""" Load test for the /query endpoint. Fires N requests all at once (concurrently) and reports how long each one took, so you can see how badly requests queue up / delay each other under a simulated burst. Usage: python load_test.py --url http://localhost:7860/query --n 100 --api-key YOUR_KEY Adjust --payload-* args to match a real query for one of your test workspaces (channel/channel_identifier must resolve to a real workspace + Shopify creds in your DB, otherwise every request will just fail fast at the workspace lookup step and you won't be testing the real inventory/LLM path). """ import argparse import asyncio import statistics import time import httpx async def fire_one(client: httpx.AsyncClient, url: str, headers: dict, payload: dict, idx: int) -> dict: start = time.perf_counter() try: resp = await client.post(url, json=payload, headers=headers, timeout=60.0) elapsed = time.perf_counter() - start return { "idx": idx, "status": resp.status_code, "elapsed": elapsed, "error": None, } except Exception as e: elapsed = time.perf_counter() - start return { "idx": idx, "status": None, "elapsed": elapsed, "error": str(e), } async def main(args): headers = {"Content-Type": "application/json"} if args.api_key: headers["X-API-Key"] = args.api_key payload = { "query": args.query, "channel": args.channel, "channel_identifier": args.channel_identifier, } if args.post_caption: payload["post_caption"] = args.post_caption print(f"Firing {args.n} concurrent requests at {args.url} ...") overall_start = time.perf_counter() limits = httpx.Limits(max_connections=args.n + 10, max_keepalive_connections=args.n + 10) async with httpx.AsyncClient(limits=limits) as client: tasks = [ fire_one(client, args.url, headers, payload, i) for i in range(args.n) ] results = await asyncio.gather(*tasks) overall_elapsed = time.perf_counter() - overall_start # --- Report --- successes = [r for r in results if r["error"] is None and r["status"] and r["status"] < 400] failures = [r for r in results if r not in successes] print("\n" + "=" * 60) print(f"Total wall-clock time for all {args.n} requests: {overall_elapsed:.2f}s") print(f"Successful responses: {len(successes)} / {args.n}") print(f"Failed/errored responses: {len(failures)} / {args.n}") if successes: elapsed_times = [r["elapsed"] for r in successes] elapsed_times.sort() print("\n--- Latency stats (successful requests only) ---") print(f" min: {min(elapsed_times):.2f}s") print(f" max: {max(elapsed_times):.2f}s") print(f" mean: {statistics.mean(elapsed_times):.2f}s") print(f" median: {statistics.median(elapsed_times):.2f}s") if len(elapsed_times) >= 20: p90 = elapsed_times[int(len(elapsed_times) * 0.90)] p99 = elapsed_times[int(len(elapsed_times) * 0.99)] print(f" p90: {p90:.2f}s") print(f" p99: {p99:.2f}s") if failures: print("\n--- Failures (first 10 shown) ---") for r in failures[:10]: print(f" idx={r['idx']} status={r['status']} error={r['error']}") print("\n--- Per-request timeline (sorted by completion time) ---") for r in sorted(results, key=lambda r: r["elapsed"]): status_str = r["status"] if r["status"] else f"ERROR: {r['error']}" print(f" idx={r['idx']:>3} | {r['elapsed']:.2f}s | status={status_str}") print("=" * 60) if __name__ == "__main__": parser = argparse.ArgumentParser(description="Load test the /query endpoint") parser.add_argument("--url", default="http://localhost:7860/query", help="Full URL to /query endpoint") parser.add_argument("--n", type=int, default=100, help="Number of concurrent requests to fire") parser.add_argument("--api-key", default=None, help="API key for auth header") parser.add_argument("--query", default="do you have this in stock?", help="Query text to send") parser.add_argument("--channel", default="instagram", help="Channel value") parser.add_argument("--channel-identifier", default="123", help="Channel identifier (must exist in your DB)") parser.add_argument("--post-caption", default=None, help="Post caption text (optional)") args = parser.parse_args() asyncio.run(main(args))