server/tests/benchmark_pipeline.py

import asyncio
import time
import sys
import os

# Add parent directory to path for imports
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from app.services.medical_orchestrator import get_medical_orchestrator
from app.services.context_manager import EntryContext

async def benchmark():
    orchestrator = get_medical_orchestrator()
    
    # Test cases
    queries = [
        "hi how are you",
        "do you have hydrogels?",
        "muje sar ma drd ho rha hy kya karo", # Roman Urdu - Medical
    ]
    
    context = EntryContext(
        tenant_id="default_healthcare_tenant",
        user_id="benchmark_user"
    )
    
    print("\n" + "="*50)
    print("🚀 PIPELINE LATENCY BENCHMARK (9-LAYER CPU FLOW)")
    print("="*50)
    
    for query in queries:
        print(f"\nQUERY: '{query}'")
        start_time = time.time()
        
        try:
            response, confidence, is_emergency = await orchestrator.process_query(query, context)
            end_time = time.time()
            latency = end_time - start_time
            
            print(f"LATENCY: {latency:.3f}s")
            print(f"CONFIDENCE: {confidence:.2f}")
            print(f"RESPONSE: {response[:100]}...")
            
            # Goal: < 3s for CPU flow
            if latency < 3.0:
                print("✅ PASS (Under 3s)")
            else:
                print("⚠️  SLOW (Over 3s)")
                
        except Exception as e:
            print(f"❌ ERROR: {e}")
            import traceback
            traceback.print_exc()

if __name__ == "__main__":
    asyncio.run(benchmark())