Home

Awesome

gcp-hosted-github-runner

GitHub Actions Workflow Status

This terraform module provides a ready to use solution for Google Cloud hosted GitHub ephemeral runner.

[!IMPORTANT] I am not responsible if this Terraform module results in high costs on your billing account. Keep an eye on your billing account and activate alerts!

Quickstart

1. Apply Terraform

Add this Terraform module to your root module and provide/adjust the values:

provider "google" {
  project = "<gcp_project>"
  region  = "<gcp_region>"
  zone    = "<gcp_zone>"
}

module "github-runner" {
  source                    = "github.com/Privatehive/gcp-hosted-github-runner"
  machine_type              = "c2d-highcpu-8" // The machine type of the VM instance.
  github_runner_group_id    = 1 // The GitHub Organization/Enterprise runner group ID. Has no effect for GitHub Repositories.

  // Provide only ONE of the following variables:
  github_enterprise         = "<enterprise_name>" // Provide the name of the GitHub Enterprise.
  github_organization       = "<organization_name>" // Provide the name of the GitHub Organization.
  github_repositories       = ["<repository_user/repository_name>"] // Provide USER/NAME of at least one GitHub Repository.
}

output "runner_webhook_config" {
  value = nonsensitive(module.github-runner.runner_webhook_config) // Remove the output after the initial setup.
}

Authenticate with gcloud and apply the terraform module (apply twice if the first apply results in an error - wait some minutes in between)

$ gcloud auth application-default login --project <gcp_project>
$ terraform init -upgrade && terraform apply

[!IMPORTANT] After a successful initial setup you should remove the runner_webhook_config output because it prints the webhook secret(s). Also make sure that the Terraform state file is stored in a safe place (e.g. in a Cloud Storage bucket). The state file contains the webhook secret as plaintext.

2. Configure GitHub webhook

Have a look at the Terraform output runner_webhook_config. There you find the Cloud Run webhook payload url(s) and the associated webhook secret(s). For each output line you have to create either an Enterprise, Organization or Repository webhook:

3. Provide PAT

This PAT is needed to automatically create a Enterprise, Organization, Repository jit-config for each ephemeral runner to join the Repository or the runner group of an Enterprise/Organization. Then open the Secret Manager in the Google Cloud Console and add a new Version to the already existing secret "github-pat-token". Paste the PAT into the Secret value field and click "ADD NEW VERSION".

[!TIP] Currently it is only possible to provide one PAT to the secret. That's why you can't combine an Enterprise with an Organization or Repository.

That's it 👍

As soon as you start a GitHub workflow, which contains a job with runs-on: self-hosted (or any other label you provided to the Terraform module variable github_runner_labels), a VM instance (with the specified machine_type) starts. The name of the VM instance starts with the github_runner_prefix, which is followed by a random string to make the name unique. The name of the VM instance is also the name of the runner in the GitHub runner group or repository. After the workflow job completed, the VM instance will be deleted again.

Advanced Configuration

Have a look at the variables.tf file how to further configure the Terraform module.

[!TIP] To find the cheapest VM machine_type use this table and sort by Spot instance cost. But remember that the price varies depending on the region.

Runner features

Magic Labels

Each workflow job can select a different machine type than the configured default machine_type. This is done by the special label @machine:c2d-standard-16.

jobs:
  example:
    runs-on: [self-hosted, @machine:c2d-standard-16] // this job will run on a c2d-standard-16 machine
    steps:
    - run: echo Hello world!

Expected Cost

The following Google Cloud resources are created that may generate cost:

Other:

Example:

A single 1 h long workflow job in europe-west1 leads to the following cost:

Ephemeral external IPv4 for Spot instance $0.0025
Spot VM Instance c2d-highcpu-8            $0.0494
Standard persistent disk 20 GiB used    ~ $0.0011
-----------------------------------------------------
                                          $0.053

Overall, only the compute instance accounts for the "majority" of the costs.

How it works

  1. As soon as a new GitHub workflow job is queued, the GitHub webhook event "Workflow jobs" invokes the Cloud Run container with path /webhook
  2. The Cloud run validates the caller source (signature) and if valid enqueues a "create-vm" Cloud task. This is necessary, because the timeout of a GitHub webhook is only 10 seconds but to start a VM instance takes about 1 minute.
  3. The Cloud task invokes the Cloud Run path /create_vm.
  4. Cloud Run creates a jit-config (using PAT from Secret Manager). The runner is then already registered (but marked as offline).
  5. The Cloud Run creates the VM instance from the instance template (preemtible spot VM instance by default) and provides it with the runner jit-config via custom metadata attribute.
  6. The runner starts working on the workflow job.
  7. As soon as the workflow job completed, the GitHub webhook event "Workflow jobs" invokes the Cloud Run again.
  8. The Cloud run validates the caller source (signature) and if valid enqueues a "delete-vm" Cloud task. This is necessary, because the timeout of a GitHub webhook is only 10 seconds but to delete a VM instance takes about 1 minute.
  9. The Cloud task invokes the Cloud Run path /delete_vm.
  10. The Cloud Run deletes the VM instance.

Troubleshooting

Public access to Cloud Run disallowed

The terraform error looks something like this:

Error applying IAM policy for cloudrun service "v1/projects/github-spot-runner/locations/us-east1/services/cloudrun-service": Error setting IAM policy for cloudrun service "v1/projects/github-spot-runner/locations/us-east1/services/cloudrun-service": googleapi: Error 400: One or more users named in the policy do not belong to a permitted customer, perhaps due to an Organization policy

  1. Solution: Use project tags: How to create public Cloud Run services when Domain Restricted Sharing is enforced

  2. Solution: Override the Organization Policy "Domain Restricted Sharing" in the project, by setting it to "Allow all".

The VM Instance immediately stops after it was created without processing a workflow job

The VM will shoutdown itself if the registration at the GitHub runner group fails. This can be caused by:

You can observer the runner registration process by connecting to the VM instance via SSH and running:

$ sudo journalctl -u google-startup-scripts.service --follow

New VM Instance not created (but a lot of instances are already running)

You exceeded your projects vCPU limit for the machine type in the region or for all regions. You may find an error log message in the Cloud Run logs stating Machine Type vCPU quota exceeded for region. Request a quota increase from google customer support for the project.

Nothing happens at all

Have a look in the Logs of the Cloud Run.