DLQ and Redrive Beat Manual Database Polling for Retrying Failed Nightly Jobs

For a small app retrying failed nightly payment-reconciliation jobs, the article recommends a queue with a dead-letter queue and redrive as the default. Manual database polling is only reasonable while concurrency stays tiny and the table itself serves as the needed audit record. Because the job runs overnight, shaving milliseconds off dispatch matters little, while duplicate payment effects and retry-state maintenance are costly. Polling tends to grow into a queue-like system once workers compete or operators need backoff and poison-message isolation. The author suggests keeping cron as the trigger, using a queue for delivery, a DLQ for exhausted jobs, and the application database for durable history, with idempotency keys preventing double-applied settlements.