Provision an autoscaled Gitlab runner on AWS

This simple Terraform code provision one Gitlab runner on AWS with a template configuration implementing the autoscaling feature using EC2 and the docker-machine executor.

Provision

In order to provision a new runner, you need to do the following:

• install terraform (tested with version 1.4.4)
• set all the variables to be replaced in the terraform.tfvars file. They include the AWS access and secret keys, the AWS profile (set to default if you did not set profiles) and the name and token of the runner to be created
• execute the usual Terraform sequence of commands to provision the runner:

terraform init
terraform plan # check what you are doing
terraform apply

After that, you will be able to connect to the runner manager instance using the appropriate key pair chosen.

The sharp bits

• Manual runner registration: a new runner must be manually registered by SSHing into the corresponding manager and execute:

sudo mv /etc/gitlab-runner/config.toml.tmpl /etc/gitlab-runner/config.toml
gitlab-runner start

• Do not mess with the remote state: the file backend.tf is auto-generated and points to a previously generate environment in the eu-west-3 region. This environment notably already contains:

  • the VPC which can be reused for hosting new runners.
  • the S3 bucket used a shared cache for the runners

Do not remove this file and always make sure that you are using a cloud-based backend to store your state, i.e. you should never have a terraform.tfstate file in your project folder, otherwise you risk to corrupt the remotely saved state.