大型Kubernetes集群运维管理实践.pdf

大型Kubernetes集群运维管理实践 Managing Large-Scale Kubernetes Clusters Effectively and Reliably Agenda • Background and Motivation • Design Concept • Cluster Management Operators • Q A Background: Cluster Scale ● Tens of clusters ○ Tens of thousands of nodes in one cluster ● Hundreds of thousands of pods ○ Tens of thousands of jobs ● Resource cost is huge Motivation ● Cluster Life Cycle Management ○ Create ○ Delete ○ Upgrade ○ … ● Fault self-recovery: ○ hardware failure ○ component service exception ● Change Controllable Design Concept: Imperative ● Imperative ○ Tell What How ○ Tell How Initial State Final State Design Concept: Declarative == ● Declarative ○ Tell What ○ Not How Final State - Controller Current State Design Concept: Self-recovery ⚫ Observe watch actual state Disturbance ⚫ Analyze Analyze Action difference from desired and actual state Cluster ⚫ Action - Operators Cluster Final State change to desired state Cluster Observe Current State == Cluster-Operator