Test your Python Kubernetes app/operator end-to-end with kind https://pypi.org/project/pytest-kind/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Henning Jacobs d277f1226c v22.9.0 3 months ago
examples/aiohttp-helloworld readme for example 3 years ago
pytest_kind kind v0.15, k8s v1.25 3 months ago
tests kind v0.15, k8s v1.25 3 months ago
.flake8 kube_cluster fixture 3 years ago
.gitignore add mypy type validation 3 years ago
.pre-commit-config.yaml update dependencies 4 months ago
.travis.yml Travis CI 3 years ago
LICENSE add license 3 years ago
Makefile remove obsolete github.com mirror 3 months ago
README.md update to kind v0.10 and K8s 1.20.2 2 years ago
poetry.lock kind v0.15, k8s v1.25 3 months ago
pyproject.toml v22.9.0 3 months ago



Build Status PyPI PyPI - Python Version License CalVer

Test your Python Kubernetes app/operator end-to-end with kind and pytest.

pytest-kind is a plugin for pytest which provides the kind_cluster fixture. The fixture will install kind 0.10.0, create a Kubernetes 1.20 cluster, and provide convenience functionality such as port forwarding.


Install pytest-kind via pip or via poetry, e.g.:

poetry add --dev pytest-kind

Write your pytest functions and use the provided kind_cluster fixture, e.g.:

def test_kubernetes_version(kind_cluster):
    assert kind_cluster.api.version == ('1', '20')

To load your custom Docker image and apply deployment manifests:

import requests
from pykube import Pod

def test_myapp(kind_cluster):
    kind_cluster.kubectl("apply", "-f", "deployment.yaml")
    kind_cluster.kubectl("rollout", "status", "deployment/myapp")

    # using Pykube to query pods
    for pod in Pod.objects(kind_cluster.api).filter(selector="app=myapp"):
        assert "Sucessfully started" in pod.logs()

    with kind_cluster.port_forward("service/myapp", 80) as port:
        r = requests.get(f"http://localhost:{port}/hello/world")
        assert r.text == "Hello world!"

See the examples directory for sample projects and also check out kube-web-view which uses pytest-kind for its e2e tests.

KindCluster object

The kind_cluster fixture is an instance of the KindCluster class with the following methods:

  • load_docker_image(docker_image): load the specified Docker image into the kind cluster
  • kubectl(*args): run the kubectl binary against the cluster with the specified arguments. Returns the process output as string.
  • port_forward(service_or_pod_name, remote_port, *args): run "kubectl port-forward" for the given service/pod and return the (random) local port. To be used as context manager ("with" statement). Pass the namespace as additional args to kubectl via "-n", "mynamespace".

KindCluster has the following attributes:

  • name: the kind cluster name
  • kubeconfig_path: the path to the Kubeconfig file to access the cluster
  • kind_path: path to the kind binary
  • kubectl_path: path to the kubectl binary
  • api: pykube HTTPClient instance to access the cluster from Python

You can also use KindCluster directly without pytest:

from pytest_kind import KindCluster

cluster = KindCluster("myclustername")
cluster.kubectl("apply", "-f", "..")
# ...

Pytest Options

The kind cluster name can be set via the --cluster-name CLI option.

The kind cluster is deleted after each pytest session, you can keep the cluster by passing --keep-cluster to pytest.

Note that you can use the PYTEST_ADDOPTS environment variable to pass these options to pytest. This also works if you call pytest from a Makefile:

# for test debugging: don't delete the kind cluster
PYTEST_ADDOPTS=--keep-cluster make test


  • The kind_cluster fixture is session-scoped, i.e. the same cluster will be used across all test modules/functions.
  • The kind and kubectl binaries will be downloaded once to the local directory ./.pytest-kind/{cluster-name}/. You can use them to interact with the cluster (e.g. when --keep-cluster is used).
  • Some cluster pods might not be ready immediately (e.g. kind's CoreDNS take a moment), add wait/poll functionality as required to make your tests predictable.