An effective parallel convolutional anomaly multi-classification model for fault diagnosis in microservice system

Microservice architecture is a new technology for deploying large-scale applications and services in the cloud. But multivariate time series data with anomalies are increasingly generated in the cloud. Effectively diagnosing the runtime system anomalies is necessary to ensure the quality of service of microservice systems. Typical anomaly detection methods are effective in data quality and computing reliability of cloud computing. However, they all focus on one-class anomaly detection, which may not perform on practical microservice frameworks with diverse types of anomalies. Furthermore, locating the root cause of anomalies to eliminate after detection is essential. To address these issues, we propose an effective parallel convolutional anomaly multi-classification model (PCAC) based on an attention mechanism for fault diagnosis in microservice system. We first construct a parallel convolutional structure that allows subnetworks to extract features independently. Then, channel and spatial attention mechanisms are applied in the parallel convolutional layers to mitigate the loss of feature representation. Finally, causal inference based on the anomalous graph is used to locate the fault in the microservice system. The experimental results clearly show that the proposed model achieves the highest F1 scores on six public microservice datasets, improved by 37.9% in average macro-F1 and 4.4% in average micro-F1 scores respectively, outperforming eight state-of-the-art methods.