Setup K3d/K3s for GitHub Actions

Install K3d/K3s and start a local Kubernetes cluster of a specific version.

K8s is Kubernetes. K3s is a lightweight K8s distribution. K3d is a wrapper to run K3s in Docker.

K3d/K3s are especially good for development and CI purposes, as it takes only 20-30 seconds of time till the cluster is ready. For comparison, Kind takes 1.5 minutes, Minikube takes 2 minutes till ready (as of Sep'2020).

Quick start

Start with the simplest way:

jobs:
  some-job:
    steps:
      - uses: nolar/setup-k3d-k3s@v1

Change versions with the verbose way:

jobs:
  some-job:
    steps:
      - uses: nolar/setup-k3d-k3s@v1
        with:
          version: v1.21  # E.g.: v1.21, v1.21.2, v1.21.2+k3s1
          github-token: ${{ secrets.GITHUB_TOKEN }}

Inputs

version

Required version of Kubernetes and/or K3s -- either full or partial.

The following notations are supported:

• v1.21.2+k3s1
• v1.21.2
• v1.21
• v1
• latest

Defaults to latest.

Keep in mind that K3d dates back only to v1.16. There are no 1.15 and older versions of K8s. Besides, 1.16 and 1.17 are broken and will not be fixed (see #11).

When the version is partial, the latest detected one will be used, as found in K3s releases, according to the basic semantical sorting (i.e. not by time of releasing).

k3d-tag

A tag/version of K3d to use. Corresponds to GitHub tags at https://github.com/rancher/k3d/releases. For example, v5.0.0. latest is also accepted, but converted to an empty string for the installation script.

By default (i.e. if no value is provided), the latest version is used.

k3d-name

A name of the cluster to be created.

By default (i.e. if no value is provided), K3d/K3s define their own name. Usually it is k3d-k3s-default.

Note: the name should not include the k3d- prefix, but must be used with it. The k3d- prefix is enforced by K3d and cannot be disabled.

k3d-args

Additional args to pass to K3d. See k3d cluster create --help for available flags.

github-token

A token for GitHub API, which is used to avoid rate limiting.

The API is used to fetch the releases from the K3s repository.

By default, or if it is empty, then the API is accessed anonymously, which implies the limit of approximately 60 requests / 1 hour / 1 worker.

Usage:

with:
  github-token: ${{ secrets.GITHUB_TOKEN }}

skip-creation

Whether to return from the action as soon as possible without the cluster creation (the cluster readiness is also skipped). This can be useful to only install the tools for manual cluster creation, or to parse the available versions and return them as the action's outputs.

By default (false), the cluster is created.

skip-readiness

Whether to return from the action as soon as possible, possibly providing a cluster that is only partially ready.

By default (false), the readiness is awaited by checking for some preselected resources to appear (e.g., for a service account named "default").

Outputs

k3d-version

The specific K3d version that was detected and used. E.g. v5.0.0.

k3s-version

The specific K3s version that was detected and used. E.g. v1.21.2+k3s1.

k8s-version

The specific K8s version that was detected and used. E.g. v1.21.2.

Examples

With the latest version of K3d/K3s/K8s:

steps:
  - uses: nolar/setup-k3d-k3s@v1

With the specific minor version of K8s, which implies the latest micro version of K8s and the latest possible version of K3s:

steps:
  - uses: nolar/setup-k3d-k3s@v1
    with:
      version: v1.21

With the very specific version of K3s:

steps:
  - uses: nolar/setup-k3d-k3s@v1
    with:
      version: v1.21.2+k3s1

The partial versions enable the build matrices with only the essential information in them, which in turn, makes it easier to configure GitHub branch protection checks while the actual versions of tools are upgraded:

jobs:
  some-job:
    strategy:
      fail-fast: false
      matrix:
        k8s: [ v1.21, v1.20, v1.19, v1.18 ]
    name: K8s ${{ matrix.k8s }}
    runs-on: ubuntu-22.04
    steps:
      - uses: nolar/setup-k3d-k3s@v1
        with:
          version: ${{ matrix.k8s }}

Multiple clusters in one job are possible, as long as there is enough memory (note: k3d- prefix is enforced by K3d):

jobs:
  some-job:
    name: Multi-cluster
    runs-on: ubuntu-22.04
    steps:
      - uses: nolar/setup-k3d-k3s@v1
        with:
          version: v1.20
          k3d-name: 1-20
      - uses: nolar/setup-k3d-k3s@v1
        with:
          version: v1.21
          k3d-name: 1-21
      - run: kubectl version --context k3d-1-20 
      - run: kubectl version --context k3d-1-21 

Custom version of K3d can be used, if needed:

jobs:
  some-job:
    name: Custom K3d version
    runs-on: ubuntu-22.04
    steps:
      - uses: nolar/setup-k3d-k3s@v1
        with:
          k3d-tag: v4.4.8
      - run: k3d --version

Custom args can be passed to K3d (and through it, to K3s & K8s):

jobs:
  some-job:
    name: Custom args
    runs-on: ubuntu-22.04
    steps:
      - uses: nolar/setup-k3d-k3s@v1
        with:
          k3d-args: --servers 2 --no-lb
      - run: kubectl get nodes  # there must be two of them

For real-life examples, see:

• https://github.com/nolar/kopf/actions (all "K3s…" jobs).