Infrastructure & Operations › Kubernetes & Orchestration

Node

A machine in the cluster that runs pods.

Also known as: kubernetes node, worker node, kubelet

A Node is a machine in a cluster — usually a virtual machine, sometimes bare metal — that actually runs your pods. Each node runs an agent called the kubelet, which starts and stops containers, plus a container runtime and networking components. The control plane decides which pods go where; the node does the work.

kubectl get nodes
kubectl describe node <name>      # capacity, allocatable, and what's scheduled
kubectl top node                  # current usage

A node can be marked unschedulable and drained so its pods move elsewhere, which is how you do maintenance or upgrades without downtime:

kubectl cordon <node>            # stop scheduling new pods here
kubectl drain <node> --ignore-daemonsets   # evict existing pods

The classic mistake is ignoring capacity. The scheduler places a pod based on its resource requests, not its real usage. If requests are missing or too small, many pods pile onto one node and compete for CPU and memory; if they’re too large, nodes sit half empty and cost more. Look at Allocatable versus Requests on a node to see how much room is really left.

The second mistake is assuming all nodes are identical. Nodes differ by size, labels and taints, and you steer pods with node affinity and taints and tolerations — for example keeping GPU work on GPU machines.

Managed Kubernetes hides most node operations, but the mental model still matters: pods are placed on bounded machines, and unhealthy nodes are replaced automatically. See cluster and container orchestration.