docker-compose.yml

# debuga.ai LLM Stack - Docker Compose de Laboratório
# Este é um exemplo genérico para experimentação local.
# NÃO representa a configuração de produção do debuga.ai.
#
# Pré-requisitos:
# - NVIDIA GPU com 24GB+ VRAM
# - NVIDIA Container Toolkit instalado
# - Docker 24.0+
# - Arquivo .env configurado (copie de .env.example)
#
# Uso:
#   docker compose up -d
#   curl http://localhost:8000/health

services:
  # ─── Motor de Inferência (vLLM) ─────────────────────────
  vllm:
    image: vllm/vllm-openai:latest
    container_name: debuga-vllm
    restart: unless-stopped
    ports:
      - "${VLLM_PORT:-8000}:8000"
    environment:
      - HUGGING_FACE_HUB_TOKEN=${HF_TOKEN}
    volumes:
      - model-cache:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    command: >
      --model ${VLLM_MODEL_ID:-Qwen/Qwen2.5-Coder-7B-Instruct}
      --tensor-parallel-size ${VLLM_TENSOR_PARALLEL_SIZE:-1}
      --gpu-memory-utilization ${VLLM_GPU_MEMORY_UTILIZATION:-0.90}
      --max-model-len ${VLLM_MAX_MODEL_LEN:-8192}
      --host 0.0.0.0
      --port 8000
      --trust-remote-code
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 120s

  # ─── Monitoramento (Opcional) ────────────────────────────
  prometheus:
    image: prom/prometheus:latest
    container_name: debuga-prometheus
    restart: unless-stopped
    ports:
      - "${PROMETHEUS_PORT:-9090}:9090"
    volumes:
      - ./monitoring/prometheus.yml:/etc/prometheus/prometheus.yml:ro
      - prometheus-data:/prometheus
    profiles:
      - monitoring

  grafana:
    image: grafana/grafana:latest
    container_name: debuga-grafana
    restart: unless-stopped
    ports:
      - "${GRAFANA_PORT:-3200}:3000"
    environment:
      - GF_SECURITY_ADMIN_PASSWORD=admin
    volumes:
      - grafana-data:/var/lib/grafana
    profiles:
      - monitoring

volumes:
  model-cache:
    name: debuga-model-cache
  prometheus-data:
    name: debuga-prometheus-data
  grafana-data:
    name: debuga-grafana-data