Behavior changes

Everything else in the v2 migration is a compile error. These two are not: your code builds and behaves differently. Both are in k8s, so skip this page if you do not use it, and see the v2 overview for the rest of the migration.

Node addresses prefer ExternalIP

k8s.FindNodeHostnameContextE, and GetServiceEndpointContextE for a NodePort service, now return the ExternalIP recorded on the Node object when one is present. They fall back to the internal hostname exactly as before when it is not.

Cloud controller managers record an instance’s public IP as an ExternalIP. Terratest previously ignored that field and, on AWS, called ec2:DescribeInstances to find the public IP instead. Reading the Node object is the same answer without the API call, so:

• ec2:DescribeInstances is no longer needed for this path
• k8s no longer depends on the aws module, which removes 22 AWS service clients from its dependency graph

What to check. If your cluster advertises an ExternalIP and your test previously received an internal hostname, it now receives the external address. That is the documented behavior and almost certainly what you wanted, but it is a different string. Tests that assert on the endpoint value, or that rely on reaching the node over its internal address, are the ones to look at.

Signatures are unchanged. The ExternalIP preference applies to every provider. For AWS-backed nodes that advertise no ExternalIP, a new pair of functions takes *KubectlOptions and consults a lookup you provide:

options := k8s.NewKubectlOptions("", kubeconfig, "default")
options.NodePublicIPLookup = aws.GetPublicIpsOfEc2InstancesContextE

hostname, err := k8s.FindNodeHostnameWithOptionsContextE(t, ctx, options, node)

NodePublicIPLookup is only consulted after the Node’s own ExternalIP has been checked, so most callers can leave it nil.

KubectlOptions carrying a RestConfig cannot be saved

json.Marshal of a KubectlOptions built by NewKubectlOptionsWithRestConfig now fails with an error wrapping k8s.ErrRestConfigNotSerializable. Match it with errors.Is, not ==: encoding/json returns a *json.MarshalerError around it. This affects k8s.SaveKubectlOptions, teststructure.SaveTestData, and any code of your own that marshals options.

This already failed in v1, with an opaque json: unsupported type: transport.WrapperFunc, because rest.Config holds func-typed fields that encoding/json rejects. What changed is that the error now says what is wrong and what to do instead.

It is deliberately an error rather than silently dropping the config. Dropped, the reloaded options would carry no cluster identity at all, fall back to the ambient kubeconfig, and run your test against a different cluster.

What to do. For staged tests, build options from a kubeconfig path or from in-cluster auth. Both round trip:

options := k8s.NewKubectlOptions(contextName, configPath, namespace)
// or
options := k8s.NewKubectlOptionsWithInClusterAuth()

If you need a rest.Config at runtime, keep building one, but rebuild it in each stage rather than saving it.