Research And Implementation Of High Availability For Container Cloud Platforms Based On Kubernetes

With the continuous development of cloud computing and container technology,container cloud has become the preferred strategy for enterprises to build efficient,flexible,and highly available application systems.Although container cloud,with Kubernetes as its core technology,provides a more efficient,flexible,and reliable way to build,deploy,and run containerized applications,there are still many shortcomings in terms of high availability.Kubernetes,with etcd as its backend storage,experiences significant latency in write requests as the cluster scales up.A large number of concurrent requests can cause exceptions in the API server,resulting in unavailability of container services.The default Horizontal Pod Autoscaler(HPA)in Kubernetes has a certain lagging issue,and a sudden surge in traffic may cause the application to crash.This Thesis mainly focuses on the analysis and implementation of high availability in the container cloud platform from three dimensions: storage,server-side,and scalability.Firstly,a multi-etcd cluster storage system is designed and implemented,proposing a cluster resource splitting strategy to distribute the read-write pressure of a single etcd cluster and implementing storage management modules for regular backup,health check,and fault recovery of etcd clusters.Secondly,In terms of the server-side,a high availability deployment solution has been proposed,which implements a proactive health check module and an admission control module based on the Admission Webhook mechanism..Then,a collaborative strategy between HPA and Vertical Pod Autoscaler(VPA)is designed,combining VPA resource recommendations with the flexibility of HPA scaling to more accurately respond to workload changes.Finally,a HPA strategy based on traffic trend prediction is proposed,which utilizes custom metric data to proactively scale the number of Pod replicas based on workload trend changes,effectively handling sudden traffic spikes.The test results show that the improved strategies for high availability in the container cloud proposed in this Thesis can enhance high availability at the levels of storage,server-side,and application scalability,thus improving the overall high availability of the container cloud platform.