What is a Dead Letter Queue?

In any queue-based architecture, some percentage of messages will fail — a downstream API is down, the payload is malformed, or a bug throws on a specific input shape. Without a dead letter queue, that failure either blocks the whole queue (a poison message stuck at the head) or vanishes, and nobody notices data went missing until a customer complains. The standard pattern caps retries with backoff, then routes the message to a separate DLQ instead of looping forever. From there, an operator or an automated reconciler can inspect the failure reason, fix the root cause, and replay the message back into the main queue. Mature systems alert on DLQ depth so a spike in failures gets noticed within minutes, not weeks. Dead letter queues are a resilience primitive, not an error-hiding one — they only help if something actually watches them.