CPVPA fails if some API is not discoverable

[rfranzke]

If there are APIServices that are not discoverable then the CPVPA is crashlooping:

$ kubectl get apiservice  | grep metrics-ad
v1beta1.custom.metrics.k8s.io          kube-system/kube-metrics-adapter   False (MissingEndpoints)   26d
v1beta1.external.metrics.k8s.io        kube-system/kube-metrics-adapter   False (MissingEndpoints)   26d

$ kubectl -n kube-system logs calico-typha-vertical-autoscaler-5557c6d7d-2sd7c
I0420 04:39:57.408180       1 autoscaler.go:46] Scaling namespace: kube-system, target: deployment/calico-typha-deploy
E0420 04:39:59.408093       1 autoscaler.go:49] failed to discover apigroup for kind "Deployment": unable to retrieve the complete list of server APIs: custom.metrics.k8s.io/v1beta1: the server is currently unable to handle the request, external.metrics.k8s.io/v1beta1: the server is currently unable to handle the request

Is this desired behaviour? If yes, why does the CPVPA need to discover the full API?

[MrHohn]

Thanks for reporting, I wasn't aware that our usage on the discovery API would be buggy like this. I will take a look when I get a chance.

[rfranzke]

@MrHohn any chance to fix this soon? I could also help if you point me to the right code location that needs adjustments.

[MrHohn]

@rfranzke Sorry for lacking update here, I haven't got a chance to take a closer look on this, so could definitely use your help if you are available :)

The exact code location is:

cluster-proportional-vertical-autoscaler/pkg/autoscaler/k8sclient/k8sclient.go

Lines 135 to 138 in a0338cc

	resourceLists, err := client.Discovery().ServerPreferredNamespacedResources()
	if err != nil {
	return "", nil, fmt.Errorf("failed to discover apigroup for kind %q: %v", kind, err)
	}

Is this desired behaviour? If yes, why does the CPVPA need to discover the full API?

I don't think CPVPA needs to discover the full API, just the relevant ones would be sufficient.

[rfranzke]

/assign