compose.yaml

# Development stack: Rails, background jobs, Tailwind watcher, and PostgreSQL.
# See docs/DOCKER.md.
#
#   docker compose up -d --build
#   docker compose logs -f web
#
# The database lives in the named volume pg_data, so it survives restarts,
# recreates, and rebuilds. Only `docker compose down -v` destroys it.

name: macronx

services:
  db:
    image: postgres:18
    environment:
      POSTGRES_USER: postgres
      # Override by setting POSTGRES_PASSWORD in .env
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-macronx}
    volumes:
      # Postgres 18 stores data in /var/lib/postgresql/<major>/docker, so the
      # volume is mounted at the parent directory. Mounting the old
      # /var/lib/postgresql/data path instead leaves the data in the container
      # layer, where it is lost on every recreate.
      - pg_data:/var/lib/postgresql
    # Port 5432 is not published: it would collide with a PostgreSQL server
    # already running on the host. Uncomment and use 5433 to reach the database
    # from the host.
    # ports:
    #   - "127.0.0.1:5433:5432"
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 5s
      timeout: 5s
      retries: 20
    restart: unless-stopped

  web:
    build:
      context: .
      dockerfile: Dockerfile.dev
    # bin/docker-entrypoint runs db:prepare for this exact command, which
    # creates and migrates the databases on first boot and is a no-op after.
    command: ["./bin/rails", "server"]
    environment:
      # Rails binds to localhost in development, which is unreachable from
      # outside the container.
      BINDING: "0.0.0.0"
      PORT: "3000"
      PGHOST: db
      PGUSER: postgres
      PGPASSWORD: ${POSTGRES_PASSWORD:-macronx}
      # Comma-separated hostnames to allow besides localhost and IP addresses,
      # for example the ngrok domain.
      # RAILS_DEVELOPMENT_HOSTS: your-domain.ngrok-free.app
    volumes:
      - .:/rails
    ports:
      - "3000:3000"
    # Lets the app reach Ollama, Whisper, and other services on the host.
    extra_hosts:
      - "host.docker.internal:host-gateway"
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:3000/up"]
      interval: 10s
      timeout: 5s
      retries: 30
      start_period: 60s
    depends_on:
      db:
        condition: service_healthy
    restart: unless-stopped
    stdin_open: true
    tty: true

  jobs:
    build:
      context: .
      dockerfile: Dockerfile.dev
    command: ["bin/jobs"]
    environment:
      PGHOST: db
      PGUSER: postgres
      PGPASSWORD: ${POSTGRES_PASSWORD:-macronx}
    volumes:
      - .:/rails
    extra_hosts:
      - "host.docker.internal:host-gateway"
    # Waits for the web container, which is the one that runs db:prepare.
    depends_on:
      web:
        condition: service_healthy
    restart: unless-stopped

  css:
    build:
      context: .
      dockerfile: Dockerfile.dev
    # The "always" argument puts Tailwind in polling mode. Without it the watcher
    # builds once and exits, leaving this container in a restart loop. Note the
    # rake argument syntax: the rails command runner rejects a bare "always".
    command: ["bin/rails", "tailwindcss:watch[always]"]
    volumes:
      - .:/rails
    restart: unless-stopped

volumes:
  pg_data: