Kubernetes List/Watch, Informer Mechanism, and Writing Controllers for Pods and Custom Resources

The article begins by describing Kubernetes as a container orchestration system and introduces the controller concept, where a controller watches resources and reconciles the actual state to the desired state.

It details the HTTP List and Watch mechanisms, showing how to list pods via GET /api/v1/namespaces/default/pods and watch changes with the ?watch=true query parameter, including sample curl commands and raw JSON event streams (ADDED, MODIFIED, DELETED).

Next, a minimal Go program that directly uses the HTTP API to watch pods is presented, followed by a more idiomatic controller built with the Kubernetes Go client, which creates a watcher , processes events in a work queue, and logs pod lifecycle events.

The Informer mechanism is then introduced as a higher‑level abstraction that caches resources locally, reduces API‑server load, improves performance, and simplifies controller code. The article outlines the internal components (Reflector, FIFO queue, Indexer) and the typical event‑handling flow.

SharedInformer is explained as a way to share a single List/Watch call and cache among multiple controllers, with code showing how to create a shared informer factory, register event handlers, and start the informer.

For custom resources, the article shows how to define a CRD (Foo), generate clientset, informers, and listers using k8s.io/code-generator , register the types in a scheme, and write a controller that watches Foo objects using the generated informer.

Finally, the article mentions higher‑level frameworks—controller‑runtime and Kubebuilder—that build on informers to provide leader election, reconciliation loops, and code generation, offering an even more streamlined way to develop Kubernetes operators.