Kubernetes

Kubernetes is a runtime orchestration platform for containers. It abstracts away the machines underneath and automates the operational work: placing workloads, restarting them, scaling them.

The moving parts

• Control plane. Holds the cluster’s state and exposes an API server. People and tools talk to it through kubectl or a dashboard. Controllers inside it run control loops.
• Worker nodes. Application workloads run in pods on worker nodes. An agent on each node, the kubelet, talks to the control plane, and workloads are exposed on the network through Services.
• Custom resource definitions (CRDs). Extend the API with new kinds of resource, so things like monitoring, CI/CD toolchains or databases can be declared and managed the same way as everything else.

The idea worth keeping: reconciliation

You don’t tell Kubernetes what to do; you tell it what you want. A controller then loops forever: observe the actual state, compare it with the desired state, take a small action to close the gap, repeat. Scale a deployment from three replicas to five and nothing “runs a script”; a loop notices the difference and fixes it.

Once you see that loop, it turns up everywhere: GitOps, self-healing machine configuration, even balancing in a handstand (Balance Is a Control Loop). It’s also why Kubernetes so often sits underneath an Internal Developer Platform as its control plane.