What is the best approach for creating a controller in Kubernetes?

Question

I'm not quite sure if which of the following approaches is the better approach to create a controller in kubernetes however I know that:

• I don't want to create a custom resource by any means.
• I do only want to fetch information about k8s native resources (pods, ...) given that there might be a lot of pods in each namespace

I have seens some patterns like:

ctrl, err := controller.New("name-here", mgr, controller.Options{
    Reconciler: &ReconcilePod{Client: mgr.GetClient(), Logger: log},
})

which ReconcilePod is a struct that has a function Reconcile that keep whole business logic.

Another approach I have seens is like following:

type Controller struct {
    indexer  cache.Indexer
    queue    workqueue.RateLimitingInterface
    informer cache.Controller
}

and then defining shared informer and watcher etc. And the third pattern that I have seen is using operators

what I don't get perhaps is what is the main differences between mentioned approaches above and which one fits my need at scale.

Answer 1

If you don't want to "control" anything, there is no need to create a controller.

If you just want to "read" and "watch" resources, you can use client-go and see e.g. Extend Kubernetes via a shared informer for inspiration about how to read and watch resources.

To stay informed about when these events get triggered you can use a primitive exposed by Kubernetes and the client-go called SharedInformer, inside the cache package. Let’s see how it works in practice.

Controllers are more complex and contains a reconciliation loop since they should realize/manage a desired state.

An "operator" is a controller as well.

Answer 2

You can simply start with operator-sdk provided to create a controller easily in which the code will be generated and you need to add business logic to your objects. https://sdk.operatorframework.io/docs/building-operators/golang/quickstart/

mkdir memcached-operator
cd memcached-operator
operator-sdk init --domain example.com --repo github.com/example/memcached-operator

You can simply then create controller eith for existing kind or a new kind

operator-sdk create api --group cache --version v1alpha1 --kind Memcached --resource --controller

It is the simplest way to manage kinds and resources using operator sdk and no need to worry about complex code of client set