Design And Implementation Of A Cross-cluster Management For Kubernetes

With the development of cloud computing,more and more companies choose to adopt the multi-cloud strategies in the application platform.On the one hand,this approach can avoid companies to be limited to a vendor,enhance the ability to deliver available services and other effects.On the other hand,this approach could introduce some potential issues.For example,it would significantly hurt the management performance with unexpected spread of the project resources in data centers and private clouds.It is extremely important to implement a unified management scheme of the resources in the multi-cloud platforms.As Kubernetes becomes the most concerned solutions based on Docker container of Platform as a Service,many large enterprises like HUAWEI,CCMC(China Mobile Communications Corporation)try to implement their own Kubernetes multi-cluster managernent.This paper provides a cross-cluster management system for Kubernetes for harmonycloud.To achieve this goal,this paper solves several problems,including cross-cluster service discovery,cross-cluster service scheduling and service consistency management.Firstly,base on the existing service discovery and load balancing solution of Kubemetes,this paper designs a new architecture,which combines server and client service discovery architecture.Users can access the service externally via domain name,while enjoying the benefit of user-request load balancing.Secondly,this paper designs a weight-based scheduling strategy to solve the multi-cluster scheduling problem.When a user deploys a service,there could be various constraints of the application in relation to different scenarios.The scheduling policy in this paper offers great flexibility for users on the constraints of service deployment.Thirdly,automatic service recovery is solved by service consistency management.service consistency management maintain the services that are running in all clusters consistent with the service information in the storage system.Service consistency management not only realizes real-time deployment service update,but also provides service rescheduling when cluster is in the abnormal state,which solves the problem of automatic service recovery.