docker-compose.yml

services:
  postgres:
    image: postgres:17-alpine
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-postgres
    command:
      - "postgres"
      - "-c"
      - "wal_level=logical"
      - "-c"
      - "max_replication_slots=10"
      - "-c"
      - "max_wal_senders=10"
    environment:
      POSTGRES_DB: ${POSTGRES_DB:-openmmes}
      POSTGRES_USER: ${POSTGRES_USER:-openmmes_user}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-openmmes_secret}
    volumes:
      - postgres_data:/var/lib/postgresql/data
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "pg_isready -U ${POSTGRES_USER:-openmmes_user} -d ${POSTGRES_DB:-openmmes}",
        ]
      interval: 5s
      timeout: 5s
      retries: 5
    # Every other service has this — without it postgres stays down after
    # a host reboot / docker restart while the app containers come back,
    # which surfaces as 502s from Caddy ("could not translate host name
    # 'postgres' to address").
    restart: unless-stopped
    networks:
      - openmmes-network

  backend:
    image: ghcr.io/mes-open/openmes:${OPENMES_VERSION:-latest}
    # Default to building from the cloned source so `docker compose up` always
    # runs the current code rather than a stale local image. Builds are
    # layer-cached → instant when nothing changed.
    #
    # Deployers who want the published image instead of a build set
    # OPENMES_PULL_POLICY=missing and OPENMES_VERSION=<tag>. Prefer that on hosts
    # where building is unreliable — NAS appliances on pre-6.6 kernels cannot
    # extract archives inside a build at all (the fchmodat2 syscall is missing and
    # older libseccomp reports the wrong error), so every image build fails there
    # while a pulled image runs fine.
    pull_policy: ${OPENMES_PULL_POLICY:-build}
    build:
      context: .
      dockerfile: backend/Dockerfile
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-backend
    environment:
      APP_NAME: ${APP_NAME:-OpenMES}
      APP_ENV: ${APP_ENV:-production}
      # Plant timezone: every rendered timestamp, report boundary and scheduled job
      # uses it (config/app.php → app.timezone, and the `timezone` Inertia prop).
      # Without this passthrough a Docker install is stuck on UTC.
      APP_TIMEZONE: ${APP_TIMEZONE:-UTC}
      APP_DEBUG: ${APP_DEBUG:-false}
      APP_URL: ${APP_URL:-http://localhost}
      # Sanctum SPA stateful hosts. Empty by default (production derives them
      # from APP_URL). Set this when serving on a host:port the app URL doesn't
      # cover — e.g. localhost:8084 or a LAN IP — or the live-sync /api
      # collection requests 401 and lists render empty.
      SANCTUM_STATEFUL_DOMAINS: ${SANCTUM_STATEFUL_DOMAINS:-}
      DB_CONNECTION: pgsql
      DB_HOST: postgres
      DB_PORT: 5432
      DB_DATABASE: ${POSTGRES_DB:-openmmes}
      DB_USERNAME: ${POSTGRES_USER:-openmmes_user}
      DB_PASSWORD: ${POSTGRES_PASSWORD:-openmmes_secret}
      # Database-backed queue so real jobs (webhook delivery, CSV import, MQTT,
      # auto-update) run async with retries instead of blocking the request. The
      # entrypoint runs a queue worker on the primary, so this works out of the
      # box. (Live-sync broadcasts are ShouldBroadcastNow and unaffected.)
      QUEUE_CONNECTION: ${QUEUE_CONNECTION:-database}
      # Anonymous usage reports. Unset means "ask the stored setting";
      # false forces reporting off without touching the database.
      OPENMES_TELEMETRY: ${OPENMES_TELEMETRY:-}
      ADMIN_USERNAME: ${ADMIN_USERNAME:-admin}
      ADMIN_EMAIL: ${ADMIN_EMAIL:-admin@openmmes.local}
      ADMIN_PASSWORD: ${ADMIN_PASSWORD:-Admin1234!}
      LOG_CHANNEL: ${LOG_CHANNEL:-stack}
      LOG_STACK: ${LOG_STACK:-single}
      LOG_LEVEL: ${LOG_LEVEL:-debug}
      # Server-side sessions — NOT the "cookie" driver (which spawns a cookie per
      # session id and bloats request headers). See .env.example.
      SESSION_DRIVER: ${SESSION_DRIVER:-database}
      # Reverb (WebSocket broadcasting) — live sync. REVERB_HOST points at the
      # reverb service for server-side publishing; clients connect via Caddy.
      BROADCAST_CONNECTION: ${BROADCAST_CONNECTION:-reverb}
      REVERB_APP_ID: ${REVERB_APP_ID:-openmes}
      REVERB_APP_KEY: ${REVERB_APP_KEY:-openmeskey}
      REVERB_APP_SECRET: ${REVERB_APP_SECRET:-openmessecret}
      REVERB_HOST: ${REVERB_HOST:-reverb}
      REVERB_PORT: ${REVERB_PORT:-8080}
      REVERB_SCHEME: ${REVERB_SCHEME:-http}
    expose:
      - "8000"
    depends_on:
      postgres:
        condition: service_healthy
    volumes:
      - backend_storage:/var/www/html/storage
      - backend_cache:/var/www/html/bootstrap/cache
    restart: unless-stopped
    networks:
      - openmmes-network

  # ── Reverb WebSocket server (live sync) ───────────────────────────────────
  # Single multiplexed WebSocket (Pusher protocol). Laravel publishes deltas to
  # it; clients connect through Caddy (/app/*). NOT exposed publicly.
  reverb:
    image: ghcr.io/mes-open/openmes:${OPENMES_VERSION:-latest}
    # Default to building from the cloned source so `docker compose up` always
    # runs the current code rather than a stale local image. Builds are
    # layer-cached → instant when nothing changed.
    #
    # Deployers who want the published image instead of a build set
    # OPENMES_PULL_POLICY=missing and OPENMES_VERSION=<tag>. Prefer that on hosts
    # where building is unreliable — NAS appliances on pre-6.6 kernels cannot
    # extract archives inside a build at all (the fchmodat2 syscall is missing and
    # older libseccomp reports the wrong error), so every image build fails there
    # while a pulled image runs fine.
    pull_policy: ${OPENMES_PULL_POLICY:-build}
    build:
      context: .
      dockerfile: backend/Dockerfile
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-reverb
    command: php artisan reverb:start --host=0.0.0.0 --port=8080
    environment:
      APP_NAME: ${APP_NAME:-OpenMES}
      APP_ENV: ${APP_ENV:-production}
      # Plant timezone: every rendered timestamp, report boundary and scheduled job
      # uses it (config/app.php → app.timezone, and the `timezone` Inertia prop).
      # Without this passthrough a Docker install is stuck on UTC.
      APP_TIMEZONE: ${APP_TIMEZONE:-UTC}
      APP_KEY: ${APP_KEY}
      DB_CONNECTION: pgsql
      DB_HOST: postgres
      DB_PORT: 5432
      DB_DATABASE: ${POSTGRES_DB:-openmmes}
      DB_USERNAME: ${POSTGRES_USER:-openmmes_user}
      DB_PASSWORD: ${POSTGRES_PASSWORD:-openmmes_secret}
      BROADCAST_CONNECTION: reverb
      REVERB_APP_ID: ${REVERB_APP_ID:-openmes}
      REVERB_APP_KEY: ${REVERB_APP_KEY:-openmeskey}
      REVERB_APP_SECRET: ${REVERB_APP_SECRET:-openmessecret}
      REVERB_SERVER_HOST: 0.0.0.0
      REVERB_SERVER_PORT: 8080
    expose:
      - "8080"
    depends_on:
      postgres:
        condition: service_healthy
    restart: unless-stopped
    networks:
      - openmmes-network

  # ── MQTT listener daemon ──────────────────────────────────────────────────
  # One supervisor process services ALL active MQTT machine_connections and
  # picks up newly-created / newly-activated ones automatically (no restart).
  # To pin it to a single connection instead, set MQTT_CONNECTION_ID and add
  # `--connection=${MQTT_CONNECTION_ID}` back to the command below.
  mqtt-listener:
    image: ghcr.io/mes-open/openmes:${OPENMES_VERSION:-latest}
    # Default to building from the cloned source so `docker compose up` always
    # runs the current code rather than a stale local image. Builds are
    # layer-cached → instant when nothing changed.
    #
    # Deployers who want the published image instead of a build set
    # OPENMES_PULL_POLICY=missing and OPENMES_VERSION=<tag>. Prefer that on hosts
    # where building is unreliable — NAS appliances on pre-6.6 kernels cannot
    # extract archives inside a build at all (the fchmodat2 syscall is missing and
    # older libseccomp reports the wrong error), so every image build fails there
    # while a pulled image runs fine.
    pull_policy: ${OPENMES_PULL_POLICY:-build}
    build:
      context: .
      dockerfile: backend/Dockerfile
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-mqtt-listener
    command:
      ["php", "artisan", "mqtt:listen"]
    environment:
      APP_ENV: ${APP_ENV:-production}
      # Plant timezone: every rendered timestamp, report boundary and scheduled job
      # uses it (config/app.php → app.timezone, and the `timezone` Inertia prop).
      # Without this passthrough a Docker install is stuck on UTC.
      APP_TIMEZONE: ${APP_TIMEZONE:-UTC}
      APP_KEY: ${APP_KEY}
      DB_CONNECTION: pgsql
      DB_HOST: postgres
      DB_PORT: 5432
      DB_DATABASE: ${POSTGRES_DB:-openmmes}
      DB_USERNAME: ${POSTGRES_USER:-openmmes_user}
      DB_PASSWORD: ${POSTGRES_PASSWORD:-openmmes_secret}
      BROADCAST_CONNECTION: log
    depends_on:
      postgres:
        condition: service_healthy
    volumes:
      - backend_storage:/var/www/html/storage
      - backend_cache:/var/www/html/bootstrap/cache
    restart: unless-stopped
    networks:
      - openmmes-network
    profiles:
      - connectivity

  # ── Modbus TCP poller daemon ──────────────────────────────────────────────
  # One instance per active Modbus machine_connection.
  # Example: MODBUS_CONNECTION_ID=1 docker compose --profile connectivity up -d modbus-poller
  modbus-poller:
    image: ghcr.io/mes-open/openmes:${OPENMES_VERSION:-latest}
    # Default to building from the cloned source so `docker compose up` always
    # runs the current code rather than a stale local image. Builds are
    # layer-cached → instant when nothing changed.
    #
    # Deployers who want the published image instead of a build set
    # OPENMES_PULL_POLICY=missing and OPENMES_VERSION=<tag>. Prefer that on hosts
    # where building is unreliable — NAS appliances on pre-6.6 kernels cannot
    # extract archives inside a build at all (the fchmodat2 syscall is missing and
    # older libseccomp reports the wrong error), so every image build fails there
    # while a pulled image runs fine.
    pull_policy: ${OPENMES_PULL_POLICY:-build}
    build:
      context: .
      dockerfile: backend/Dockerfile
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-modbus-poller
    command:
      ["php", "artisan", "modbus:poll", "--connection=${MODBUS_CONNECTION_ID:-1}"]
    environment:
      APP_ENV: ${APP_ENV:-production}
      # Plant timezone: every rendered timestamp, report boundary and scheduled job
      # uses it (config/app.php → app.timezone, and the `timezone` Inertia prop).
      # Without this passthrough a Docker install is stuck on UTC.
      APP_TIMEZONE: ${APP_TIMEZONE:-UTC}
      APP_KEY: ${APP_KEY}
      DB_CONNECTION: pgsql
      DB_HOST: postgres
      DB_PORT: 5432
      DB_DATABASE: ${POSTGRES_DB:-openmmes}
      DB_USERNAME: ${POSTGRES_USER:-openmmes_user}
      DB_PASSWORD: ${POSTGRES_PASSWORD:-openmmes_secret}
    depends_on:
      postgres:
        condition: service_healthy
    volumes:
      - backend_storage:/var/www/html/storage
      - backend_cache:/var/www/html/bootstrap/cache
    restart: unless-stopped
    networks:
      - openmmes-network
    profiles:
      - connectivity

  # ── OPC UA gateway sidecar (Node.js) ──────────────────────────────────────
  # Bridges an OPC UA server into the OpenMES ingest API. OpenMES cannot speak
  # OPC UA natively. Requires an API token (Admin → API Tokens).
  # Example: OPCUA_CONNECTION_ID=1 OPENMES_API_TOKEN=xxx \
  #          docker compose --profile connectivity up -d opcua-gateway
  opcua-gateway:
    build:
      context: ./opcua-gateway
      dockerfile: Dockerfile
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-opcua-gateway
    environment:
      OPENMES_API_URL: ${OPENMES_API_URL:-http://backend:8000}
      OPENMES_API_TOKEN: ${OPENMES_API_TOKEN}
      OPCUA_CONNECTION_ID: ${OPCUA_CONNECTION_ID:-1}
      OPCUA_PASSWORD: ${OPCUA_PASSWORD:-}
    depends_on:
      - backend
    restart: unless-stopped
    networks:
      - openmmes-network
    profiles:
      - connectivity

  # ── Queue worker ──────────────────────────────────────────────────────────
  # Consumes queued jobs (e.g. MQTT message processing). Start in production if
  # you use MQTT or other queued work.
  # Example: docker compose --profile workers up -d queue-worker
  queue-worker:
    image: ghcr.io/mes-open/openmes:${OPENMES_VERSION:-latest}
    # Default to building from the cloned source so `docker compose up` always
    # runs the current code rather than a stale local image. Builds are
    # layer-cached → instant when nothing changed.
    #
    # Deployers who want the published image instead of a build set
    # OPENMES_PULL_POLICY=missing and OPENMES_VERSION=<tag>. Prefer that on hosts
    # where building is unreliable — NAS appliances on pre-6.6 kernels cannot
    # extract archives inside a build at all (the fchmodat2 syscall is missing and
    # older libseccomp reports the wrong error), so every image build fails there
    # while a pulled image runs fine.
    pull_policy: ${OPENMES_PULL_POLICY:-build}
    build:
      context: .
      dockerfile: backend/Dockerfile
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-queue-worker
    command: ["php", "artisan", "queue:work", "--sleep=1", "--tries=5", "--timeout=120"]
    environment:
      APP_ENV: ${APP_ENV:-production}
      # Plant timezone: every rendered timestamp, report boundary and scheduled job
      # uses it (config/app.php → app.timezone, and the `timezone` Inertia prop).
      # Without this passthrough a Docker install is stuck on UTC.
      APP_TIMEZONE: ${APP_TIMEZONE:-UTC}
      APP_KEY: ${APP_KEY}
      QUEUE_CONNECTION: ${QUEUE_CONNECTION:-database}
      DB_CONNECTION: pgsql
      DB_HOST: postgres
      DB_PORT: 5432
      DB_DATABASE: ${POSTGRES_DB:-openmmes}
      DB_USERNAME: ${POSTGRES_USER:-openmmes_user}
      DB_PASSWORD: ${POSTGRES_PASSWORD:-openmmes_secret}
    depends_on:
      postgres:
        condition: service_healthy
    volumes:
      - backend_storage:/var/www/html/storage
      - backend_cache:/var/www/html/bootstrap/cache
    restart: unless-stopped
    networks:
      - openmmes-network
    profiles:
      - workers

  caddy:
    image: caddy:2-alpine
    container_name: ${OPENMES_NAME_PREFIX:-openmmes}-caddy
    ports:
      - "${HTTP_PORT:-80}:80"
      - "${HTTPS_PORT:-443}:443"
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile:ro
      - caddy_data:/data
      - caddy_config:/config
    environment:
      DOMAIN: ${DOMAIN:-localhost}
    depends_on:
      - backend
    restart: unless-stopped
    networks:
      - openmmes-network

volumes:
  postgres_data:
  backend_storage:
  backend_cache:
  caddy_data:
  caddy_config:

networks:
  openmmes-network:
    driver: bridge