Persistent Storage And Scheduling Technology Of Docker Stateful Service

With the development of container technology,docker has become one of the most popular open source container tools.In the whole life cycle from application development to release,docker can not only save the cost of maintenance development,simplify the online migration process of application services,but also improve the horizontal expansion ability of application services.At present,there are several problems in persistent storage and scheduling of stateful services on the container cloud platform.First,the stateful service processing business needs to rely on the previous request,and the stateful service container node needs to store the business data related to the request.When the container fails,the business data in the stateful service container node is at risk of loss and cannot be migrated across hosts.Therefore,the persistent storage problem of stateful services is one of the key problems to be solved in the container cloud.Second,for stateful services,the docker cluster scheduling strategy does not consider the future workload of physical nodes,and the scheduling factors are not comprehensive enough.This leads to the possibility that the container may be allocated to the physical nodes that are about to be overloaded,and the resource distribution of the container cluster is unbalanced.Therefore,the problem of stateful service scheduling is one of the important problems to improve the performance of container cloud.Firstly,aiming at the problem of persistent storage of stateful service,this paper takes ceph distributed storage as the back-end storage of the container,and designs a general persistent storage technology for database services.This technology shortens the downtime recovery time of database service and improves the reliability of database service data.Secondly,aiming at the problem of the docker cluster scheduling strategy does not consider the future workload of nodes,the paper proposes a read-write load forecasting model based on ARIMA-BP combination.It models and predicts the read-write load of container in the physical node,so that the docker cluster can predict the read-write low load physical nodes.Finally,aiming at the problem of the default scheduling strategy only considers the CPU cores,this paper additionally considers the memory,IO and network factors,and uses the genetic algorithm based on elitist reservation strategy to implement stateful service scheduling strategy.The experimental results show that the combined model is more efficient than the single model,and the scheduling strategy improves the load balancing ability of container cluster scheduling stateful service.