services:
  genai2openai:
    build:
      context: .
      dockerfile: Dockerfile
      args:
        GENAI_BUILD_COMMIT: ${GENAI_BUILD_COMMIT:-}
        GENAI_BUILD_COMMIT_TIME: ${GENAI_BUILD_COMMIT_TIME:-}
        GENAI_BUILD_DIRTY: ${GENAI_BUILD_DIRTY:-0}
    restart: unless-stopped
    init: true
    stop_grace_period: 45s
    ports:
      - "${HOST_PORT:-5000}:${APP_PORT:-5000}"
    environment:
      APP_PORT: ${APP_PORT:-5000}
      GENAI_TOKEN: ${GENAI_TOKEN:-}
      KEYSTORE_PATH: ${KEYSTORE_PATH:-}
      TOKEN_CHECK_INTERVAL: ${TOKEN_CHECK_INTERVAL:-60}
      GENAI_MAX_RETRIES: ${GENAI_MAX_RETRIES:-10}
      GENAI_RETRY_BACKOFF: ${GENAI_RETRY_BACKOFF:-0.5}
      GENAI_MODEL_CACHE: ${GENAI_MODEL_CACHE:-/app/.cache/models/models.json}
      GENAI_TOKENIZER_CACHE: /opt/genai2openai/hf-assets
      GENAI_TOKENIZER_OFFLINE: "1"
      HF_HUB_OFFLINE: "1"
      API_KEY: ${PROXY_API_KEY:-}
      APP_DEBUG: ${APP_DEBUG:-0}
      CLAUDE_HAIKU_MODEL: ${CLAUDE_HAIKU_MODEL:-deepseek-chat}
      CLAUDE_SONNET_MODEL: ${CLAUDE_SONNET_MODEL:-chatglm}
      CLAUDE_OPUS_MODEL: ${CLAUDE_OPUS_MODEL:-chatglm}
      GUNICORN_WORKERS: ${GUNICORN_WORKERS:-2}
      GUNICORN_THREADS: ${GUNICORN_THREADS:-8}
      GUNICORN_TIMEOUT: ${GUNICORN_TIMEOUT:-180}
      GUNICORN_GRACEFUL_TIMEOUT: ${GUNICORN_GRACEFUL_TIMEOUT:-30}
      GUNICORN_KEEPALIVE: ${GUNICORN_KEEPALIVE:-10}
      GUNICORN_MAX_REQUESTS: ${GUNICORN_MAX_REQUESTS:-0}
      GUNICORN_MAX_REQUESTS_JITTER: ${GUNICORN_MAX_REQUESTS_JITTER:-0}
    volumes:
      - type: bind
        source: ${KEYSTORE_HOST_PATH:-./docker-deploy.keystore}
        target: /app/docker-deploy.keystore
        read_only: false
      - model-cache:/app/.cache/models
    command:
      - /bin/sh
      - -lc
      - |
        set -eu
        if [ -z "$${GENAI_TOKEN:-}" ] && [ -z "$${KEYSTORE_PATH:-}" ]; then
          echo "Either GENAI_TOKEN or KEYSTORE_PATH must be set." >&2
          exit 1
        fi

        if [ "$${APP_DEBUG:-0}" = "1" ]; then
          log_level=debug
        else
          log_level=info
        fi

        exec gunicorn \
          --bind "0.0.0.0:$${APP_PORT}" \
          --worker-class gthread \
          --workers "$${GUNICORN_WORKERS:-2}" \
          --threads "$${GUNICORN_THREADS:-8}" \
          --timeout "$${GUNICORN_TIMEOUT:-180}" \
          --graceful-timeout "$${GUNICORN_GRACEFUL_TIMEOUT:-30}" \
          --keep-alive "$${GUNICORN_KEEPALIVE:-10}" \
          --max-requests "$${GUNICORN_MAX_REQUESTS:-0}" \
          --max-requests-jitter "$${GUNICORN_MAX_REQUESTS_JITTER:-0}" \
          --log-level "$${log_level}" \
          --access-logfile - \
          --error-logfile - \
          --capture-output \
          genai_proxy.wsgi:app
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "python -c \"import os, urllib.request; urllib.request.urlopen(f'http://127.0.0.1:{os.environ.get(\\\"APP_PORT\\\", \\\"5000\\\")}/health', timeout=5)\"",
        ]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

volumes:
  model-cache: