Microservice Anomaly Monitoring Method based on Multi-dimensional Spatiotemporal Feature Reconstruction

In a modern microservice architecture, the largescale distribution of services and high concurrency access make anomaly monitoring particularly complex. Traditional monitoring methods usually rely on a single data dimension, such as request delay or resource utilization, making capturing complex abnormal patterns challenging. Therefore, this paper proposes a microservice anomaly monitoring method based on multidimensional spatiotemporal feature reconstruction, aiming to reconstruct the complete feature space by integrating different data sources to improve the accuracy and robustness of anomaly detection. This paper’s method first extracts adequate data, such as service response time and service call chain from the tracking dataset, with time and space as the core and constructs it into a multi-dimensional feature matrix, fully considering the spatial and temporal correlation between microservices. Second, this paper introduces a deep learning model combining convolutional neural networks and gated recurrent cells to capture dynamic dependence and feature interactions in spatiotemporal dimensions. The convolutional neural network module extracts the spatial dependence through the convolution operation, while the gating cycle unit module captures the changing trend of the time series. By paralleling two modules and introducing the attention mechanism, the model is optimized with the reconstruction error, and finally, the network abnormality is judged by setting the threshold.