docker-compose.yml

# docker-compose.yml (repo root)
# ----------------------------------------------------------------------
# Local DEV environment for arac-hasar-v2.
#
# What it brings up:
#   - backend (FastAPI, hot-reload, mounts ./services/backend:/app)
#   - worker  (Celery, same image)
#   - postgres 16
#   - redis 7
#   - minio + minio-init (S3-mock, bucket auto-created)
#
# What it does NOT bring up (run with pnpm dev):
#   - apps/web      -> cd apps/web      && pnpm dev   (port 3000)
#   - apps/mobile   -> cd apps/mobile   && pnpm start
#   - apps/desktop  -> cd apps/desktop  && pnpm tauri dev
#
# Quick start:
#   docker compose up -d postgres redis minio minio-init
#   docker compose up backend worker        # foreground, see logs
#
# For the full GPU/CUDA compose with bundled web container,
# see services/backend/docker-compose.yml.
# ----------------------------------------------------------------------

name: arac-hasar-v2-dev

services:
  backend:
    build:
      context: ./services/backend
      dockerfile: Dockerfile
    image: hasarui-api:dev
    container_name: hasarui-backend
    # NOTE: --reload-exclude tames watchfiles churn from ML shim copies,
    # __pycache__ writes, training log spew, and visualization PNGs that
    # otherwise trigger a uvicorn restart loop in dev.
    command: ["sh", "-c", "uvicorn main:app --host 0.0.0.0 --port 8000 --reload --reload-dir /app --reload-exclude '*.log' --reload-exclude '*.png' --reload-exclude '*.jpg' --reload-exclude '__pycache__/*' --reload-exclude 'tests/*' --reload-exclude 'migrations/versions/*'"]
    ports:
      - "8000:8000"
    env_file:
      - ./services/backend/.env
    environment:
      # Override URLs to point to compose service names
      DATABASE_URL: "postgresql://postgres:postgres@postgres:5432/arac_hasar"
      REDIS_URL: "redis://redis:6379/0"
      S3_ENDPOINT: "http://minio:9000"
      S3_ACCESS_KEY: "minioadmin"
      S3_SECRET_KEY: "minioadmin"
      S3_BUCKET: "inspections"
      S3_REGION: "us-east-1"
      ML_DEVICE: "cpu"
      # Entrypoint skips S3 model fetch — we mount weights as a volume.
      SKIP_MODEL_FETCH: "1"
      RUN_MIGRATIONS: "1"
      ENVIRONMENT: "development"
      # Web (3000), Tauri desktop dev (1420) + prod webview origins, Expo mobile (8081)
      CORS_ORIGINS: "http://localhost:3000,http://localhost:1420,http://tauri.localhost,tauri://localhost,http://localhost:8081"
      # Single-source ML modules: services/ml/ mounted read-only at /app/ml.
      # PYTHONPATH includes /app/ml so `from pipeline import DamagePipeline`
      # resolves to services/ml/pipeline.py without scripts/prepare-ml-shim.*
      # having to copy files into services/backend/. /app still takes
      # priority — backend-local override remains possible during a cutover.
      PYTHONPATH: "/app:/app/ml"
    volumes:
      # Hot-reload: source mounted on top of image
      - ./services/backend:/app
      # ML single-source mount: services/ml/ read-only at /app/ml.
      # Eliminates services/backend/pipeline.py drift vs services/ml/pipeline.py.
      # prepare-ml-shim.{sh,ps1} is now only needed for the PRODUCTION docker
      # build (CI bakes files in); dev runs straight off this mount.
      # NOT: pretrained/ alt klasoru runtime'da yazilabilir (model_manager
      # download bir kez .pt indirir, sonraki boot'larda cache hit). Kalan
      # services/ml/ icerigi kod ve egitim verisi — dev'de read-write OK.
      - ./services/ml:/app/ml
      # Local model weights for dev (point at your snapshot dir or symlink).
      # If this path is missing, run:
      #   $env:MODEL_SNAPSHOT_DIR = "<absolute path to latest snapshot>"
      # or edit the path below to the latest run under services/ml/runs/bundles/.
      - ${MODEL_SNAPSHOT_DIR:-./services/ml/runs/bundles/full_20260515_044630/_SNAPSHOT_FOR_BUILD}:/app/models:ro
    networks:
      - hasarui-net
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
      minio-init:
        condition: service_completed_successfully
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
    # Dev guardrail. Observed peak ~1.7 GiB (uvicorn 2 workers + YOLO ensemble
    # + cv2/numpy + torch CPU). Render starter (512 MB) is undersized — we
    # have to deploy on standard (2 GB) or trim to a single uvicorn worker
    # + lazy model load. NOT mirrored to render.yaml in this change.
    deploy:
      resources:
        limits:
          memory: 3G
          cpus: "4.0"
        reservations:
          memory: 1G
    restart: unless-stopped

  worker:
    build:
      context: ./services/backend
      dockerfile: Dockerfile
    image: hasarui-api:dev
    container_name: hasarui-worker
    command: ["sh", "-c", "celery -A worker.celery_app worker --loglevel=info --concurrency=1"]
    env_file:
      - ./services/backend/.env
    environment:
      DATABASE_URL: "postgresql://postgres:postgres@postgres:5432/arac_hasar"
      REDIS_URL: "redis://redis:6379/0"
      S3_ENDPOINT: "http://minio:9000"
      S3_ACCESS_KEY: "minioadmin"
      S3_SECRET_KEY: "minioadmin"
      S3_BUCKET: "inspections"
      S3_REGION: "us-east-1"
      ML_DEVICE: "cpu"
      SKIP_MODEL_FETCH: "1"
      RUN_MIGRATIONS: "0"
      ENVIRONMENT: "development"
      CORS_ORIGINS: "http://localhost:3000,http://localhost:1420,http://tauri.localhost,tauri://localhost,http://localhost:8081"
      # Celery prefork spawn temizler sys.path '' entry'sini; explicit PYTHONPATH
      # zorunlu — yoksa worker `from pipeline import DamagePipeline` başarısız.
      # /app/ml eklendi: services/ml/ tek-kaynak modülleri (pipeline.py vb.)
      PYTHONPATH: "/app:/app/ml"
    volumes:
      - ./services/backend:/app
      # Worker da ML modüllerini aynı tek-kaynak mount'tan okur.
      # NOT: pretrained/ alt klasoru runtime'da yazilabilir (model_manager
      # download bir kez .pt indirir, sonraki boot'larda cache hit). Kalan
      # services/ml/ icerigi kod ve egitim verisi — dev'de read-write OK.
      - ./services/ml:/app/ml
      - ${MODEL_SNAPSHOT_DIR:-./services/ml/runs/bundles/full_20260515_044630/_SNAPSHOT_FOR_BUILD}:/app/models:ro
    networks:
      - hasarui-net
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
      minio-init:
        condition: service_completed_successfully
      backend:
        condition: service_started
    # Override Dockerfile's HTTP /health check — worker has no HTTP server.
    # Use Celery's own broker ping (returns "pong" from each running worker).
    healthcheck:
      test: ["CMD-SHELL", "celery -A worker.celery_app inspect ping -d celery@$$HOSTNAME -t 5 || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
    # Worker ensemble inference 12 foto için peak ~2 GiB gözlemlendi.
    # Concurrency=1 ile zaten serialize ediyoruz; 3 GiB tampon yeterli.
    deploy:
      resources:
        limits:
          memory: 3G
          cpus: "4.0"
        reservations:
          memory: 1G
    restart: unless-stopped

  postgres:
    image: postgres:16-alpine
    container_name: hasarui-postgres
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: postgres
      POSTGRES_DB: arac_hasar
    ports:
      - "5432:5432"
    volumes:
      - pg_data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres -d arac_hasar"]
      interval: 5s
      timeout: 5s
      retries: 10
      start_period: 10s
    networks:
      - hasarui-net
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
    restart: unless-stopped

  redis:
    image: redis:7-alpine
    container_name: hasarui-redis
    command: ["redis-server", "--appendonly", "yes", "--save", "60", "1"]
    ports:
      - "6379:6379"
    volumes:
      - redis_data:/data
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 5s
      timeout: 3s
      retries: 5
    networks:
      - hasarui-net
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
    restart: unless-stopped

  minio:
    image: minio/minio:latest
    container_name: hasarui-minio
    command: ["server", "/data", "--console-address", ":9001"]
    environment:
      MINIO_ROOT_USER: minioadmin
      MINIO_ROOT_PASSWORD: minioadmin
    ports:
      - "9000:9000"   # S3 API
      - "9001:9001"   # Web console -> http://localhost:9001
    volumes:
      - minio_data:/data
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 10s
    networks:
      - hasarui-net
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"
    restart: unless-stopped

  minio-init:
    image: minio/mc:latest
    container_name: hasarui-minio-init
    depends_on:
      minio:
        condition: service_healthy
    networks:
      - hasarui-net
    # Idempotent: --ignore-existing makes mb / anonymous safe to re-run.
    entrypoint:
      - /bin/sh
      - -c
      - |
        set -e
        until mc alias set local http://minio:9000 minioadmin minioadmin >/dev/null 2>&1; do
          echo "minio-init: waiting for minio..."
          sleep 1
        done
        mc mb --ignore-existing local/inspections
        mc mb --ignore-existing local/models
        mc anonymous set download local/inspections || true
        echo "MinIO buckets ready: inspections, models"
    restart: "no"

networks:
  hasarui-net:
    driver: bridge

volumes:
  pg_data:
  redis_data:
  minio_data: