I recently shared how I [set up Packer to build Proxmox templates](/building-proxmox-templates-packer/) in my homelab. That post covered storing (and retrieving) environment-specific values in Vault, the `cloud-init` configuration for definiting the installation parameters, the various post-install scripts for further customizing and hardening the template, and the Packer template files that tie it all together. By the end of the post, I was able to simply run `./build.sh ubuntu2204` to kick the build of a new Ubuntu 22.04 template without having to do any other interaction with the process.
That's pretty cool, but *The Dream* is to not have to do anything at all. So that's what this post is about: describing setting up a rootless self-hosted GitHub Actions Runner to perform the build, and the GitHub Actions workflows to trigger it.
When a GitHub Actions workflow fires, it schedules the job(s) to run on GitHub's own infrastructure. That's easy and convenient, but can make things tricky when you need a workflow to interact with on-prem infrastructure. I've worked around that in the past by [configuring the runner to connect to my tailnet](/gemini-capsule-gempost-github-actions/#publish-github-actions), but given the amount of data that will need to be transferred during the Packer build I decided that a [self-hosted runner](https://docs.github.com/en/actions/hosting-your-own-runners/managing-self-hosted-runners/about-self-hosted-runners) would be a better solution.
I wanted my runner to execute the build inside of a Docker container so that I could control that environment a bit more, and I also wanted to ensure that it would run [without elevated permissions](https://docs.docker.com/engine/security/rootless/). It took a bit of fiddling to get there, but I'm pretty pleased with the result!
GitHub [strongly recommends](https://docs.github.com/en/actions/hosting-your-own-runners/managing-self-hosted-runners/about-self-hosted-runners#self-hosted-runner-security) that you only use self-hosted runners with **private** repositories. You don't want a misconfigured workflow to allow a pull request submitted from a fork to run potentially-malicious code on your system(s).
So while I have a [public repo](https://github.com/jbowdre/packer-proxmox-templates/) to share my Packer work, my runner environment is attached to an otherwise-identical private repo. I'd recommend following a similar setup.
I started by cloning a fresh Ubuntu 22.04 VM off of my new template. After doing the basic initial setup (setting the hostname and IP, connecting it Tailscale), I then created a user account for the runner to use. That account will need sudo privileges during the initial setup, but then I can revoke that access. I also set a password for the account.
I then installed the `systemd-container` package so that I could use [`machinectl`](https://www.man7.org/linux/man-pages/man1/machinectl.1.html) to log in as the new user (since [`sudo su` won't work for the rootless setup](https://docs.docker.com/engine/security/rootless/#unable-to-install-with-systemd-when-systemd-is-present-on-the-system)).
```shell
sudo apt update # [tl! .cmd:2]
sudo apt install systemd-container
sudo machinectl shell github@
```
And I installed the `uidmap` package since rootless Docker requires `newuidmap` and `newgidmap`:
```shell
sudo apt install uidmap # [tl! .cmd]
```
At this point, I just followed the usual [Docker installation instructions](https://docs.docker.com/engine/install/ubuntu/#install-using-the-repository):
sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
sudo apt-get update # [tl! .cmd]
sudo apt-get install \ # [tl! .cmd]
docker-ce \
docker-ce-cli \
containerd.io \
docker-buildx-plugin \
docker-compose-plugin
```
Then the actual rootless setup can begin. That starts by disabling the existing Docker service and socket and then running the `dockerd-rootless-setuptool.sh` script:
After that, I started and enabled the service in the user context and enabled "linger" for the `github` user so that its systemd instance can continue to function even while the user is not logged in:
```shell
systemctl --user start docker # [tl! .cmd:2]
systemctl --user enable docker
sudo loginctl enable-linger $(whoami)
```
That should take care of setting up Docker, and I can quickly confirm by spawning the `hello-world` container:
```shell
docker run hello-world # [tl! .cmd]
Unable to find image 'hello-world:latest' locally # [tl! .nocopy:25]
So the Docker piece is sorted; now for setting up the runner.
#### Install/Configure Runner
I know I've been talking about a singular runner, but I actually set up multiple instances of the runner on the same host to allow running jobs in parallel. I could probably support four simultaneous builds in my homelab but I'll settle two runners for now (after all, I only have two build flavors so far anyway).
Each runner instance needs its own folder structure so I started by setting that up under `/opt/github/`:
- Extracted the runner software into the designated directory and `cd`'d to there:
```shell
tar xzf ./actions-runner-linux-x64-2.317.0.tar.gz --directory=runner1 # [tl! .cmd:1]
cd runner1
```
- Went to my private GitHub repo and navigated to **Settings > Actions > Runners** and clicked the big friendly **New self-hosted runner** button at the top-right of the page. All I really need from that is the token which appears in the **Configure** section. Once I had that token, I...
- Ran the configuration script, accepting the defaults for every prompt *except* for the runner name, which must be unique within the repository (so `runner1`, `runner2`, so on):
And I can see that my new runners are successfully connected to my *private* GitHub repo:
I now have a place to execute the Packer builds, I just need to tell the runner how to do that. And that's means it's time to talk about the...
### GitHub Actions Workflow
My solution for this consists of a Github Actions workflow which calls a custom action to spawn a Docker container and do the work. We'll cover this from the inside out to make sure we have a handle on all the pieces.
#### Docker Image
I opted to use a customized Docker image consisting of Packer and associated tools with the addition of the [wrapper script](/building-proxmox-templates-packer/#wrapper-script) that I used for local builds. That image will be integrated with a custom action called `packerbuild`.
So I commenced this part of the journey by creating a folder to hold my new action (and Dockerfile):
```shell
mkdir -p .github/actions/packerbuild # [tl! .cmd]
```
I don't want to maintain two copies of the `build.sh` script, so I moved it into this new folder and created a symlink to it back at the top of the repo:
It borrows from Hashicorp's minimal `alpine` image and installs a few common packages and `xorriso` to support the creation of ISO images. It then downloads the indicated version of the Packer installer and extracts it to `/bin/`. Finally it copies the `build.sh` script into the image and sets it as the `ENTRYPOINT`.
#### Custom Action
Turning this Docker image into an action only needs a very minimal amount of YAML to describe how to interact with the image.
So here is `.github/actions/packerbuild/action.yml`:
```yaml
# torchlight! {"lineNumbers":true}
name: 'Execute Packer Build'
description: 'Performs a Packer build'
inputs:
build-flavor:
description: 'The build to execute'
required: true
runs:
using: 'docker'
image: 'Dockerfile'
args:
- ${{ inputs.build-flavor }}
```
As you can see, the action expects (nay, requires!) a `build-flavor` input to line up with `build.sh`'s expected parameter. The action will run in Docker using the image defined in the local `Dockerfile`, and will pass `${{ inputs.build-flavor }}` as the sole argument to that image.
And that brings us to the workflow which will tie all of this together.
#### The Workflow
The workflow is defined as another bit of YAML in `.github/workflows/build.yml`. It starts simply enough with a name and a declaration of when the workflow should be executed.
```yaml
# torchlight! {"lineNumbers":true}
name: Build VM Templates
on:
workflow_dispatch:
schedule:
- cron: '0 8 * * 1'
```
`workflow_dispatch` just indicates that I should be able to manually execute the workflow from the GitHub Actions UI, and the `cron` schedule means that the workflow will run every Monday at 8:00 AM (UTC).
Rather than rely on an environment file (which, again, should *not* be committed to version control!), I'm using [repository secrets](https://docs.github.com/en/actions/security-guides/using-secrets-in-github-actions) to securely store the `VAULT_ADDR` and `VAULT_TOKEN` values. So I introduce those into the workflow like so:
When I did the [Vault setup](/building-proxmox-templates-packer/#vault-configuration), I created the token with a `period` of `336` hours; that means that the token will only remain valid as long as it gets renewed at least once every two weeks. So I start the `jobs:` block with a simple call to Vault's REST API to renew the token before each run:
--request POST "${VAULT_ADDR}v1/auth/token/renew-self" | grep -q auth
```
Assuming that token is renewed successfully, the Build job uses a [matrix strategy](https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idstrategymatrixinclude) to enumerate the `build-flavor`s that will need to be built. All of the following steps will be repeated for each flavor.
And the first step is to simply check out the GitHub repo so that the runner has all the latest code.
To get the runner to interact with the rootless Docker setup we'll need to export the `DOCKER_HOST` variable and point it to the Docker socket registered by the user... which first means obtaining the UID of that user and echoing it to the special `$GITHUB_OUTPUT` variable so it can be passed to the next step:
And now, finally, for the actual build. The `Build template` step calls the `.github/actions/packerbuild` custom action, sets the `DOCKER_HOST` value to the location of `docker.sock` (using the UID obtained earlier) so the runner will know how to interact with rootless Docker, and passes along the `build-flavor` from the matrix to influence which template will be created.
If it fails for some reason, the `Retry on failure` step will try again, just in case it was a transient glitch like a network error or a hung process.
All that's left at this point is to `git commit` and `git push` this to my *private* repo. I can then visit the repo on the web, head to the **Actions** tab, select the new **Build VM Templates** workflow on the left, and click the **Run workflow** button. That fires off the build, and I can check back a few minutes later to confirm that it completed successfully:
And I can also check my Proxmox host to confirm that the new VM templates were indeed created:
For future builds, I don't have to actually do anything at all. GitHub will automatically trigger this workflow every Monday morning so my templates will never be more than a week out-of-date. Pretty slick, right?
You can check out my *public* repo at [github.com/jbowdre/packer-proxmox-templates/](https://github.com/jbowdre/packer-proxmox-templates/) to explore the full setup - and to follow along as I add support for additional OS flavors.
