In Kubernetes, you rarely interact with individual worker nodes. Instead, all the API requests go to the control plane. The control plane then decides on the actions to take based on the requests, and then it communicates with the worker nodes.
The interaction with the control plane can take several forms:
- Using the kubectl CLI
- Using a web dashboard
- Using the Kubernetes API from inside an application other than kubectl
Control plane nodes usually run the API server, scheduler, a configuration store (etcd), and possibly some additional processes to handle the specific needs. For example, Kubernetes clusters deployed in a public cloud such as Google Cloud Platform have cloud controllers running on control plane nodes. The cloud controller interacts with the cloud provider's API to replace the failed machines, provision load balancers, or assign external IP addresses.
