between microservice instances in concurrent request scenarios, making it difficult to perceive the propagation patterns of anomalies between microservices and reducing detection accuracy. This paper proposes a spatial-temporal attention network called DyTrace for microservice anomaly detection. The proposed method first preprocesses metric and trace data to construct a microservice system state graph, then uses a graph attention neural network to learn the micro service spatial dependencies at each time point. A temporal attention network is employed to capture the time-dependent relationships of the system state, and finally, anomaly detection is performed based on Deep Support Vector Data Description (Deep SVDD). Performance evaluation results on the AIOps2022 dataset show that DyTrace outperforms the current state-of-the-art methods, with an average improvement of 20.9% in Precision, 22.8% in Recall, and 21.7% in F1 Score.