Skip to content

Latest commit

 

History

History
202 lines (175 loc) · 7.41 KB

README.md

File metadata and controls

202 lines (175 loc) · 7.41 KB

testutil

Go Documentation Build Status Go Report Card Coverage Status Maintainability

This library is a collection of helpers to ease implementing integration tests utilizing kind, helm and kubernetes without the need to install any CLI tools.

If it improves the code readability and ergonomics, the library will utilize panic for rare errors, e.g. filesystem issues, and performance is not a primary requirement. Therefore it should not be used as part of a production service.

Getting started

Before we can interact with a kubernetes cluster, we have to either create one or connect to an existing one. If you just want to connect to an external cluster feel free to skip the following section.

Creating a kind cluster

Create a new kind cluster with a randomized name is as simple as running:

cluster, err := kind.NewCluster()
if err != nil {}
defer cluster.Close()

However in most use-cases you want to provide more options to kind to configure it as you require it. You might want to set a deadline for the creation or use an existing cluster.

clusterOptions := []kind.ClusterOption{
    kind.ClusterWithWaitForReady(2*time.Minute),
}
if existingClusterName != "" {
    clusterOptions = append(clusterOptions,
        kind.ClusterWithName(existingClusterName),
        kind.ClusterUseExisting(),
        kind.ClusterDoNotDelete(),
    )
}
cluster, err := kind.NewCluster(clusterOptions...)
if err != nil {}
defer cluster.Close() // If do not delete is set, the cluster will not be deleted

The node image used by kind can be specified via kind.ClusterWithNodeImage. By default this will fallback to a sensible default, but should be set to the kubernetes version of your choice.

If you require even deeper configurability, you can pass in a cluster definition via kind.ClusterWithConfig.

Setting up helm

Before the helm-client can be setup you have to retrieve the raw kubeconfig.

NOTE:This will most likely change to rest.Config at some point in the future.

For an existing kind.Cluster this will look as follows:

kubeConfig, err := cluster.GetKubeConfig()
if err != nil {}

With the kubeconfig at our disposal the client can then be created:

helmClient, err = helm.NewClient(kubeConfig)
if err != nil {}
defer helmClient.Close()

It is important to always clean up the helm.Client via Close as it manages its own temporary cache and repository indices, which will otherwise leak.

To be able to install a remote chart, you will have to add your repositories, e.g.:

err := helmClient.AddRepository(&helm.RepositoryEntry{
    Name: "bitnami",
    URL:  "https://charts.bitnami.com/bitnami",
})
if err != nil {}

Installing a chart

You can install both remote and local charts. In context of this getting started guide we will install the bitnami/nginx-chart:

rls, err := helmClient.Install("bitnami/nginx", "", helm.ValuesOptions{})
if err != nil {}

The second parameter is the version to use. If it is an empty string, the most recent version will be installed.

helm.ValuesOptions is a struct offering several ways to provide both from filesystem and by definition. Before those are used they are merged according to helm's conventions.

By default a release name will be generated, but can be retrieved from the returned helm.Release.

The install procedure can be further customized using InstallOptions. For example to use a predefined release name:

rls, err := client.Install("bitnami/nginx", "", ValuesOptions{},
    InstallWithReleaseName(name),
)
if err != nil {}

Interacting with kubernetes objects

For most tests you will not stop after installing a helm chart, but rather you want to interact with the resulting objects.

So to show off some of the available helpers, let's create a kube.Client (also embed the kubernetes controller-runtime client) and illustrate some of its usage.

restConfig, err = cluster.GetRESTConfig()
if err != nil {}
k8sClient, err = NewClient(restConfig)
if err != nil {}

Let's create a context with a timeout first, so that commands can not exceed a deadline in our tests:

ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()

The installed nginx has a Deployment, so let's retrieve it first:

deployment := kube.DeploymentWithNamespacedName(rls.Namespace, rls.Name+"-nginx")
err := k8sClient.Get(ctx, NamespacedName(deployment), deployment)
if err != nil {}

We want to retrieve the scheduled pod, but at the beginning it won't be available, so let's wait until the deployment is scheduled.

err := k8sClient.WaitUntil(ctx, kube.DeploymentIsScheduled(deployment))
if err != nil {}

Once we know the deployment was scheduled, we can either retrieve pods with the required labels via k8sClient.List or for demonstration purposes find the replicasets for the deployment and then the pods for the active replicaset (using k8sClient.ListForOwner also particularly useful for jobs):

replicaSetList := &appsv1.ReplicaSetList{}
err := k8sClient.ListForOwner(context.Background(), replicaSetList, deployment)
if err != nil {}
if len(replicaSetList) < 1 {}
replicaSet := &replicaSetList.Items[0]
err = k8sClient.WaitUntil(ctx, kube.ReplicaSetIsAvailable(replicaSet), kube.ReplicaSetIsReady(replicaSet))
if err != nil {}
podList := &corev1.PodList{}
err = k8sClient.ListForOwner(ctx, podList, replicaSet)
if err != nil {}
if len(podList) < 1 {}
pod := &podList.Items[0]

Now we actually retrieved our nginx pod. At this point we probably want to interact with the pod, e.g. create a port forward, however let's wait until the pod it ready to retrieve traffic first:

err := k8sClient.WaitUntil(ctx, kube.PodIsReady(&pod))
if err != nil {}

The pod is ready, so we can create a port-forward:

pf, err := k8sClient.PortForward(pod, PortAny, 8080)
if err != nil {}
defer pf.Close()

The above code will use an available port on the host rather than a predefined one. So connecting to nginx might look as follows:

_, err := http.Get(fmt.Sprintf("http://localhost:%d", pf.LocalPort))

For some integration tests it might make sense to retrieve logs of a pod, conveniently k8sClient.Logs is here to help:

logs, err := k8sClient.LogsString(ctx, pod)

Last but not least events for a specific object can be retrieved using k8sClient.Events. This is particularly useful for operators which interact with several resources and create events for their interactions:

events, err := k8sClient.Events(ctx, pod)

Notes (temporary)

  • uses panic do not use in live code just tests
  • make TEST_FLAGS="-kind-cluster=testutil" test
  • still missing ds, statefulsets