claude-forge: Skill for Claude Code

Set Up Celery for Background Jobs (claudeforge)

Follow async-and-queues/celery.md (deeper) and backend/fastapi/05-async-and-celery.md. Steps:

1. Confirm with user:
   • Existing app (FastAPI / Django / Flask)?
   • Use cases: emails, scheduled jobs, heavy CPU work?
   • Already have Redis? If not, set up via Docker compose first.
2. Install + configure:
   • Add celery[redis]>=5.4 and redis>=5.2 to pyproject.toml
   • Create workers/celery_app.py with the locked config (acks_late, prefetch=1, max_tasks_per_child, queue routing, beat schedule)
   • Update env to include REDIS_URL
3. Define tasks:
   • Create workers/{feature}_tasks.py per feature
   • Each task: @celery_app.task(bind=True, autoretry_for=..., retry_backoff=True, max_retries=5)
   • Idempotency: every task either checks a DB flag, uses an idempotency key, or is naturally idempotent. Do not skip this.
   • For async DB operations: wrap in asyncio.run(...) inside the task
4. Wire into the app:
   • Add task.delay(args) calls from routes/services
   • Don't put heavy work directly in route handlers
5. Run workers:
   • Dev: celery -A {{project-slug}}.workers.celery_app worker -Q emails -c 4 -n emails@%h
   • Plus separate worker per queue, plus exactly one beat
6. Production setup:
   • systemd template ({{project-slug}}[email protected]) so you can systemctl enable {{project-slug}}-worker@emails
   • Or Docker Compose with separate worker services per queue
   • Configure beat with persisted schedule file
7. Monitoring:
   • Install Flower: flower --basic-auth=admin:CHANGEME --port=5555 (lock down with auth + IP filter)
   • Add OpenTelemetry: CeleryInstrumentor().instrument()
   • Add Sentry: CeleryIntegration() in sentry_sdk.init
8. Verify: enqueue a test task from a route, observe it process, confirm metric/trace appears.

Always make tasks idempotent. Always run exactly one beat. Always split queues by SLA.