services:
  db:
    image: postgres:15
    container_name: rss_db
    environment:
      - POSTGRES_DB=${DB_NAME}
      - POSTGRES_USER=${DB_USER}
      - POSTGRES_PASSWORD=${DB_PASS}
    volumes:
      - postgres_data:/var/lib/postgresql/data
      - ./init-db:/docker-entrypoint-initdb.d
    restart: always
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${DB_USER} -d ${DB_NAME}"]
      interval: 5s
      timeout: 5s
      retries: 5

  web:
    build:
      context: .
      args:
        TORCH_CUDA: cu121
    container_name: rss_web
    command: gunicorn --bind 0.0.0.0:8000 --workers 3 app:app
    ports:
      - "8001:8000"
    environment:
      - DB_HOST=db
      - DB_PORT=5432
      - DB_NAME=${DB_NAME}
      - DB_USER=${DB_USER}
      - DB_PASS=${DB_PASS}
      - SECRET_KEY=${SECRET_KEY}
      - WEB_TRANSLATED_DEFAULT=1
      - DEFAULT_LANG=es
      - TRANSLATION_PREFERRED_LANGS=es
    depends_on:
      db:
        condition: service_healthy
    restart: always

  scheduler:
    build:
      context: .
      args:
        TORCH_CUDA: cu121
    container_name: rss_scheduler
    command: python scheduler.py
    environment:
      - DB_HOST=db
      - DB_PORT=5432
      - DB_NAME=${DB_NAME}
      - DB_USER=${DB_USER}
      - DB_PASS=${DB_PASS}
      - SECRET_KEY=${SECRET_KEY}
    depends_on:
      db:
        condition: service_healthy
    restart: always

  translator:
    build:
      context: .
      args:
        TORCH_CUDA: cu121
    container_name: rss_translator
    command: python translation_worker.py
    environment:
      - DB_HOST=db
      - DB_PORT=5432
      - DB_NAME=${DB_NAME}
      - DB_USER=${DB_USER}
      - DB_PASS=${DB_PASS}

      - TARGET_LANGS=es
      - TRANSLATOR_BATCH=8
      - ENQUEUE=200
      - TRANSLATOR_SLEEP_IDLE=5

      - MAX_SRC_TOKENS=680
      - MAX_NEW_TOKENS=400

      - NUM_BEAMS_TITLE=2
      - NUM_BEAMS_BODY=1

      - UNIVERSAL_MODEL=facebook/nllb-200-1.3B

      - CHUNK_BY_SENTENCES=True
      - CHUNK_MAX_TOKENS=700
      - CHUNK_OVERLAP_SENTS=1
      - CLEAN_ARTICLE=1

      - DEVICE=cuda

      - PYTHONUNBUFFERED=1
      - HF_HOME=/root/.cache/huggingface
      - TOKENIZERS_PARALLELISM=false
      - PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:64,garbage_collection_threshold:0.9

      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    volumes:
      - hf_cache:/root/.cache/huggingface
    depends_on:
      db:
        condition: service_healthy
    restart: always
    gpus: all

  ner:
    build:
      context: .
      args:
        TORCH_CUDA: cu121
    container_name: rss_ner
    command: python ner_worker.py
    environment:
      - DB_HOST=db
      - DB_PORT=5432
      - DB_NAME=${DB_NAME}
      - DB_USER=${DB_USER}
      - DB_PASS=${DB_PASS}
      - NER_LANG=es
      - NER_BATCH=64
    depends_on:
      db:
        condition: service_healthy
    restart: always

volumes:
  postgres_data:
  hf_cache: