dev/predeploy.py

"""Dev only: refuse a Space deploy while any job depends on the live Space relay.

A push rebuilds the Space, which drops every relay WebSocket and fails the requests in flight. A job that is still
starting has no relay yet (it connects about 6 minutes in), and a relay that just dropped keeps its pending requests
for a 120 s grace period, so both count as live. Run before `git push`:

    python dev/predeploy.py [SPACE_URL]    # exit 0: safe to deploy; exit 1: live relays or PostTrain jobs (listed)
"""
import json, sys, urllib.request
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from compute_jobs import kind  # noqa: E402  (the same job purposes the dashboard shows)

url = (sys.argv[1] if len(sys.argv) > 1 else 'https://benchflow-posttrain-arena.hf.space').rstrip('/')
blockers = []
try:
    status = json.load(urllib.request.urlopen(url + '/api/relay/status', timeout=30))
except Exception as error:
    sys.exit(f'Could not read {url}/api/relay/status ({error}); check by hand before deploying.')
for r in status.get('connected', []):  # open or inside the reconnect grace period
    blockers.append(f"relay {r['run_id']}: {'open' if r.get('open') else 'reconnecting'}, {r.get('requests', 0)} requests, {r.get('in_flight', 0)} in flight")
try:
    from huggingface_hub import HfApi
    for job in HfApi().list_jobs(namespace='benchflow'):
        stage = str(job.status.stage).split('.')[-1]
        what = kind(getattr(job, 'labels', None) or {})
        if what and stage in ('RUNNING', 'SCHEDULING'):
            blockers.append(f"job {(job.labels or {}).get('name') or job.id} ({what}): {stage.lower()}")
except Exception as error:
    sys.exit(f'Could not list HF jobs ({type(error).__name__}); check https://huggingface.co/jobs/benchflow by hand before deploying.')
for line in blockers:
    print(line)
if blockers:
    sys.exit(f'{len(blockers)} live relay(s) or PostTrain job(s); deploying now could break them. Wait until they finish.')
print('No relays and no running PostTrain jobs: safe to deploy.')