Research And Design Of Container Resource Scheduling Strategy Based On Kubernetes Cluster

The formation of a new generation of Caa S cloud service pattern provides users with a rapid construction of an agile production environment,and speeds up the product development and release cycle.Container virtualization technology,represented by Docker,has been rapidly applied in the industry due to its advantages of lightweight,easy to transplant and quick to start up.Subsequently,Kubernetes,which is used to manage and deploy large-scale cluster containers,has become the mainstream container orchestration system in the industry due to its powerful management function,good compatibility and open source.However,at present,the most cores scheduling function of Kubernetes only contains a single scheduling strategy,which is difficult to adapt to the requirements of complex and changeable production scenarios.Therefore,the Kubernetes scheduler is targeted as the main research object in this thesis.Scheduling process and the current related scheduling mechanism are deeply studied,and the specific production environment requirements and the defects existing in the existing scheduling mechanism are optimized and improved.The main contents are as follows:1.Static scheduling algorithm library is expanded.According to the disaster tolerance requirements of specific Pod applications,a safe KNN node classification uniform scheduling strategy is designed in the form of plug-in.Based on this scheduling strategy,KNN node load classification is carried out in advance to filter the nodes under heavy load.Meanwhile,the alarm mechanism is triggered to inform the user to deal with the crisis node in a timely manner.Then,the multi-copy Pod application is distributed and uniformly deployed to the secure cluster node meeting the requirements.The experimental results show that this scheduling mechanism can meet the disaster tolerance requirements and protect the nodes,and it can effectively shorten the scheduling time compared with the default built-in scheduling strategy in the case of high-concurrency Pod scheduling.2.Dynamic load balancing scheduling is optimized.In view of the latest v1.15 version of Kubernetes static load balancing scheduling strategy,the resources assessment factors are expanded.AHP(analytic hierarchy process)is applied to determine the node real-time load and performance resource assessment factor weights,and comprehensively weight node performance and real-time load to score and screen target node.In addition,weighted least connections algorithm is used for multiple secondary screening score the highest target node.The experimental results show that the optimal strategy is more effective than the original scheduling strategy when Pod is scheduled concurrently in the cluster.3.Dynamic priority preemptive scheduling is designed.Accoding to the analysis on the basis priority in the latest v1.15 version Kubernetes object static priority preemptive scheduling,and depending on the type of controller manager and the current load status of cluster Pod type and prioritization are determined.Comprehensively taking into the Qo S of the Pod,resource capacity and the execution time,a dynamic priority preemptive scheduling strategy is designed,which can promote high-priority Pod to preempt low-priority Pod in the condition of insufficient node resources.Experimental results show that this optimization strategy can effectively Pod preemption.Compared with single static preemption and dynamic preemption scheduling,this optimization strategy can effectively improve the utilization of cluster resources and shorten the average scheduling time of the cluster.