A Kubernetes worker node is a machine that supplies computing resources for running containers in a cluster. It is usually a virtual machine, but it can also be a bare-metal machine. Nodes are managed as a shared pool rather than as individually special hosts, so workloads can move when a machine fails. section 1.2 section 3.1.1
A worker node combines platform components with automatic workload management:
- Runs the kubelet, which communicates with the control plane and manages container lifecycles. section 3.1.1
- Uses a container runtime to execute containers inside Pods. section 3.1 section 3.1.1
- Starts and stops containers as instructed, restarts crashed containers locally, and can reclaim memory when resources are low. section 3.1.1
- Receives Pods selected by the scheduler according to available resources and placement requirements. section 5.1
The control plane decides where a Pod should run; the worker node performs the execution. If a node disappears, the scheduler can place replacement Pods on a healthy node to preserve the desired state. This differs from a local kubelet restart: node failure triggers reallocation, while container health failures trigger local restarts. section 5.1 section 3.1.1
Replacing a failed node
A Deployment requesting two Pod replicas can have those replicas scheduled across available nodes. When one node fails, the control plane detects the problem and the scheduler creates a replacement Pod on a healthy node, restoring the requested replica count. section 5.1
Use the cluster diagram to see worker nodes connecting to a hosted control plane in a cloud-based deployment. section 3.1.1
