aitbc/apps/gpu-service

AITBC GPU Service

Manages GPU resource operations.

Installation

cd /opt/aitbc
poetry install --with gpu-service

Database Setup

Create a separate database for the GPU service:

sudo -u postgres psql -f apps/gpu-service/scripts/setup-database.sql

Or manually:

CREATE DATABASE aitbc_gpu;
CREATE USER aitbc_gpu WITH PASSWORD 'password';
GRANT ALL PRIVILEGES ON DATABASE aitbc_gpu TO aitbc_gpu;

Running

# Development
python -m gpu_service.main

# Production (systemd)
sudo systemctl start gpu-service
sudo systemctl enable gpu-service

Endpoints

• GET /health - Health check
• GET /gpu/status - Get GPU status
• GET /v1/marketplace/edge-gpu/profiles - Get consumer GPU profiles
• GET /v1/marketplace/edge-gpu/metrics/{gpu_id} - Get edge GPU metrics
• POST /v1/marketplace/edge-gpu/scan/{miner_id} - Scan and register edge GPUs
• POST /v1/marketplace/edge-gpu/optimize/inference/{gpu_id} - Optimize ML inference

Testing

Prerequisites

• PostgreSQL running and aitbc_gpu database created
• Poetry dependencies installed

Database Setup

sudo -u postgres psql -f scripts/setup-database.sql

Start Service (Development)

python -m gpu_service.main

Health Check

curl http://localhost:8101/health

Expected response:

{"status": "healthy", "service": "gpu-service"}

GPU Status

curl http://localhost:8101/gpu/status

Expected response:

{
  "status": "operational",
  "service": "gpu-service",
  "message": "GPU service is running"
}

Get Consumer GPU Profiles

curl http://localhost:8101/v1/marketplace/edge-gpu/profiles

Expected response:

[
  {
    "profile_id": "consumer_nvidia_a100",
    "name": "NVIDIA A100",
    "architecture": "NVIDIA",
    "memory_gb": 80,
    "cuda_cores": 6912,
    "tensor_cores": 432,
    "compute_capability": "8.0",
    "typical_use_cases": ["ml_training", "inference", "hpc"]
  }
]

Test Through Gateway

1. Start the API gateway:

   python -m api_gateway.main
   

2. Test GPU endpoints through the gateway:

   curl http://localhost:8080/gpu/health
   curl http://localhost:8080/gpu/v1/marketplace/edge-gpu/profiles
   

For comprehensive testing procedures, see MICROSERVICES_TESTING_GUIDE.md.

Service Configuration

• Port: 8101
• Database: aitbc_gpu
• Gateway route: /gpu/*

Migration Status

Completed:

• Extracted GPU domain models (GPUArchitecture, GPURegistry, ConsumerGPUProfile, EdgeGPUMetrics, GPUBooking, GPUReview)
• Extracted GPU services (EdgeGPUService)
• Extracted GPU data (consumer_gpu_profiles)
• Set up database session management
• Extracted GPU router endpoints
• Removed edge_gpu router from coordinator-api
• Created systemd service configuration
• Created database setup script

Remaining:

• Extract additional GPU routers (gpu_multimodal_health.py, miner.py) if needed
• Run database migration script to create aitbc_gpu database
• Install and enable systemd service
• End-to-end testing with gateway