| """32 spawned CPU processes against a local fake teacher. No paid API requests.""" |
| import argparse |
| import json |
| import os |
| from http.server import BaseHTTPRequestHandler,ThreadingHTTPServer |
| import threading |
| import time |
| from pathlib import Path |
| from .config import load_config |
| from .io import run_path,write_json,sha256,digest |
| from .rewards.api_group import score_groups |
| from .rewards.base import SLOTS |
|
|
| def main(): |
| p=argparse.ArgumentParser();p.add_argument("--run-id",default="smoke_v1");p.add_argument('--backend',choices=['processes','threads'],default='processes');a=p.parse_args() |
| c=load_config("rl_posttrain/configs/api_b_only.yaml");run=run_path(a.run_id);root=run/"reports/fake_api_stress"/str(time.time_ns()) |
| from rl_posttrain.tests.conftest import fake_group |
| groups=[fake_group(root,c,i) for i in range(32)] |
| mutex=threading.Lock();active=0;peak=0;requests=0 |
| barrier=threading.Barrier(32) |
| class Handler(BaseHTTPRequestHandler): |
| def log_message(self,*args):pass |
| def do_POST(self): |
| nonlocal active,peak,requests |
| body=json.loads(self.rfile.read(int(self.headers["Content-Length"]))) |
| images=sum(p["type"]=="image_url" for p in body["messages"][0]["content"]) |
| if images!=124:raise ValueError("Incorrect fake stress payload") |
| with mutex:active+=1;peak=max(peak,active);requests+=1 |
| if a.backend=='threads':barrier.wait(timeout=60) |
| else:time.sleep(2) |
| data=json.dumps({"choices":[{"message":{"content":json.dumps(dict(zip(SLOTS,[25,50,75,100])))}}]}).encode() |
| self.send_response(200);self.send_header("Content-Type","application/json");self.send_header("Content-Length",str(len(data)));self.end_headers();self.wfile.write(data) |
| with mutex:active-=1 |
| class Server(ThreadingHTTPServer):request_queue_size=128 |
| server=Server(("127.0.0.1",0),Handler);thread=threading.Thread(target=server.serve_forever,daemon=True);thread.start() |
| c["reward"].update(endpoint=f"http://127.0.0.1:{server.server_port}",workers=32,concurrency_backend=a.backend,rpm=100000,tpm=1000000000,timeout_seconds=60) |
| previous=os.environ.get("STATEPLAY_GPT_API_KEY");os.environ["STATEPLAY_GPT_API_KEY"]="software-test-placeholder" |
| start=time.monotonic() |
| try: |
| values=score_groups(c,root,groups,32) |
| |
| again=score_groups(c,root,groups,32) |
| assert values==again and requests==32 |
| report={"kind":"mock_local_API_server","workers":32,"backend":"threads" if a.backend=='threads' else "spawned_processes","requests":requests,"peak_server_inflight":peak,"groups":32,"images_per_request":124,"seconds":time.monotonic()-start,"cached_repeat_added_requests":0,"paid_API_requests":0} |
| write_json(run/("reports/api_32_thread_stress.json" if a.backend=='threads' else "reports/api_32_process_stress.json"),report);print(report,flush=True) |
| finally: |
| server.shutdown() |
| if previous is None:os.environ.pop("STATEPLAY_GPT_API_KEY",None) |
| else:os.environ["STATEPLAY_GPT_API_KEY"]=previous |
| if __name__=="__main__":main() |
|
|
