DEV Community

Cover image for Create a Docker based Self Hosted GitHub runner Windows container
Marcel.L
Marcel.L

Posted on • Updated on

Create a Docker based Self Hosted GitHub runner Windows container

Overview

All the code used in this tutorial can be found on my GitHub project: docker-github-runner-windows.

Welcome to Part 1 of my series: Self Hosted GitHub Runner containers on Azure.

In part one of this series, we will focus and look at how we can create a windows container image using docker that will essentially be a packaged up image we can use to deploy and run self hosted GitHub runners as containers. We will focus more on the docker image itself and how we can build our image and run our image on a local server or VM running docker for windows and also scaling out multiple instances of our image using docker-compose.

Part two will focus on building a Linux based Ubuntu image and in subsequent parts, we will look at how we can utilize Azure to store and run our containers in the cloud using technologies such as Azure Container Registry (ACR), Azure Container Instances (ACI) and Azure Container Apps (ACA) to run and scale our self hosted GitHub runners, instead of using a VM based approach with docker running inside of a VM.

Setup environment

Before building and running docker images we need to set a few things up first. For my environment I will be using a Windows 11 virtual machine running WSL2. Here is more information on running docker on Windows Server. Things that we will need on our VM are:

Open PowerShell as administrator and run:

wsl --install
Enter fullscreen mode Exit fullscreen mode

After WSL is installed, run:

Enable-WindowsOptionalFeature -Online -FeatureName $("Microsoft-Hyper-V", "Containers") -All
Enter fullscreen mode Exit fullscreen mode

NOTE: You will need to reboot the system after adding the relevant features above.

  • Download and Install Docker Desktop For Windows (This will automatically also install Docker-Compose)

  • Once Docker Desktop For Windows is installed you need to switch to Windows containers. Use the Docker item in the Windows system tray:

image.png

Prepare PowerShell Scripts used in image creation

Now that we have Docker-Desktop as well as Docker-Compose installed and set to use Windows Containers we can start to build out our self hosted GitHub runner docker image.

Open VSCode, you can clone the repo found on my GitHub project docker-github-runner-windows which contains all the files or simply follow along with the following steps. We will prepare a few PowerShell scripts that will be needed as part of our docker image creation.

Create a root folder called docker-github-runner-windows and then another sub folder called scripts. Inside of the scripts folder you can create the following three PowerShell scripts:

image.png

Cleanup-Runners.ps1

Because we will run and scale self hosted runners using docker/docker-compose dynamically using our image, this script will be used to remove and unregister any old/offline GitHub runner registrations against our GitHub repository when we scale containers up and down based on our needs. This PowerShell script uses GitHub-CLI. If you are running this script locally ensure you have GitHub-CLI installed.

#This script invokes GitHub-CLI (Pre-installed on container image)
Param (
    [Parameter(Mandatory = $false)]
    [string]$owner = $env:GH_OWNER,
    [Parameter(Mandatory = $false)]
    [string]$repo = $env:GH_REPOSITORY,
    [Parameter(Mandatory = $false)]
    [string]$pat = $env:GH_TOKEN
)

#Use --with-token to pass in a PAT token on standard input. The minimum required scopes for the token are: "repo", "read:org".
#Alternatively, gh will use the authentication token found in environment variables. See gh help environment for more info.
#To use gh in GitHub Actions, add GH_TOKEN: $ to "env". on Docker run: Docker run -e GH_TOKEN='myPatToken'
gh auth login --with-token $pat

#Cleanup#
#Look for any old/stale dockerNode- registrations to clean up
#Windows containers cannot gracefully remove registration via powershell due to issue: https://github.com/moby/moby/issues/25982#
#For this reason we can use this scrip to cleanup old offline instances/registrations
$runnerBaseName = "dockerNode-"
$runnerListJson = gh api -H "Accept: application/vnd.github.v3+json" "/repos/$owner/$repo/actions/runners"
$runnerList = (ConvertFrom-Json -InputObject $runnerListJson).runners

Foreach ($runner in $runnerList) {
    try {
        If (($runner.name -like "$runnerBaseName*") -and ($runner.status -eq "offline")) {
            write-host "Unregsitering old stale runner: $($runner.name)"
            gh api --method DELETE -H "Accept: application/vnd.github.v3+json" "/repos/$owner/$repo/actions/runners/$($runner.id)"
        }
    }
    catch {
        Write-Error $_.Exception.Message
    }
}

#Remove PAT token after cleanup
$pat=$null
$env:GH_TOKEN=$null
Enter fullscreen mode Exit fullscreen mode

Install-Choco.ps1

This script will be used to install Chocolatey (Windows package manager) into our docker image when we build the image.

$securityProtocolSettingsOriginal = [System.Net.ServicePointManager]::SecurityProtocol
try {
    # Set TLS 1.2 (3072), then TLS 1.1 (768), then TLS 1.0 (192), finally SSL 3.0 (48)
    # Use integers because the enumeration values for TLS 1.2 and TLS 1.1 won’t
    # exist in .NET 4.0, even though they are addressable if .NET 4.5+ is
    # installed (.NET 4.5 is an in-place upgrade).
    [System.Net.ServicePointManager]::SecurityProtocol = 3072 -bor 768 -bor 192 -bor 48
}
catch {
    Write-Warning "Unable to set PowerShell to use TLS 1.2 and TLS 1.1 check .NET Framework installed. If you see underlying connection closed or trust errors, try the following: (1) upgrade to .NET Framework 4.5 (2) specify internal Chocolatey package location (set $env:chocolateyDownloadUrl prior to install or host the package internally), (3) use the Download + PowerShell method of install. See https://chocolatey.org/install for all install options."
}
Invoke-Expression ((New-Object System.Net.WebClient).DownloadString('https://community.chocolatey.org/install.ps1'))
[System.Net.ServicePointManager]::SecurityProtocol = $securityProtocolSettingsOriginal
Enter fullscreen mode Exit fullscreen mode

start.ps1

This script will be used as our 'ENTRYPOINT' script and will be used to bootstrap our docker container when we start/run a container from the image we will be creating. The main purpose of this script is to register a new self hosted GitHub runner instance on the repo we pass into the docker environment each time a new container is spun up or scaled up from the image.

#This script invokes GitHub-CLI (Already installed on container image)
#To use this entrypoint script run: Docker run -e GH_TOKEN='myPatToken' -e GH_OWNER='orgName' -e GH_REPOSITORY='repoName' -d imageName
Param (
    [Parameter(Mandatory = $false)]
    [string]$owner = $env:GH_OWNER,
    [Parameter(Mandatory = $false)]
    [string]$repo = $env:GH_REPOSITORY,
    [Parameter(Mandatory = $false)]
    [string]$pat = $env:GH_TOKEN
)

#Use --with-token to pass in a PAT token on standard input. The minimum required scopes for the token are: "repo", "read:org".
#Alternatively, gh will use the authentication token found in environment variables. See gh help environment for more info.
#To use gh in GitHub Actions, add GH_TOKEN: $ to "env". on Docker run: Docker run -e GH_TOKEN='myPatToken'
gh auth login

#Get Runner registration Token
$jsonObj = gh api --method POST -H "Accept: application/vnd.github.v3+json" "/repos/$owner/$repo/actions/runners/registration-token"
$regToken = (ConvertFrom-Json -InputObject $jsonObj).token
$runnerBaseName = "dockerNode-"
$runnerName = $runnerBaseName + (((New-Guid).Guid).replace("-", "")).substring(0, 5)

try {
    #Register new runner instance
    write-host "Registering GitHub Self Hosted Runner on: $owner/$repo"
    ./config.cmd --unattended --url "https://github.com/$owner/$repo" --token $regToken --name $runnerName

    #Remove PAT token after registering new instance
    $pat=$null
    $env:GH_TOKEN=$null

    #Start runner listener for jobs
    ./run.cmd
}
catch {
    Write-Error $_.Exception.Message
}
finally {
    # Trap signal with finally - cleanup (When docker container is stopped remove runner registration from GitHub)
    # Does not currently work due to issue: https://github.com/moby/moby/issues/25982#
    # Perform manual cleanup of stale runners using Cleanup-Runners.ps1
    ./config.cmd remove --unattended --token $regToken
}
Enter fullscreen mode Exit fullscreen mode

NOTE: This PowerShell script uses GitHub-CLI to register new agents onto the GitHub repository we specify. Thus we will load GitHub-CLI into our container when we build the image later on, so it is part of the container image.

Prepare dockerfile to build image (Windows)

Now with our scripts ready, we can get to the fun part... Building the windows docker image. Navigate back to the root folder and create a file called: dockerfile:

image.png

dockerfile

This dockerfile contains the instructions to build our container image.

##### BASE IMAGE INFO ######
#Using servercore insider edition for compacted size.
#For compatibility on "your" host running docker you may need to use a specific tag.
#E.g. the host OS version must match the container OS version.
#If you want to run a container based on a newer Windows build, make sure you have an equivalent host build.
#Otherwise, you can use Hyper-V isolation to run older containers on new host builds.
#The default entrypoint is for this image is Cmd.exe. To run the image:
#docker run mcr.microsoft.com/windows/servercore/insider:10.0.{build}.{revision}
#tag reference: https://mcr.microsoft.com/en-us/product/windows/servercore/insider/tags

#Win10
#FROM mcr.microsoft.com/windows/servercore/insider:10.0.19035.1

#Win11
FROM mcr.microsoft.com/windows/servercore/insider:10.0.20348.1

#input GitHub runner version argument
ARG RUNNER_VERSION

LABEL Author="Marcel L"
LABEL Email="pwd9000@hotmail.co.uk"
LABEL GitHub="https://github.com/Pwd9000-ML"
LABEL BaseImage="servercore/insider:10.0.20348.1"
LABEL RunnerVersion=${RUNNER_VERSION}

SHELL ["powershell", "-Command", "$ErrorActionPreference = 'Stop';"]

#Set working directory
WORKDIR /actions-runner

#Install chocolatey
ADD scripts/Install-Choco.ps1 .
RUN .\Install-Choco.ps1 -Wait; \
    Remove-Item .\Install-Choco.ps1 -Force

#Install Git, GitHub-CLI, Azure-CLI and PowerShell Core with Chocolatey (add more tooling if needed at build)
RUN choco install -y \
    git \
    gh \
    powershell-core \
    azure-cli

#Download GitHub Runner based on RUNNER_VERSION argument (Can use: Docker build --build-arg RUNNER_VERSION=x.y.z)
RUN Invoke-WebRequest -Uri "https://github.com/actions/runner/releases/download/v$env:RUNNER_VERSION/actions-runner-win-x64-$env:RUNNER_VERSION.zip" -OutFile "actions-runner.zip"; \
    Expand-Archive -Path ".\\actions-runner.zip" -DestinationPath '.'; \
    Remove-Item ".\\actions-runner.zip" -Force

#Add GitHub runner configuration startup script
ADD scripts/start.ps1 .
ADD scripts/Cleanup-Runners.ps1 .
ENTRYPOINT ["pwsh.exe", ".\\start.ps1"]
Enter fullscreen mode Exit fullscreen mode

Let's take a closer look and see what this docker build file will actually do, step by step:

#Win11
FROM mcr.microsoft.com/windows/servercore/insider:10.0.20348.1
Enter fullscreen mode Exit fullscreen mode

The 'FROM' instruction will tell our docker build to fetch and use a windows OS base image. Because windows base images can be fairly large we are using servercore insider edition, because the size is compact and optimized.

For compatibility on "your" host/VM running docker you may need to use a different tag, mcr.microsoft.com/windows/servercore/insider:10.0.{build}.{revision}

The host OS version must be higher than the base image OS version. You can use Hyper-V isolation to run older containers on new host builds also.

Because docker is running on my Windows 11 host build version: 10.0.22000.0. I'm using a container OS version of: mcr.microsoft.com/windows/servercore/insider:10.0.20348.1

Just make sure that your host build version running docker is higher than the base image build version you are using in the dockerfile when building the image. You can use the following Servercore insider tag reference: https://mcr.microsoft.com/en-us/product/windows/servercore/insider/tags

NOTE: To check your host OS build version you can run the following powershell command: [System.Environment]::OSVersion.Version

image.png

#input GitHub runner version argument
ARG RUNNER_VERSION

LABEL Author="Marcel L"
LABEL Email="pwd9000@hotmail.co.uk"
LABEL GitHub="https://github.com/Pwd9000-ML"
LABEL BaseImage="servercore/insider:10.0.20348.1"
LABEL RunnerVersion=${RUNNER_VERSION}
Enter fullscreen mode Exit fullscreen mode

Next we define an input argument using 'ARG'. This is so that we can instruct the docker build command to load a specific version of the GitHub runner agent into the image when building the image. Because we are using a windows container, 'ARG' will create a system variable $env:RUNNER_VERSION which will be accessible to PowerShell inside the container.

In addition we can also label our image with some metadata using 'LABEL' to add more information about the image. You can change these values as necessary.

NOTE: 'LABEL RunnerVersion=${RUNNER_VERSION}', this label is dynamically updated from the build argument we will be passing into the docker build command later.

SHELL ["powershell", "-Command", "$ErrorActionPreference = 'Stop';"]

#Set working directory
WORKDIR /actions-runner
Enter fullscreen mode Exit fullscreen mode

We then configure PowerShell as our default 'SHELL' for running scripts or commands and also set a working directory named actions-runner with 'WORKDIR'. This directory will contain our GitHub runner binaries and scripts, under the path: C:\actions-runner inside of the windows container.

#Install chocolatey
ADD scripts/Install-Choco.ps1 .
RUN .\Install-Choco.ps1 -Wait; \
    Remove-Item .\Install-Choco.ps1 -Force

#Install Git, GitHub-CLI, Azure-CLI and PowerShell Core with Chocolatey (add more tooling if needed at build)
RUN choco install -y \
    git \
    gh \
    powershell-core \
    azure-cli
Enter fullscreen mode Exit fullscreen mode

The 'ADD' instruction will copy our Install-Choco.ps1 script into the working directory and 'RUN' the script which will install Chocolatey into the image, and then cleanup/remove the script.

The second 'RUN' will then use Chocolatey to install Git, GitHub-CLI, Azure-CLI and PowerShell Core into the image. You can add any additional tooling you want to add to the image at build time here.

NOTE: Try not to install too many packages at build time to keep the image as lean, compact and re-usable as possible. You can always use a GitHub Action later in a workflow when running the container and use Chocolatey which is now loaded into the image/container to install more software.

I will also be showing how we can add more software and tooling e.g. Terraform later on when we run our container, using a GitHub Action.

#Download GitHub Runner based on RUNNER_VERSION argument (Can use: Docker build --build-arg RUNNER_VERSION=x.y.z)
RUN Invoke-WebRequest -Uri "https://github.com/actions/runner/releases/download/v$env:RUNNER_VERSION/actions-runner-win-x64-$env:RUNNER_VERSION.zip" -OutFile "actions-runner.zip"; \
    Expand-Archive -Path ".\\actions-runner.zip" -DestinationPath '.'; \
    Remove-Item ".\\actions-runner.zip" -Force
Enter fullscreen mode Exit fullscreen mode

The next 'RUN' instruction will run a series of PowerShell commands to download and extract a specific version of the GitHub runner binaries based on the build argument 'ARG' value passed into the container build process that sets the environment variable: $env:RUNNER_VERSION as described earlier.

#Add GitHub runner configuration startup script
ADD scripts/start.ps1 .
ADD scripts/Cleanup-Runners.ps1 .
ENTRYPOINT ["pwsh.exe", ".\\start.ps1"]
Enter fullscreen mode Exit fullscreen mode

The last section will 'ADD' the Cleanup-Runners.ps1 as well as an 'ENTRYPOINT' script named start.ps1 into the working directory. The entrypoint script will run each time a new container is created. It acts as a bootstrapper that will, based on specific environment variables we pass into the Docker Run command, such as, $env:GH_OWNER, $env:GH_REPOSITORY and $env:GH_TOKEN to register the containers self hosted runner agent against a specific repository in the GitHub organisation we specify.

Note that the 'ENTRYPOINT' script will be run using PowerShell Core with "pwsh.exe". Remember we used Chocolatey to install PowerShell Core as part of the image creation.

Now that we have our scripts as well as our dockerfile ready we can build our image.

NOTE: We can build and run the windows container images using docker-desktop or docker-compose, I will show both methods next.

Building the Docker Image - Docker Desktop (Windows)

In VSCode terminal or a PowerShell session, navigate to the root folder containing the docker file and run the following command. Remember we need to pass in a build argument to tell docker what version of the GitHub runner agent to use in the image creation. GitHub Runner Releases

#Build container: docker build [OPTIONS] PATH
docker build --build-arg RUNNER_VERSION=2.292.0 --tag docker-github-runner-win .
Enter fullscreen mode Exit fullscreen mode

The build process can take a little while to complete:

image.png

Once the process is complete, you will see the new image in Docker Desktop for Windows under images:

image.png

Run the Docker Image - Docker Desktop (Windows)

To run and provision a new self hosted GitHub runner windows container from the image we just created, run the following command. We have to pass in some environment variables using the '-e' option to specify the PAT (Personal Access Token), GitHub Organisation and Repository to register the runner against.

#Run container from image:
docker run -e GH_TOKEN='myPatToken' -e GH_OWNER='orgName' -e GH_REPOSITORY='repoName' -d image-name
Enter fullscreen mode Exit fullscreen mode

See creating a personal access token on how to create a GitHub PAT token. PAT tokens are only displayed once and are sensitive, so ensure they are kept safe.

The minimum permission scopes required on the PAT token to register a self hosted runner are: "repo", "read:org":

image.png

Tip: I recommend only using short lived PAT tokens and generating new tokens whenever new agent runner registrations are required.

After running this command, under the GitHub repository settings, you will see a new self hosted GitHub runner. (This is our docker container):

image.png

You will also be able to see the running container under Docker Desktop for Windows under Containers:

image.png

Lets test our new docker container self hosted GitHub runner by creating a GitHub workflow to run a few GitHub Actions by installing Terraform on the running container.

You can use this test workflow from my GitHub project: docker-github-runner-windows.

Create a new workflow under the GitHub repository where you deployed the self hosted runner where it's running:

name: Local runner test

on:
  workflow_dispatch:

jobs:
  testRunner:
    runs-on: [self-hosted]
    defaults:
      run:
        shell: pwsh
    steps:
      - uses: actions/checkout@v2
      - name: Setup Terraform
        run: choco install terraform -y
      - name: Refresh Environment
        run: refreshenv
      - name: Display Terraform Version
        run: terraform --version
      - name: Display Azure-CLI Version
        run: az --version
Enter fullscreen mode Exit fullscreen mode

Notice that the workflow 'runs-on: [self-hosted]' and that the default shell is set to PowerShell Core, 'shell: pwsh'.

Because Chocolatey is part of the image we built earlier, we can use the following step to install Terraform using Chocolatey:

steps:
- name: Setup Terraform
    run: choco install terraform -y
- name: Display Terraform Version
    run: terraform --version
Enter fullscreen mode Exit fullscreen mode

image.png

To add additional docker runners (containers), we simply re-run the docker command we ran earlier (Each run will create an additional runner instance/container):

#Run container from image:
docker run -e GH_TOKEN='myPatToken' -e GH_OWNER='orgName' -e GH_REPOSITORY='repoName' -d image-name
Enter fullscreen mode Exit fullscreen mode

image.png

Next we will look at stopping/destroying our running docker instances and cleaning up the registrations for all the self hosted runners registered against our GitHub repository.

To stop and remove all running containers simply run:

docker stop $(docker ps -aq) && docker rm $(docker ps -aq)
Enter fullscreen mode Exit fullscreen mode

You will notice that all the running containers under Docker Desktop for Windows are no longer there, but we still have the registrations against our GitHub repository which now shows as 'Offline':

image.png

To unregister or cleanup these stale registrations just run the script we created earlier under the ./scripts folder called Cleanup-Runners.ps1 (If you are running this script locally ensure you have GitHub-CLI installed as the script invokes GitHub-CLI to remove the stale 'Offline' runner registrations):

.\scripts\Cleanup-Runners.ps1 -owner "orgName" -repo "repoName" -pat "myPatToken"
Enter fullscreen mode Exit fullscreen mode

image.png

NOTE: for convenience, the same cleanup script is also copied to each container under the working directory 'C:\actions-runner\Cleanup-Runners.ps1'

After running the cleanup script you will notice that the stale 'offline' registrations against our repository are now removed.

Next we will look how we can build the image and also run our image at scale using docker-compose.

Building the Docker Image - Docker Compose (Windows)

As we saw earlier, it is pretty easy to build our image using docker commands, but we can also use docker-compose with a configuration file to make things a bit easier. So following on, navigate to the root folder again that contains the dockerfile we created earlier, and create a new 'YAML' file called docker-compose.yml:

---
version: '3.8'

services:
  runner:
    image: pwd9000-github-runner-win:latest
    build:
      context: .
      args:
        RUNNER_VERSION: '2.292.0'
    environment:
      GH_TOKEN: ${GH_TOKEN}
      GH_OWNER: ${GH_OWNER}
      GH_REPOSITORY: ${GH_REPOSITORY}
Enter fullscreen mode Exit fullscreen mode

In the docker compose configuration file we can set out the parameters for our docker image by specifying things like the image name, GitHub runner version, as well as our environment variables.

Note that we have to set these environment variables on our host, windows 11 machine in order for docker compose to be able to interpret the values specified on the 'YAML' file inside of the '${}' symbols. This can easily be done by running the following PowerShell commands on the windows 11 host:

#set system environment with $env: (or use .env file to pass GH_TOKEN, GH_OWNER, GH_REPOSITORY)
$env:GH_OWNER='Org/Owner'
$env:GH_REPOSITORY='Repository'
$env:GH_TOKEN='myPatToken'
Enter fullscreen mode Exit fullscreen mode

NOTE: You can also use an environment file instead to pass environment variables onto the docker compose build process using a docker-compose.yml file like this instead:

---
version: '3.8'

services:
  runner:
    image: pwd9000-github-runner-win:latest
    build:
      context: .
      args:
        RUNNER_VERSION: '2.292.0'
    env_file:
      - ./variables.env
Enter fullscreen mode Exit fullscreen mode

This method however requires us to create another file in the root of our working folder called ./variables.env and populating this file with our environment variables like so:

GH_OWNER=orgName
GH_REPOSITORY=repoName
GH_TOKEN=myPatToken
Enter fullscreen mode Exit fullscreen mode

IMPORTANT: Don't use this method, and don't commit this file to source control if you are using sensitive values and storing your code in a remote source control repository. Add this environment file to your '.gitignore' file if needed, so that it is not pushed into source control.

Which ever method you decide to use, you can kick off the build process after creating this docker-compose.yml file by running the following PowerShell command:

docker-compose build
Enter fullscreen mode Exit fullscreen mode

image.png

Once the process is complete, you will see the new image in Docker Desktop for Windows under images:

image.png

Run and scale the Docker Image - Docker Compose (Windows)

What's nice about using docker-compose is that we can easily scale the amount of runners we want to use simply by running the following command:

docker-compose up --scale runner=3 -d
Enter fullscreen mode Exit fullscreen mode

Because all of our configuration and details are kept in environment variables and the docker-compose 'YAML' file, we don't really have to run long docker commands as we did earlier, and we simply scale the amount of runners we want by specifying the '--scale' parameter.

image.png

NOTE: The '--scale runner=3 -d' parameter is based on the docker compose file, 'services:' setting, which in our case is called 'runner':

services:
  runner:
Enter fullscreen mode Exit fullscreen mode

To scale down to one runner, we can simply rerun the command as follow:

docker-compose up --scale runner=1 -d
Enter fullscreen mode Exit fullscreen mode

To stop and remove all running containers simply run:

docker-compose stop
docker rm $(docker ps -aq)
Enter fullscreen mode Exit fullscreen mode

As described earlier, you will notice that all the running containers under Docker Desktop for Windows are no longer there, but still have the registrations against our GitHub repository which now shows as 'Offline':

image.png

Simply re-run the cleanup script we ran earlier under the ./scripts folder called Cleanup-Runners.ps1:

.\scripts\Cleanup-Runners.ps1 -owner "orgName" -repo "repoName" -pat "myPatToken"
Enter fullscreen mode Exit fullscreen mode

In this part of the series we have covered how you can build and run self hosted Github runners as windows containers using docker-desktop and docker-compose. In the next part of this blog series we will look at building a Linux based Ubuntu image container instead, for our self hosted GitHub runners.

I hope you have enjoyed this post and have learned something new. You can find the code samples used in this blog post on my Github page. ❤️

Author

Like, share, follow me on: 🐙 GitHub | 🐧 Twitter | 👾 LinkedIn

Discussion (0)