After signup I do background_tasks.add_task(send_welcome_email, user.email). Most users get it, maybe 1 in 20 don't. No error in the logs that I can find. Deployed on Railway.
Added logging inside send_welcome_email, and for the missing ones the log line never appears. ChatGPT says to use Celery but that seems like a lot for one email.
BackgroundTasks run in the same process after the response is sent. If that process restarts (deploy, crash, Railway recycling the container, scaling down), queued tasks are simply gone. No retry, no trace. That fits "1 in 20, no log line".
For anything that must happen, use a real queue. Doesn't have to be Celery: arq or rq with Redis is ~20 lines. Or, simplest of all, a pending_emails table that a small cron picks up and marks sent.
I'd push back slightly: before adding infra, check whether the task is throwing. Exceptions in background tasks are easy to miss because the response already went out. Wrap the body in try/except with logger.exception. In my experience "silently missing" is more often an unhandled SMTP timeout than a restart. If it still goes missing with logging in place, then yes, a table + cron.
AskVibe