When Kubernetes runs a container on a node, the control plane decides what should run, but the node must carry out that work. The kubelet is the node-level process that communicates with the control plane and manages the lifecycle of containers through the container runtime. section 3.1.1
The kubelet operates alongside the operating system and container runtime. It starts and stops containers as requested and can take local action without waiting for the control plane. section 3.1.1
Separate kubelet responsibilities from scheduler responsibilities:
- Restarts a container when a liveness or readiness failure requires a local Pod restart. section 5.1
- Runs the containers assigned to its node and manages their lifecycle. section 3.1.1
- Does not reallocate Pods after a node disappears; the scheduler detects the node problem and places a replacement Pod elsewhere. section 5.1
A failed container
If a container fails a liveness or readiness check, the kubelet handles the restart locally. If the entire node fails, the control plane and scheduler arrange for a new Pod on a healthy node. section 5.1
Use the node-process diagram to locate the kubelet beside the operating system and container runtime. section 3.1.1


