KubeCon + CloudNativeCon China 2018

A Kubernetes Node depends on many on-host softwares and configurations, including container runtime, network plugins, kubelet. The maintenance of these dependencies is tedious and error-prone. At Alibaba and Ant Financial, a typical cluster admin needs to maintain tens of thousands of Kubernetes' Nodes on average. We developed Node operator to make the task easy and less risky.
In this talk, we will share how we use Node Operator to maintain the software and configuration of the node. The declaritive API we designed allows the cluster admins to interactive with the node CRD resources to manage the life cycle of any Node. The Node Operator is also responsible for reacting to Nodes' status changes, and take recovery actions when needed. The Node Operator has an extensible design, which allows it to manage other on-host softwares that is not part of Kubernetes.