This repository contains the code of GitHub Actions Runner modified to spawn preemptible GCP instances with Singularity containers and to perform run steps within them.
The software was designed to run in Google Compute Engine. Therefore, it is necessary to prepare some virtual infrastructure prior to installing the runner.
The repositories listed below contain the definitions of the required components:
- github-actions-runner-scalerunner - the image used by preemptible GCP instances that serve as workers (one worker per job).
- github-actions-runner-terraform - a Terraform module used to create the virtual network, firewall rules, cloud NAT and coordinator instance for the runner.
For convenience, an installation script is available that installs dependencies, configures the system, clones the repository and builds the runner.
The manual below assumes that Debian Buster is used to deploy the runner.
The following packages must be installed:
build-essential
- Terraform
- Google Cloud SDK
With all prerequisites in place, in order to install the software, follow the steps below:
Install the Google Cloud SDK and setup the project:
# Authenticate with GCP.
gcloud auth login
# Create a GCP project for your runner.
export PROJECT=example-runner-project
gcloud projects create $PROJECT
gcloud config set $PROJECT
# Create and setup a service account.
export SERVICE_ACCOUNT_ID=runner-manager
gcloud iam service-accounts create $SERVICE_ACCOUNT_ID
gcloud projects add-iam-policy-binding $PROJECT \
--member="serviceAccount:$SERVICE_ACCOUNT_ID@$PROJECT \
--role="roles/compute.admin"
gcloud projects add-iam-policy-binding $PROJECT \
--member="serviceAccount:$SERVICE_ACCOUNT_ID@$PROJECT \
--role="roles/iam.serviceAccountCreator"
gcloud projects add-iam-policy-binding $PROJECT \
--member="serviceAccount:$SERVICE_ACCOUNT_ID@$PROJECT \
--role="roles/iam.serviceAccountUser"
# Create and download SA key.
# WARNING: the export below will be used by Terraform later.
export GOOGLE_APPLICATION_CREDENTIALS
gcloud iam service-accounts keys create $GOOGLE_APPLICATION_CREDENTIALS \
--iam-account=$SERVICE_ACCOUNT_ID@$PROJECT
# Create a GCP bucket for worker image.
export BUCKET=$PROJECT-worker-bucket
gsutil mb gs://$BUCKET
Build and upload the worker image:
# Clone the repository
git clone https://github.com/antmicro/github-actions-runner-scalerunner.git
cd github-actions-runner-scalerunner
# Compile bzImage
cd buildroot && make BR2_EXTERNAL=../overlay/ scalenode_gcp_defconfig && make
# Prepare a disk for GCP
./make_gcp_image.sh
# Upload the resulting tar archive
./upload_gcp_image.sh $PROJECT $BUCKET
Setup virtual infrastructure using Terraform:
git clone https://github.com/antmicro/github-actions-runner-terraform.git
terraform init && terraform apply
Connect to the coordinator instance created in the previous step:
gcloud compute --zone <COORDINATOR_ZONE> ssh <COORDINATOR_INSTANCE>
Install and configure the runner on the coordinator instance:
# Download and run the installation script.
wget https://raw.githubusercontent.com/antmicro/runner/vm-runners/scripts/install.sh | bash
# The runner software runs as the 'runner' user, so let's sudo into it.
sudo -i -u runner
cd /home/runner/github-actions-runner
# Copy the .vm_specs.json file and adjust the parameters accordingly.
cp .vm_specs.example.json .vm_specs.json
vim .vm_specs.json
# Register the runner in the desired repository.
./config.sh --url https://github.com/$REPOSITORY_ORG/$REPOSITORY_NAME --token $TOKEN --num $SLOTS
In order to start the runners manually, run SCALE=<number of slots> supervisord -n -c supervisord.conf
.
Start the runner by running sudo systemctl start gha-main@$SLOTS
replacing $SLOTS
with the number of runner slots you'd like to allocate.
If you want the software to start automatically, run the command above with the enable
action instead of start
.