import json, os, subprocess, time, urllib.request
from pathlib import Path
root=Path('/home/lab/bench/ornith/modular/sglang')
shared=Path('/home/lab/bench/ornith/modular/shared')
ready=False
for attempt in range(450):
    try:
        with urllib.request.urlopen('http://127.0.0.1:8130/health',timeout=2) as r:
            if r.status==200:ready=True;break
    except Exception:pass
    if attempt>15 and subprocess.check_output(['tmux','display-message','-p','-t','%68','#{pane_current_command}'],text=True).strip()=='bash':
        raise SystemExit('FlashInfer server exited before ready; inspect server log.')
    time.sleep(2)
if not ready:raise SystemExit('FlashInfer startup timeout')
base=['--base','http://127.0.0.1:8130/v1','--model','ornith-sglang']
quality=root/'results/quality-flashinfer-prefill.json'
subprocess.run(['python3',str(shared/'quality.py'),*base,'--out',str(quality)])
rows=json.loads(quality.read_text())['results']
must_pass={'multiply','arithmetic','json','square','sum_list','tool_call'}
if any(not row['pass'] for row in rows if row['case'] in must_pass):
    raise SystemExit('FlashInfer quality regression: do not promote or run long comparison.')
cmd=['/home/lab/venvs/vllm/bin/python',str(shared/'bench.py'),*base,
     '--out',str(root/'results/flashinfer-prefill-compare'),'--lengths','32768,131072',
     '--concurrency','1','--repeats','1','--seed','4400',
     '--gpu','GPU-f4a988cb-c310-b618-5a9c-4462d4d33add']
raise SystemExit(subprocess.run(cmd,env={**os.environ,'CUDA_VISIBLE_DEVICES':''}).returncode)
