Testing Cluster API
This document presents testing guideline and conventions for Cluster API.
IMPORTANT: improving and maintaining this document is a collaborative effort, so we are encouraging constructive feedback and suggestions.
Unit tests
Unit tests focus on individual pieces of logic - a single func - and don’t require any additional services to execute. They should be fast and great for getting the first signal on the current implementation, but unit test have the risk that to allow integration bugs to slip through.
Historically, in Cluster API unit test were developed using go test, gomega and the fakeclient; see the quick reference below.
However, considered some changes introduced in the v0.3.x releases (e.g. ObservedGeneration, Conditions), there is a common agreement among Cluster API maintainers that usage fakeclient should be progressively deprecated in favor of usage of envtest; see the quick reference below.
Integration tests
Integration tests are focuses on testing the behavior of an entire controller or the interactions between two or more Cluster API controllers.
In older versions of Cluster API, integration test were based on a real cluster and meant to be run in CI only; however, now we are considering a different approach base on envtest and with one or more controllers configured to run against the test cluster.
With this approach it is possible to interact with Cluster API like in a real environment, by creating/updating Kubernetes objects and waiting for the controllers to take action.
Please note that while using this mode, as of today, when testing the interactions with an infrastructure provider some infrastructure components will be generated, and this could have relevant impacts on test durations (and requirements).
While, as of today this is a strong limitation, in the future we might consider to have a “dry-run” option in CAPD or a fake infrastructure provider to allow test coverage for testing the interactions with an infrastructure provider as well.
Running unit and integration tests
Using the test
target through make
will run all of the unit and integration tests.
End-to-end tests
The end-to-end tests are meant to verify the proper functioning of a Cluster API management cluster in an environment that resemble a real production environment.
Following guidelines should be followed when developing E2E tests:
- Use the [Cluster API test framework].
- Define test spec reflecting real user workflow, e.g. [Cluster API quick start].
- Unless you are testing provider specific features, ensure your test can run with different infrastructure providers (see Writing Portable Tests).
See e2e development for more information on developing e2e tests for CAPI and external providers.
Running the end-to-end tests
make docker-build-e2e
will build the images for all providers that will be needed for the e2e test.
make test-e2e
will run e2e tests by using whatever provider images already exist on disk.
After running make docker-build-e2e
at least once, this can be used for a faster test run if there are no provider code changes.
Additionally, test-e2e
target supports the following env variables:
GINKGO_FOCUS
to set ginkgo focus (default empty - all tests)GINKGO_NODES
to set the number of ginkgo parallel nodes (default to 1)E2E_CONF_FILE
to set the e2e test config file (default to ${REPO_ROOT}/test/e2e/config/docker.yaml)ARTIFACTS
to set the folder where test artifact will be stored (default to ${REPO_ROOT}/_artifacts)SKIP_RESOURCE_CLEANUP
to skip resource cleanup at the end of the test (useful for problem investigation) (default to false)USE_EXISTING_CLUSTER
to use an existing management cluster instead of creating a new one for each test run (default to false)GINKGO_NOCOLOR
to turn off the ginko colored output (default to false)
Quick reference
envtest
envtest is a testing environment that is provided by the controller-runtime project. This environment spins up a local instance of etcd and the kube-apiserver. This allows test to be executed in an environment very similar to a real environment.
Additionally, in Cluster API there is a set of utilities under test/helpers that helps developers in setting up a envtest ready for Cluster API testing, and most specifically:
- With the required CRDs already pre-configured.
- With all the Cluster API webhook pre-configured, so there are enforced guarantees about the semantic accuracy of the test objects you are going to create.
This is an example of how to create an instance of envtest that can be shared across all the tests in a package;
by convention, this code should be in a file named suite_test.go
:
var (
testEnv *helpers.TestEnvironment
ctx = context.Background()
)
func TestMain(m *testing.M) {
// Bootstrapping test environment
testEnv = helpers.NewTestEnvironment()
go func() {
if err := testEnv.StartManager(); err != nil {
panic(fmt.Sprintf("Failed to start the envtest manager: %v", err))
}
}()
// Run tests
code := m.Run()
// Tearing down the test environment
if err := testEnv.Stop(); err != nil {
panic(fmt.Sprintf("Failed to stop the envtest: %v", err))
}
// Report exit code
os.Exit(code)
}
Most notably, envtest provides not only a real API server to user during test, but it offers the opportunity to configure one or more controllers to run against the test cluster; by using this feature it is possible to use envtest for developing Cluster API integration tests.
func TestMain(m *testing.M) {
// Bootstrapping test environment
...
if err := (&MyReconciler{
Client: testEnv,
Log: log.NullLogger{},
}).SetupWithManager(testEnv.Manager, controller.Options{MaxConcurrentReconciles: 1}); err != nil {
panic(fmt.Sprintf("Failed to start the MyReconciler: %v", err))
}
// Run tests
...
}
Please note that, because envtest uses a real kube-apiserver that is shared across many tests, the developer should take care of ensuring each test run in isolation from the others, by:
- Creating objects in separated namespaces.
- Avoiding object name conflict.
However, developers should be aware that in some ways, the test control plane will behave differently from “real” clusters, and that might have an impact on how you write tests.
One common example is garbage collection; because there are no controllers monitoring built-in resources, objects do not get deleted, even if an OwnerReference is set up; as a consequence, usually test implements code for cleaning up created objects.
This is an example of a test implementing those recommendations:
func TestAFunc(t *testing.T) {
g := NewWithT(t)
// Generate namespace with a random name starting with ns1; such namespace
// will host test objects in isolation from other tests.
ns1, err := testEnv.CreateNamespace(ctx, "ns1")
g.Expect(err).ToNot(HaveOccurred())
defer func() {
// Cleanup the test namespace
g.Expect(testEnv.DeleteNamespace(ctx, ns1)).To(Succeed())
}()
obj := &clusterv1.Cluster{
ObjectMeta: metav1.ObjectMeta{
Name: "test",
Namespace: ns1.Name, // Place test objects in the test namespace
},
}
// Actual test code...
}
In case of object used in many test case within the same test, it is possible to leverage on Kubernetes GenerateName
;
For objects that are shared across sub-tests, ensure they are scoped within the test namespace and deep copied to avoid
cross-test changes that may occur to the object.
func TestAFunc(t *testing.T) {
g := NewWithT(t)
// Generate namespace with a random name starting with ns1; such namespace
// will host test objects in isolation from other tests.
ns1, err := testEnv.CreateNamespace(ctx, "ns1")
g.Expect(err).ToNot(HaveOccurred())
defer func() {
// Cleanup the test namespace
g.Expect(testEnv.DeleteNamespace(ctx, ns1)).To(Succeed())
}()
obj := &clusterv1.Cluster{
ObjectMeta: metav1.ObjectMeta{
GenerateName: "test-", // Instead of assigning a name, use GenerateName
Namespace: ns1.Name, // Place test objects in the test namespace
},
}
t.Run("test case 1", func(t *testing.T) {
g := NewWithT(t)
// Deep copy the object in each test case, so we prevent side effects in case the object changes.
// Additionally, thanks to GenerateName, the objects gets a new name for each test case.
obj := obj.DeepCopy()
// Actual test case code...
}
t.Run("test case 2", func(t *testing.T) {
g := NewWithT(t)
obj := obj.DeepCopy()
// Actual test case code...
}
// More test cases.
}
fakeclient
fakeclient is another utility that is provided by the controller-runtime project. While this utility is really fast and simple to use because it does not require to spin-up an instance of etcd and kube-apiserver, the fakeclient comes with a set of limitations that could hamper the validity of a test, most notably:
- it does not handle properly a set of field which are common in the Kubernetes API objects (and Cluster API objects as well)
like e.g.
creationTimestamp
,resourceVersion
,generation
,uid
- API calls does not execute defaulting or validation webhooks, so there are no enforced guarantee about the semantic accuracy of the test objects.
Historically, fakeclient is widely used in Cluster API, however, given the growing relevance of the above limitations with regard to some changes introduced in the v0.3.x releases (e.g. ObservedGeneration, Conditions), there is a common agreement among Cluster API maintainers that usage fakeclient should be progressively deprecated in favor of usage of envtest.
ginkgo
Ginkgo is a Go testing framework built to help you efficiently write expressive and comprehensive tests using Behavior-Driven Development (“BDD”) style.
While Ginkgo is widely used in the Kubernetes ecosystem, Cluster API maintainers found the lack of integration with the most used golang IDE somehow limiting, mostly because:
- it makes interactive debugging of tests more difficult, since you can’t just run the test using the debugger directly
- it makes it more difficult to only run a subset of tests, since you can’t just run or debug individual tests using an IDE,
but you now need to run the tests using
make
or theginkgo
command line and override the focus to select individual tests
In Cluster API you MUST use ginkgo only for E2E tests, where it is required to leverage on the support for running specs
in parallel; in any case, developers MUST NOT use the table driven extension DSL (DescribeTable
, Entry
commands)
which is considered unintuitive.
gomega
Gomega is a matcher/assertion library. It is usually paired with the Ginkgo BDD test framework, but it can be used with other test frameworks too.
More specifically, in order to use Gomega with go test you should
func TestFarmHasCow(t *testing.T) {
g := NewWithT(t)
g.Expect(f.HasCow()).To(BeTrue(), "Farm should have cow")
}
In Cluster API all the test MUST use Gomega assertions.
go test
go test testing provides support for automated testing of Go packages.
In Cluster API Unit and integration test MUST use go test.