This Terraform project deploys the load-testing infrastructure for Codabench across three AWS regions. It creates the machines that generate the load - the Codabench target infrastructure (the thing we’re testing) is deployed separately and already exists.
What gets deployed
Paris (eu-west-1) - same VPC as Codabench
Lives inside the existing Codabench VPC so it can reach the ALB internally and monitor RabbitMQ directly.
| Resource | Type | Role |
|---|---|---|
| Locust master | 1 × t3.medium | Orchestrates workers, serves the web UI on :8089 |
| Locust workers | ASG 2–6 × t3.medium | Generate HTTP load against the ALB |
| Playwright | 1 × t3.large | Measures perceived frontend latency with headless Chromium |
New private subnet: 10.0.21.0/24 - internet access goes through the existing NAT Gateway.
US East (us-east-1) - separate VPC
| Resource | Type | Role |
|---|---|---|
| Locust workers | ASG 2–4 × t3.medium | Headless load from the US |
New VPC 10.1.0.0/16 with a public subnet. Traffic goes over the public internet to the ALB in Paris - no VPC peering, because we want to measure real user latency from North America.
Asia Pacific (ap-southeast-1) - separate VPC
| Resource | Type | Role |
|---|---|---|
| Locust workers | ASG 2–4 × t3.medium | Headless load from Singapore |
Same design as US East, VPC 10.2.0.0/16. Measures real latency from Asia.
Shared resources
| Resource | Region | Role |
|---|---|---|
NLB codabench-loadtest-master-nlb | eu-west-1 | Exposes master :5557-5558 to US/Asia workers (TCP passthrough) |
S3 bucket codabench-loadtest-results | eu-west-1 | Collects CSV results |
IAM role codabench-loadtest-role | global | SSM access + S3 write for results |
| 3 instance profiles | one per region | Attach the IAM role to EC2 in each region |
Architecture diagram
Prerequisites
- AWS CLI configured with a
codabenchprofile - Terraform >= 1.0
- The Codabench infrastructure must already be deployed (VPC, ALB, app)
Usage
# 1. Initialize Terraform (downloads providers, configures S3 backend)
cd cloud/aws/deploy
terraform init
# 2. Preview what will be created
terraform plan
# 3. Deploy
terraform apply
# 5. Connect to any instance via SSM
aws ssm start-session \
--target <instance-id> \
--profile codabench
Tear down
terraform destroy
This destroys only the load-generator infrastructure. The Codabench target infrastructure is untouched (separate Terraform state).