Forgejo Stack

A production-ready docker-compose.yml with Forgejo + PostgreSQL + Redis + act_runner, plus a setup script and an example workflow to verify everything works.

The stack has four files. Here's the full picture.

What's included

Service	Image	Purpose
forgejo	codeberg.org/forgejo/forgejo:9	Git forge + web UI
db	postgres:16-alpine	Persistent storage
redis	redis:7-alpine	Cache, sessions, queue
dind	docker:dind	Isolated Docker for CI jobs
runner	code.forgejo.org/forgejo/act_runner:latest	Actions CI runner

First-time startup flow

# 1. Copy and edit the env file
cp .env.example .env
nano .env   # set POSTGRES_PASSWORD and FORGEJO_DOMAIN at minimum

# 2. Run setup script (fixes runner directory permissions)
bash setup.sh

# 3. Bring up everything except the runner first
docker compose up -d forgejo db redis dind

# 4. Open http://<your-ip>:3000 — complete the web installer
#    (DB settings are pre-filled via env vars)

# 5. Go to: Admin Area → Actions → Runners → Create new runner
#    Copy the token into .env as RUNNER_TOKEN

# 6. Start the runner
docker compose up -d runner

A few things worth noting

• Runner token: the runner can't register until Forgejo is fully initialized via the web installer, so it's a deliberate two-step. Once RUNNER_TOKEN is set, docker compose restart runner is all you need.
• DinD vs socket mount: the compose uses Docker-in-Docker rather than mounting /var/run/docker.sock, which gives better job isolation — important if you're running anything sensitive.
• SSH port: mapped to 2222 on the host to avoid conflicts with your existing SSH. Your git remote URLs will be ssh://git@<host>:2222/<user>/<repo>.git.
• Forgejo image tag 9: pins to the major version so you get patch updates but not surprise major bumps. Update intentionally by changing the tag.

The example workflow file goes into any repo at .forgejo/workflows/ci.yml and will immediately exercise the runner on push.