Field manual / system map
Operate the runner control plane as inspected infrastructure.
Gondola installs into an AWS account you control. A small controller maps queued GitHub jobs to ephemeral EC2 runners selected from your Terraform-defined fleet catalog.
Runtime
ECS Fargate controller and EC2 runners in your account
Workflow interface
Standard GitHub Actions runner labels
Supported fleets
Linux x64 and ARM64, On-Demand or Spot
Network path
Outbound HTTPS; no Gondola-hosted job proxy
Installation path
System boundary
GitHub supplies the queue and a short-lived just-in-time runner configuration. Gondola’s controllers long-poll GitHub, coordinate one active listener through a small DynamoDB lease, and launch a matching EC2 instance. The runner executes one job and terminates.
Queued job
Gondola controller
One-job EC2 runner
Current support
| Area | Supported |
|---|---|
| Source control | GitHub.com organizations and user-owned repositories |
| Cloud | One AWS account and region per controller deployment |
| Compute | Linux x64 and ARM64 EC2 runner fleets |
| Capacity | On-Demand, Spot, or Spot with On-Demand fallback |
| Networking | Existing VPCs and outbound HTTPS; no inbound listener |
| Not supported | Azure, Google Cloud, GitHub Enterprise Server, Windows, and macOS |
Responsibility split
| Owner | Responsible for |
|---|---|
| Gondola | Controller behavior, Terraform module, release artifacts, documentation, and published support |
| Customer platform team | AWS account, VPC and egress, IAM, images, Terraform state, fleet policy, budgets, and incident response |
| Repository owner | Workflow code, triggers, third-party Actions, branch protection, secrets, and runner-label selection |
| AWS and GitHub | Their respective cloud and source-control services, availability, billing, and platform terms |
| Stripe | Hosted payment collection and customer billing portal; never the runner execution path |
How to use this manual
- Use Install on AWS to collect prerequisites and produce a reviewed Terraform plan.
- Use Fleet records to turn workload classes into stable, least-privilege routing policy.
- Read the architecture before setting readiness, failover, or network expectations.
- Adopt the operations and troubleshooting procedures before admitting important work.
- Model AWS costs and confirm the subscription scope separately.
The current build is a release candidate. Begin in a non-critical repository, verify every fleet and failure path, and promote the exact tested image digest—not a mutable tag—under your own change control.