docker-compose.yml

# Docker Compose file for Food Rescue RL — local-dev orchestration.
#
# Usage:
#   docker compose up mlflow                                # tracking server only
#   docker compose --profile train run train                # one-shot training
#   docker compose --profile serve up serve                 # FastAPI server
#   docker compose --profile serve --profile train up       # all services
#
# ── Shared network so containers can reach each other by name ──────────────
networks:
  rescue_net:
    driver: bridge

# ── Named volumes ──────────────────────────────────────────────────────────
volumes:
  mlflow_data: # persists MLflow runs/artifacts between restarts
  experiments_data: # trained policies shared between train & serve

# ══════════════════════════════════════════════════════════════════════════
services:
  # ── 1. MLflow tracking server ──────────────────────────────────────────
  mlflow:
    image: ghcr.io/mlflow/mlflow:v2.10.0
    container_name: food_rescue_mlflow
    networks: [rescue_net]
    ports:
      - "5000:5000"
    volumes:
      - mlflow_data:/mlruns
    command: >
      mlflow server
        --host 0.0.0.0
        --port 5000
        --backend-store-uri /mlruns
        --default-artifact-root /mlruns
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:5000/health"]
      interval: 10s
      timeout: 5s
      retries: 5

  # ── 2. Training container (run-once, profile-gated) ────────────────────
  train:
    profiles: [train]
    build:
      context: .
      dockerfile: Dockerfile.train
    container_name: food_rescue_train
    networks: [rescue_net]
    depends_on:
      mlflow:
        condition: service_healthy
    volumes:
      - experiments_data:/app/experiments # persist trained policies
    environment:
      - CONFIG=${CONFIG:-configs/qlearning_v1.yaml}
      - MLFLOW_TRACKING_URI=http://mlflow:5000
    # Override CONFIG on the CLI:
    # docker compose --profile train run -e CONFIG=configs/dqn_v2_holiday.yaml train

  # ── 3. FastAPI inference server (profile-gated) ────────────────────────
  serve:
    profiles: [serve]
    build:
      context: .
      dockerfile: Dockerfile.serve
    container_name: food_rescue_api
    networks: [rescue_net]
    depends_on:
      mlflow:
        condition: service_healthy
    ports:
      - "8000:8000"
    volumes:
      - experiments_data:/app/experiments # reads trained policies
    environment:
      - MLFLOW_TRACKING_URI=http://mlflow:5000
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 15s
      timeout: 5s
      retries: 3