[hetzner] cluster-autoscaler --nodes flag uses wrong format

[bjornharrtell]

Summary

When kops generates the cluster-autoscaler --nodes flag for Hetzner instance groups, it produces a 3-field format (min:max:name.cluster) that the Hetzner autoscaler cloud provider does not recognise. The autoscaler starts but no node groups are registered, so scale-out never happens.

Root cause

GetClusterAutoscalerNodeGroups() in upup/pkg/fi/cloudup/template_functions.go sets group.Other = ig.Name + "." + cluster.Name for all non-GCE cloud providers. This produces e.g.:

--nodes=1:5:nodes-hel1.k8s-test-135

The Hetzner autoscaler cloud provider (upstream code) parses the --nodes argument expecting 5 fields: min:max:instanceType:region:name. The instanceType and region fields are required for Hetzner to create correctly-typed servers in the correct location.

Expected format

--nodes=1:5:cpx32:hel1:nodes-hel1

For Hetzner, region is the Hetzner location name, which equals the subnet name stored in ig.Spec.Subnets[0].

Fix

Add a Hetzner branch in GetClusterAutoscalerNodeGroups():

} else if cluster.GetCloudProvider() == kops.CloudProviderHetzner {
    region := ig.Spec.Subnets[0]
    group.Other = fmt.Sprintf("%s:%s:%s", ig.Spec.MachineType, region, ig.Name)
}

Fix included in the PR that accompanies this issue.