Deploy Apache Airflow 3.x on your AWS account in minutes. CDK-based, for development and testing.

Deploy in ~15 min. Tear down instantly.

Prerequisites: AWS CLI configured, Node.js 18+, SSM plugin

# One-time: bootstrap CDK
cd cdk && npx cdk bootstrap

# Deploy (default: apache/airflow main, LocalExecutor)
make deploy

# Or deploy a specific fork/branch
make deploy REPO=https://github.com/yourfork/airflow.git BRANCH=my-feature

# SSH in and run first-time setup (~10 min)
make ssh
# on EC2:  bash /opt/airflow-scripts/setup-airflow.sh

# Access Airflow UI
make tunnel
# open http://localhost:8080

No redeploy needed. Stops services, checks out branch, reinstalls, rebuilds UI, restarts.

# From laptop
make switch-branch BRANCH=another-feature

# Or from EC2
af switch another-feature

# Switch to a different fork
af switch my-feature https://github.com/otherfork/airflow.git

Switch between LocalExecutor and ECS multi-team executor at runtime. Regenerates airflow.cfg and restarts.

# Deploy with ECS stacks (needed once for ECS executor)
make deploy-ecs

# On EC2: switch between executors
af switch-executor ecs      # multi-team ECS Fargate workers
af switch-executor local    # back to LocalExecutor

From your laptop: make targets run commands remotely via SSM — no SSH needed. From the EC2 shell: make ssh to get a shell, then use the af CLI interactively.

af status                 Service health + port + DB check
af restart                Restart all services
af logs [service]         Tail logs (api-server, scheduler, dag-processor)
af switch <branch>        Switch Airflow branch
af switch-executor <type> Switch executor (local|ecs)
af deploy-dags            Upload test DAGs to S3
af rebuild                Build + push worker image to ECR
af db                     psql into metadata DB
af config                 Show airflow.cfg
af dags                   List DAGs
af teams                  List teams
af ecs-tasks              List running ECS tasks
af tunnel                 Show SSM tunnel command

Laptop ──SSM tunnel──> EC2 (api-server + scheduler + dag-processor)
                            |
                            |──> RDS PostgreSQL (metadata)
                            |──> S3 (DAG bundles + logs)
                            |──> ECS Fargate (worker tasks, optional)

Private subnets only. No public endpoints. No SSH keys. SSM Session Manager access.

All targets accept SUFFIX=name for multi-stack deployments and REPO=url BRANCH=name for fork/branch selection.

~$6/day while running (EC2 t3.large + RDS t3.small + NAT Gateway). ECS/Batch workers are pay-per-use on top.

make destroy removes everything. No lingering costs.

• Getting Started -- Step-by-step with timing and troubleshooting
• EC2 Setup & Scripts -- What each script does
• Known Issues -- Current limitations and workarounds
• Architecture -- VPC, security groups, IAM, cost breakdown
• Example Configs -- LocalExecutor and ECS multi-team airflow.cfg

Apache 2.0