Research And Application Of Cascading Failure Prediction In Service Mesh

With the continuous expansion of enterprise application development,the traditional monolithic architecture is large and complex,and it can no longer meet market demand.With its effective scalability,lightweight,agile deployment and resource isolation,microservice architecture has gradually become the mainstream deployment model.But there is also a problem that cannot be ignored in microservice architecture,it is the problem of cascading failure.A cloud platform will deploy a large number of microservices.These microservices are related to each other,and form a complex network of relationships.Once one or two microservices fail,other microservices may be implicated.This will cause large-scale failures,and lead to global degraded performance or even system crash.In order to avoid cascading failures,timely pre-judgment is required before failures occur.The prediction of cascading failures in traditional microservice system is based on service access traffic and service monitoring.The prediction is inaccurate,and there may be over-response or poor responses.On the one hand,it will affect access to some services,and on the other hand,it may exacerbate the cascading failure.First,this paper analyzes the causal relationship between microservices in cascading failures,and proposes a set of cascading failure prediction methods based on the cascading relationship awareness model among microservices.This method starts from the perspective of microservice load balancing and resource bottlenecks.It tracks the potential conduction path of cascading failures,and uses the conditions of cascading failures as the criterion to classify the nodes and microservices on this path.Then GRU neural network,which is based on time series,is used for cascading failure prediction.In addition,a feature selection method based on information entropy is designed to reduce parameter redundancy and improve the accuracy of model prediction.Secondly,this paper builds a microservice management platform and design a cascading failure prediction subsystem.The platform is deployed with the technology architecture of Docker,Kubernetes and Istio,and realizes related functional modules such as cluster management and service monitoring.The failure prediction subsystem is deployed in the service mesh in the form of a container,and interacts with various components of Istio.It is equipped with the cascading failure prediction method proposed in this paper,and realizes the prediction and alarm.Finally,this paper conducts system tests and method verification.The usability of each functional module of the system is tested.And the method is verified through comparative experiments.The experiments show that the method in this paper can accurately predict cascading failures and ensure the efficient operation of the system.