Research And Implementation Of High Availability Platform Based On Docker Swarm Cluster

With the development of cloud computing and micro-service technology,it is very common to deploy micro-services in the cloud platform.Containers become one of the technologies,and the resources in the container cloud platform can be fully utilized.Whether the service can run stably in the container is the main problem at this stage.In the actual production environment,the load of the container cloud platform tends to show periodicity,while the local load shows instability.For sudden changes in local load,it is necessary to dynamically limit the container resource usage based on the remaining resources of the node.For periodic load changes,it is necessary to increase the computing resources to maintain the quality of service during the high load period,and to reduce the idle resources according to the load amount to reduce the cost during the low load period.Therefore,the resource limitation and dynamic scaling of the cloud platform container cluster are very necessary.In the past,the research on cloud platform resource scheduling problem mainly focused on the scheduling of container creation,but the resource load in the container runtime is uncertain.Forecasting future loads by predictive models is a hot topic in current research.However,current research focuses on real-time prediction using a single predictive model.Due to large changes in resource load in a short period of time,it will cause low prediction accuracy and repeated scheduling problems.For the dynamic scheduling problem of cloud platform,this paper first studies the related frontier technologies in the container cloud platform.Secondly,the mainstream prediction algorithm is compared and analyzed.The machine learning and deep learning prediction models have the disadvantages of low precision and long time consumption for small samples.The problem of real-time load prediction is solved by the characteristics of gray model.Finally,the low-order prediction accuracy of traditional ARIMA model is not high,and the high-order parameter estimation is difficult.By combining Kalman filtering algorithm,a hybrid prediction algorithm based on ARIMA is proposed.Based on the above research,a Docker Swarm platform-based SME container resource load monitoring,forecasting and resource dynamic scheduling system is designed.By monitoring the container load in real time and combining the relationship between the container and the available resources of the node,the gray model and ARIMA are used respectively.The hybrid model predicts the container load and constrains and scales the container based on the predicted results.Finally,a comparative experiment was designed to test the performance of the platform.The experimental results show that the system ensures that the microservice container runs stably in the node under full load without exiting.The elastic expansion module ensures that the microservice cluster has different load conditions.The best performance makes the platform highly available.