docker-compose.prod.yml

# ─────────────────────────────────────────────────────────────────────────────
# Production compose file
#
# Service → Image mapping:
#   backend     → Dockerfile.web   (slim Python, no conda)  ~seconds to rebuild
#   celery-beat → Dockerfile.web   (slim Python, no conda)  reuses backend image
#   celery      → Dockerfile       (worker runtime; reuses prebuilt webkinpred-envs image)
#   frontend    → frontend/Dockerfile.prod
#
# Use deploy.sh for all deploys — it handles BuildKit flags and prunes dangling
# images automatically so disk space does not accumulate:
#
#   ./deploy.sh                  # full prod deploy
#   ./deploy.sh prod celery      # rebuild only the celery worker
# ─────────────────────────────────────────────────────────────────────────────

# ── Reusable fragments ────────────────────────────────────────────────────────
x-common-env: &common-env
  DEBUG: "0"
  DJANGO_SETTINGS_MODULE: webKinPred.settings_docker
  DJANGO_SECRET_KEY: ${DJANGO_SECRET_KEY:?DJANGO_SECRET_KEY must be set}
  REDIS_HOST: redis
  REDIS_PORT: "6379"
  GPU_EMBED_SERVICE_URL: ${GPU_EMBED_SERVICE_URL:-}
  GPU_EMBED_SERVICE_TOKEN: ${GPU_EMBED_SERVICE_TOKEN:-}
  GPU_EMBED_HEALTH_TTL_SECONDS: ${GPU_EMBED_HEALTH_TTL_SECONDS:-10}
  GPU_EMBED_JOB_TIMEOUT_SECONDS: ${GPU_EMBED_JOB_TIMEOUT_SECONDS:-1200}
  GPU_EMBED_FAIL_CLOSED: ${GPU_EMBED_FAIL_CLOSED:-0}

x-web-build: &web-build
  build:
    context: .
    dockerfile: Dockerfile.web
  image: webkinpred-web:latest

services:
  # ── Redis ───────────────────────────────────────────────────────────────────
  redis:
    image: redis:7-alpine
    volumes:
      - redis_data:/data
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 10s
      timeout: 3s
      retries: 3

  # ── Backend (Django / Gunicorn) ─────────────────────────────────────────────
  # Lightweight web image — handles HTTP only, queues tasks to celery.
  # No ML inference here, no conda envs needed.
  backend:
    <<: *web-build
    ports:
      - "8000:8000"
    volumes:
      - ./media:/app/media
      - ./db.sqlite3:/app/db.sqlite3
      - ./staticfiles:/app/staticfiles
      - ./fastas/dbs:/app/fastas/dbs:ro
    environment:
      <<: *common-env
    depends_on:
      redis:
        condition: service_healthy
    restart: unless-stopped
    command: >-
      sh -c "python manage.py migrate &&
             python manage.py migrate --database=seqmap &&
             python manage.py collectstatic --noinput --clear &&
             gunicorn webKinPred.wsgi:application --bind 0.0.0.0:8000 --workers 2 --worker-class sync --timeout 300 --graceful-timeout 30 --access-logfile - --error-logfile - --capture-output"

  # ── Frontend (nginx + React SPA) ────────────────────────────────────────────
  frontend:
    build:
      context: ./frontend
      dockerfile: Dockerfile.prod
    ports:
      - "3000:80"
    depends_on:
      - backend
    restart: unless-stopped

  # ── Celery worker ───────────────────────────────────────────────────────────
  # Runtime worker image layered on top of prebuilt conda env image.
  # --max-tasks-per-child: worker subprocess is replaced after N tasks,
  #   freeing all ML model memory that accumulated during inference.
  celery:
    build:
      context: .
      dockerfile: Dockerfile
      args:
        WEBKINPRED_ENVS_IMAGE: ${WEBKINPRED_ENVS_IMAGE:-webkinpred-envs:latest}
    image: webkinpred-worker:latest
    volumes:
      - ./media:/app/media
      - ./db.sqlite3:/app/db.sqlite3
      - ./media/sequence_info/seqmap.sqlite3:/app/media/sequence_info/seqmap.sqlite3
      - ./staticfiles:/app/staticfiles
      - ./fastas/dbs:/app/fastas/dbs:ro
      - ./models/EITLEM/Weights:/app/models/EITLEM/Weights:ro
      - ./models/TurNup/data/saved_models:/app/models/TurNup/data/saved_models:ro
      - ./models/UniKP-main/models:/app/models/UniKP-main/models:ro
      - ./models/DLKcat/DeeplearningApproach:/app/models/DLKcat/DeeplearningApproach:ro
      - ./models/KinForm/results/trained_models:/app/models/KinForm/results/trained_models:ro
      - ./models/CataPro:/app/models/CataPro:ro
      - ./models/CatPred:/app/models/CatPred:ro
    environment:
      <<: *common-env
    depends_on:
      redis:
        condition: service_healthy
      backend:
        condition: service_started
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "celery", "-A", "webKinPred", "inspect", "ping"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    command: >
      celery -A webKinPred worker
        --loglevel=info
        --queues=webkinpred
        --concurrency=1
        --max-tasks-per-child=10

  # ── Celery beat scheduler ───────────────────────────────────────────────────
  # Reuses the lightweight web image — only schedules tasks,
  # no model inference, no conda envs needed.
  celery-beat:
    <<: *web-build
    volumes:
      - ./media:/app/media
      - ./db.sqlite3:/app/db.sqlite3
      - ./media/sequence_info/seqmap.sqlite3:/app/media/sequence_info/seqmap.sqlite3
    environment:
      <<: *common-env
    depends_on:
      redis:
        condition: service_healthy
      backend:
        condition: service_started
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "pgrep", "-f", "celery.*beat"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    command: celery -A webKinPred beat --loglevel=info

volumes:
  redis_data: