一种基于模式识别的微服务异常检测方法

An Anomaly Detection Approach for Microservices Based on Pattern Recognition

云计算数据中心面向多租户部署异构软件应用,能够优化资源配置以有效提升资源利用率,然而基于微服务架构的软件应用的异构性和动态性对云计算数据中心的运行维护管理带来了困难。已有运行监测技术通常采用统一模型分析应用运行状态,然而基于微服务架构的多样化软件应用的行为差异巨大,难以通过单一模型刻画。该文提出一种基于系统调用模式识别的微服务异常检测方法。使用系统调用追踪技术监测微服务的运行状态,基于贝叶斯学习刻画软件应用的行为,根据行为特点对微服务进行自动化分类,在特定类型下基于自动编码器建立微服务运行状态模型,根据模型检测当前微服务的异常行为。实验结果表明,所提出的方法具有较低的性能开销,能够有效区分微服务类型,并且准确检测微服务异常。

The data centers of cloud computing deploy heterogeneous software applications for multi-tenants,which can optimize resource allocation to effectively improve resource utilization.However,the heterogeneity and dynamics of emerging microservice architectures raise difficulties to the operation and maintenance of data centers.The existing operation monitoring technology usually adopts unified model to analyze the application operation state,but the behavior of diversified software applications based on micro-service architecture is quite different,so it is difficult to describe through a single model.We propose an anomaly detection approach for microservices based on system call pattern recognition.The system call tracing technology is employed to monitor the running status of microservices and describe the behavior of software application based on Bayesian learning.Automatic classification of microservices is carried out according to the behavior characteristics.In a specific type,a microservices operation state model is established based on the automatic encoder.According to the model,abnormal behaviors of the current microservices are detected.Experiment shows that the proposed approach has low performance overhead,can effectively distinguish the types of microservices,and accurately detect typical anomalies.