Research And Implementation Of Scheduling Self-healing System Based On Kubernetes Cluster

This thesis aims to provide better self-healing effect for container clusters by optimizing the resource scheduling and auto-scaling policies of Kubernetes.To this end,this thesis optimizes the default resource scheduling and auto-scaling policies,and implements a self-healing module to extend the fault detection of schedulers and nodes,which significantly improves the self-healing ability of the cluster for node failures and the scaling effect of the repaired scheduling.First,the self-healing mechanism of Kubernetes is studied and the relationship between resource scheduling,auto-scaling and self-healing is analyzed.For the shortcomings of the default scheduling and scaling policies,this thesis provides an optimized design.Through node health detection and probe mechanism,cluster nodes and Pod failures are detected in time,and the optimized scheduling and scaling policies are used to perform scheduling self-healing.The optimized policy can ensure the high availability of cluster applications,and at the same time make the resource allocation of the whole cluster more balanced,the service operation more stable and the health state more durable.Second,in terms of resource scheduling,the scheduling algorithm of the default scheduling policy is optimized for the IO-intensive scenarios often faced by clusters,which significantly improves resource allocation; the real-time data indicators of nodes and static data indicators of nodes’ own performance are considered,including CPU,memory,network IO,disk IO,number of deployed Pods,etc.,to achieve a more comprehensive and balanced cluster Resource status consideration.To address the shortcomings of single metric and frequent scaling of auto-scaling policy,this thesis adds weight tolerance and step control to achieve faster scaling and more stable than the default scaling policy.We also implement a self-healing module for nodes based on auto-scaling,and extend the default Kube-scheduler to implement our own scaling scheduler.Finally,to verify the improved effect of the optimized scheduling and scaling strategy,a Kubernetes test cluster was built and tested for comparison.The experimental results show that the optimized policy can significantly improve the scheduling self-healing effect and the efficiency of auto-scaling,thus enhancing the self-healing ability of the cluster system for node failures and the scheduling scaling effect after repair.