Installing GitHub Actions Runner Controller (ARC) on k3s
To reach a broader audience, this article has been translated from Japanese.
You can find the original version here.
Hello, I'm Shoji.
k3s currently supports x86_64, armhf, arm64/aarch64, and s390x architectures. Additionally, the runner image of GitHub Actions Runner Controller (ARC) supports linux/amd64 (i.e., x86_64) and linux/arm64 (i.e., arm64/aarch64).
Therefore, ARC can be executed in a k3s environment that is either x86_64 or arm64/aarch64.
For more information about GitHub Actions Runner Controller, please refer to the article "GitHub Actions Runner Controller (ARC) - On-demand execution of self-hosted runners in Kubernetes".
You might wonder why ARC instead of a Self-Hosted Runner. There is definitely a need in the IoT world to containerize. In the IoT world, not only computing resources but also various external devices are often handled, such as cameras and GPUs. It is necessary to verify whether applications using these external devices can operate in a Kubernetes environment. Therefore, we focused on using the actual k3s environment rather than a simple Self-Hosted Runner.
Deploying ARC
#Deploying ARC uses helm. This is almost no different from the procedure described in the introduced article.
The only difference is to set the environment variable KUBECONFIG before executing the helm command.
export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
Install ARC.
helm install arc \
--namespace arc-systems \
--create-namespace \
oci://ghcr.io/actions/actions-runner-controller-charts/gha-runner-scale-set-controller
Next, install actions-runner. As mentioned in the article by Mr. Kondo, a GitHub Personal Access Token (PAT) is required for installation.
GITHUB_CONFIG_URL="https://github.com/<your_account/repo>"
GITHUB_PAT="<PAT>"
helm install arc-runner-set \
--namespace arc-runners \
--create-namespace \
--set githubConfigUrl="${GITHUB_CONFIG_URL}" \
--set githubConfigSecret.github_token="${GITHUB_PAT}" \
oci://ghcr.io/actions/actions-runner-controller-charts/gha-runner-scale-set
Using a custom actions-runner image
#The actions-runner image provided by GitHub contains only basic binaries. It is generally assumed to work with the definition of a normal GitHub Actions workflow. However, I encountered an error when trying to execute a workflow that includes the execution of the sbt command. Upon investigation, I noticed a significant difference in the number of files in the /usr/bin directory between the Ubuntu managed by GitHub and the actions-runner image. Therefore, I decided to create a custom actions-runner image following the Ubuntu image managed by GitHub.
The Dockerfile is as follows.
FROM ghcr.io/actions/actions-runner:2.312.0
USER root
RUN apt update && apt install -y curl && \
curl -fLo /tmp/sbt.tgz https://github.com/sbt/sbt/releases/download/v1.9.8/sbt-1.9.8.tgz && \
tar zxf /tmp/sbt.tgz -C /usr/share && ln -s /usr/share/sbt/bin/sbt /usr/bin/sbt && rm -f /tmp/sbt.tgz
WORKDIR /home/runner
USER runner
This Dockerfile was built using GitHub Actions and pushed to GitHub Packages.
You can install the actions-runner using a custom container image with the following command.
export GITHUB_CONFIG_URL="https://github.com/<your_account/repo>"
export GITHUB_PAT="<PAT>"
helm install self-hosted \
--namespace arc-runners \
--create-namespace \
--set githubConfigUrl="${GITHUB_CONFIG_URL}" \
--set githubConfigSecret.github_token="${GITHUB_PAT}" \
--set template.spec.containers[0].name="runner" \
--set template.spec.containers[0].image="ghcr.io/takesection-sandbox/actions-runner:latest" \
--set template.spec.containers[0].command[0]="/home/runner/run.sh" \
oci://ghcr.io/actions/actions-runner-controller-charts/gha-runner-scale-set
Conclusion
#In IoT development, there are cases where special peripherals are used. In such cases, you may want to perform continuous integration (CI) in an environment closer to production rather than the standard cloud environment provided by GitHub Actions. Kubernetes has Device Plugins for handling external devices. USB devices can also be handled. In such an environment, ARC using a custom actions-runner is considered useful.
