Issue deploying `deploy/kne/external-multinode.yaml` on existing Kubernetes cluster

[nleiva]

Hi,

I'm attempting to deploy deploy/kne/external-multinode.yaml on an existing Kubernetes cluster. I modified the Docker network name from multinode to docker0 to avoid manually creating a new network (network: docker0).

The MetalLB pods appear healthy.

$ kubectl get pods -n metallb-system
NAME                         READY   STATUS    RESTARTS   AGE
controller-fdfbfbc77-jnf4q   1/1     Running   0          4m40s
speaker-l8g66                1/1     Running   0          4m40s
speaker-lp7bv                1/1     Running   0          4m40s
speaker-vrj7z                1/1     Running   0          4m40s

However, I get the following error message, which I don't fully understand:

$ kne deploy deploy/kne/external-multinode.yaml
I0306 18:34:02.667047  465400 deploy.go:195] Deploying cluster...
I0306 18:34:02.667437  465400 deploy.go:381] Deploy is a no-op for the external cluster type
I0306 18:34:02.667458  465400 deploy.go:199] Cluster deployed
...
I0306 18:34:07.049975  465400 deploy.go:1348] Waiting on deployment "metallb-system" to be healthy
Error: failed to deploy ingress: metallb not healthy: invalid object type: *v1.Deployment

Is this the deployment is complaining about?

$ kubectl get deploy -n metallb-system
NAME         READY   UP-TO-DATE   AVAILABLE   AGE
controller   1/1     1            1           28m

Thanks in advance for your help!

[chrisy]

I can't help with that specific issue, but fwiw on my physical cluster I setup metallb independently and tell KNE to not try to (I set the manifest to "").

[nleiva]

Long time no see Chris! I'll give that a go, thank you!

[nleiva]

A small update in case someone runs into this. TL;DR I'm probably running a K8s version that's too new.

The error is triggered in deploy/deploy.go

func deploymentHealthy(ctx context.Context, c kubernetes.Interface, name string) error {
	log.Infof("Waiting on deployment %q to be healthy", name)
	w, err := c.AppsV1().Deployments(name).Watch(ctx, metav1.ListOptions{})
	// ...
	ch := w.ResultChan()
	for {
		select {
		//...
		case e, ok := <-ch:
			if !ok {
				return fmt.Errorf("watch channel closed before %q healthy", name)
			}

			d, ok := e.Object.(*appsv1.Deployment)
			
			if !ok {
				return fmt.Errorf("invalid object type: %T", d)
			}

I added this before the return statement:

status, ok := e.Object.(*metav1.Status)
if ok {
	fmt.Printf("EVENT -> %v, %v\n", status.Status, status.Message)
}

Got: EVENT -> Failure, Timeout: Too large resource version: 2906652, current: 2903959

According to ChatGPT, in Kubernetes, "Too large resource version" means a client (like a kubelet or operator) is trying to access the Kubernetes API using a resource version that is older than the current version held by the API server's watch cache, and therefore out of date. This can lead to loop errors.