terraform-aws-gitlab-runner-fleet¶

A lean OpenTofu module for a GitLab Runner fleeting fleet on AWS: one always-on manager plus scale-to-zero spot workers (the docker-autoscaler executor with fleeting-plugin-aws). Purpose-built for the phpboyscout ops account to replace cattle-ops/gitlab-runner, giving full control over the levers that module hid: spot allocation strategy (no forced spot_instance_pools), the workers' Docker install and disk size, an EFS and S3 cache layer, and no plan-time Lambda (so CI can plan and apply in separate jobs).

The design is spec 0015, A hand-rolled terraform-aws-gitlab-runner-fleet module, in the phpboyscout/infra wiki. That project is private, so it is named here rather than linked.

Usage¶

module "runner_fleet" {
  source  = "gitlab.com/phpboyscout/gitlab-runner-fleet/aws"
  version = "~> 0.2.0" # pre-1.0, a minor release may break: take patches only

  name_prefix                     = "pbs-ops-runner"
  vpc_id                          = var.vpc_id
  subnet_ids                      = var.subnet_ids
  runner_token_ssm_parameter_name = aws_ssm_parameter.runner_token.name
  ebs_kms_key_arn                 = data.aws_kms_key.ebs_default.arn

  max_instances = 4 # hard spot-instance ceiling (cost guardrail)

  tags = { Project = "phpboyscout", Environment = "ops" }
}

Every input and output, with its default, is in the generated tables in the README.

Further reading¶

The blog carries a curated route through this subject: Infrastructure with AWS and OpenTofu collects everything written about it, ordered so you can start at the beginning rather than newest-first.