Commit ·
45dd526
1
Parent(s): 5db9737
feat: container entrypoint provisions Redis + supervises Celery worker
Browse files- entrypoint.py: resolves REDIS_URL/CELERY_BROKER_URL/CELERY_RESULT_BACKEND
from env (Space secrets win), else provisions Upstash free-tier DBs with
deterministic v4-form idempotency keys (restart-stable, re-fetchable);
runs Celery worker + uvicorn, restarts worker on death.
- celery_tasks.py: explicit broker_use_ssl/redis_backend_use_ssl with
CERT_REQUIRED for rediss:// brokers (TLS verification stays ON).
- Dockerfile / Dockerfile.hf: CMD -> python entrypoint.py.
Co-Authored-By: Claude Code <noreply@anthropic.com>
- Dockerfile +3 -2
- Dockerfile.hf +1 -1
- app/tasks/celery_tasks.py +8 -0
- entrypoint.py +144 -0
Dockerfile
CHANGED
|
@@ -12,9 +12,10 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
|
|
| 12 |
COPY requirements.txt .
|
| 13 |
RUN pip install --no-cache-dir -r requirements.txt
|
| 14 |
|
| 15 |
-
# Copy app source
|
| 16 |
COPY app/ ./app/
|
|
|
|
| 17 |
|
| 18 |
EXPOSE 7860
|
| 19 |
|
| 20 |
-
CMD ["
|
|
|
|
| 12 |
COPY requirements.txt .
|
| 13 |
RUN pip install --no-cache-dir -r requirements.txt
|
| 14 |
|
| 15 |
+
# Copy app source + container entrypoint (provisions Redis, supervises worker)
|
| 16 |
COPY app/ ./app/
|
| 17 |
+
COPY entrypoint.py .
|
| 18 |
|
| 19 |
EXPOSE 7860
|
| 20 |
|
| 21 |
+
CMD ["python", "entrypoint.py"]
|
Dockerfile.hf
CHANGED
|
@@ -16,4 +16,4 @@ COPY . .
|
|
| 16 |
|
| 17 |
EXPOSE 7860
|
| 18 |
|
| 19 |
-
CMD ["
|
|
|
|
| 16 |
|
| 17 |
EXPOSE 7860
|
| 18 |
|
| 19 |
+
CMD ["python", "entrypoint.py"]
|
app/tasks/celery_tasks.py
CHANGED
|
@@ -33,6 +33,14 @@ celery_app.conf.update(
|
|
| 33 |
task_reject_on_worker_lost=True,
|
| 34 |
)
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
|
| 37 |
@celery_app.task(bind=True, max_retries=3, rate_limit="100/m")
|
| 38 |
def bulk_validate_task(self, emails: list, webhook_url: str = None):
|
|
|
|
| 33 |
task_reject_on_worker_lost=True,
|
| 34 |
)
|
| 35 |
|
| 36 |
+
# Hosted Redis (Upstash) arrives over rediss:// — always verify the server
|
| 37 |
+
# certificate. Explicit CERT_REQUIRED so Celery never silently downgrades.
|
| 38 |
+
import ssl as _ssl
|
| 39 |
+
if settings.celery_broker_url.startswith("rediss"):
|
| 40 |
+
celery_app.conf.broker_use_ssl = {"ssl_cert_reqs": _ssl.CERT_REQUIRED}
|
| 41 |
+
if settings.celery_result_backend.startswith("rediss"):
|
| 42 |
+
celery_app.conf.redis_backend_use_ssl = {"ssl_cert_reqs": _ssl.CERT_REQUIRED}
|
| 43 |
+
|
| 44 |
|
| 45 |
@celery_app.task(bind=True, max_retries=3, rate_limit="100/m")
|
| 46 |
def bulk_validate_task(self, emails: list, webhook_url: str = None):
|
entrypoint.py
ADDED
|
@@ -0,0 +1,144 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Container entrypoint for BRIDGE Modul-X on Hugging Face Spaces.
|
| 4 |
+
|
| 5 |
+
Responsibilities
|
| 6 |
+
----------------
|
| 7 |
+
1. Ensure Redis/Celery env vars point at a reachable Redis:
|
| 8 |
+
- Space secrets win: REDIS_URL / CELERY_BROKER_URL / CELERY_RESULT_BACKEND.
|
| 9 |
+
- Otherwise provision Upstash free-tier databases (cache, celery-broker,
|
| 10 |
+
celery-result) via the unauthenticated start-redis endpoint with
|
| 11 |
+
*stable* Idempotency-Keys, so every container restart re-fetches the
|
| 12 |
+
same databases. Free DBs expire after ~3 days unless claimed at the
|
| 13 |
+
console URL printed in the boot logs (or unless the account-backed
|
| 14 |
+
secrets above are set).
|
| 15 |
+
2. Supervise two child processes:
|
| 16 |
+
- Celery worker (bulk validation queue)
|
| 17 |
+
- uvicorn (FastAPI web app on $PORT / 7860)
|
| 18 |
+
The worker is restarted if it dies; if the web process dies the
|
| 19 |
+
container exits and the platform restarts it.
|
| 20 |
+
"""
|
| 21 |
+
import os
|
| 22 |
+
import signal
|
| 23 |
+
import subprocess
|
| 24 |
+
import sys
|
| 25 |
+
import time
|
| 26 |
+
import uuid
|
| 27 |
+
|
| 28 |
+
# Fixed namespace so the per-store Idempotency-Keys are deterministic.
|
| 29 |
+
_NS = uuid.UUID("7c9e6679-7425-40de-944b-e07fc1f90ae7")
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
def _stable_db_id(store: str) -> str:
|
| 33 |
+
"""Deterministic UUIDv4-FORM id (Upstash rejects non-v4 idempotency keys):
|
| 34 |
+
derive bytes from uuid5, then patch version/variant bits to v4/RFC-4122."""
|
| 35 |
+
b = bytearray(uuid.uuid5(_NS, f"bridge-modul-x/{store}").bytes)
|
| 36 |
+
b[6] = (b[6] & 0x0F) | 0x40
|
| 37 |
+
b[8] = (b[8] & 0x3F) | 0x80
|
| 38 |
+
return str(uuid.UUID(bytes=bytes(b)))
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
def _provision(store: str):
|
| 42 |
+
"""Create/re-fetch a free-tier Upstash DB. Returns a rediss:// URL or None."""
|
| 43 |
+
from app.utils.upstash_free import create_database
|
| 44 |
+
|
| 45 |
+
# First try the stable key (re-fetch after restart); if Upstash rejects
|
| 46 |
+
# it (expired DB), fall back to one fresh random key.
|
| 47 |
+
for attempt, db_id in enumerate([_stable_db_id(store), ""]):
|
| 48 |
+
try:
|
| 49 |
+
db = create_database(db_id=db_id)
|
| 50 |
+
print(
|
| 51 |
+
f"[ENTRYPOINT] {store}: upstash db {db.db_id} ok "
|
| 52 |
+
f"(expires {db.expires}) - claim to keep it: {db.console_url}",
|
| 53 |
+
flush=True,
|
| 54 |
+
)
|
| 55 |
+
return db.connection_string(0)
|
| 56 |
+
except Exception as exc:
|
| 57 |
+
print(f"[ENTRYPOINT] {store}: provision failed (try {attempt}): {exc}", flush=True)
|
| 58 |
+
return None
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
def _resolve(env_name: str, store: str):
|
| 62 |
+
url = os.environ.get(env_name, "").strip()
|
| 63 |
+
if url and "localhost" not in url and "127.0.0.1" not in url:
|
| 64 |
+
return url
|
| 65 |
+
print(f"[ENTRYPOINT] {env_name} unset/local - provisioning '{store}' on Upstash free tier", flush=True)
|
| 66 |
+
return _provision(store)
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
def main():
|
| 70 |
+
cache_url = _resolve("REDIS_URL", "cache")
|
| 71 |
+
broker_url = _resolve("CELERY_BROKER_URL", "celery-broker") or cache_url
|
| 72 |
+
result_url = _resolve("CELERY_RESULT_BACKEND", "celery-result") or broker_url
|
| 73 |
+
|
| 74 |
+
if cache_url:
|
| 75 |
+
os.environ["REDIS_URL"] = cache_url
|
| 76 |
+
if broker_url:
|
| 77 |
+
os.environ["CELERY_BROKER_URL"] = broker_url
|
| 78 |
+
if result_url:
|
| 79 |
+
os.environ["CELERY_RESULT_BACKEND"] = result_url
|
| 80 |
+
if not broker_url:
|
| 81 |
+
print("[ENTRYPOINT] WARNING: no Redis reachable - bulk jobs use the synchronous fallback", flush=True)
|
| 82 |
+
|
| 83 |
+
port = os.environ.get("PORT", "7860")
|
| 84 |
+
children = {}
|
| 85 |
+
state = {"stopping": False}
|
| 86 |
+
|
| 87 |
+
def spawn(kind):
|
| 88 |
+
if kind == "worker":
|
| 89 |
+
cmd = [
|
| 90 |
+
sys.executable, "-m", "celery",
|
| 91 |
+
"-A", "app.tasks.celery_tasks:celery_app", "worker",
|
| 92 |
+
"--loglevel=info", "--concurrency=4",
|
| 93 |
+
]
|
| 94 |
+
else:
|
| 95 |
+
cmd = [
|
| 96 |
+
sys.executable, "-m", "uvicorn", "app.main:app",
|
| 97 |
+
"--host", "0.0.0.0", "--port", port,
|
| 98 |
+
]
|
| 99 |
+
proc = subprocess.Popen(cmd, env=os.environ)
|
| 100 |
+
children[kind] = proc
|
| 101 |
+
print(f"[ENTRYPOINT] started {kind} (pid {proc.pid})", flush=True)
|
| 102 |
+
return proc
|
| 103 |
+
|
| 104 |
+
def on_signal(signum, frame):
|
| 105 |
+
state["stopping"] = True
|
| 106 |
+
for proc in list(children.values()):
|
| 107 |
+
try:
|
| 108 |
+
proc.terminate()
|
| 109 |
+
except Exception:
|
| 110 |
+
pass
|
| 111 |
+
|
| 112 |
+
signal.signal(signal.SIGTERM, on_signal)
|
| 113 |
+
signal.signal(signal.SIGINT, on_signal)
|
| 114 |
+
|
| 115 |
+
spawn("worker")
|
| 116 |
+
web = spawn("web")
|
| 117 |
+
|
| 118 |
+
while not state["stopping"]:
|
| 119 |
+
if web.poll() is not None:
|
| 120 |
+
print(f"[ENTRYPOINT] web exited ({web.returncode}) - shutting down", flush=True)
|
| 121 |
+
break
|
| 122 |
+
worker = children["worker"]
|
| 123 |
+
if worker.poll() is not None:
|
| 124 |
+
print(f"[ENTRYPOINT] worker exited ({worker.returncode}) - restarting in 5s", flush=True)
|
| 125 |
+
time.sleep(5)
|
| 126 |
+
if not state["stopping"]:
|
| 127 |
+
spawn("worker")
|
| 128 |
+
time.sleep(2)
|
| 129 |
+
|
| 130 |
+
for proc in list(children.values()):
|
| 131 |
+
try:
|
| 132 |
+
proc.terminate()
|
| 133 |
+
proc.wait(timeout=10)
|
| 134 |
+
except Exception:
|
| 135 |
+
try:
|
| 136 |
+
proc.kill()
|
| 137 |
+
except Exception:
|
| 138 |
+
pass
|
| 139 |
+
|
| 140 |
+
sys.exit(web.returncode or 0)
|
| 141 |
+
|
| 142 |
+
|
| 143 |
+
if __name__ == "__main__":
|
| 144 |
+
main()
|