Never A Dead Letter

This worker has run in production for a year. The queue has a redrive policy with maxReceiveCount of 5 and a dead-letter queue. The error log shows about 4,000 render failed lines a day. The dead-letter queue has never held a single message.

while True:
    resp = sqs.receive_message(QueueUrl=Q, MaxNumberOfMessages=10, WaitTimeSeconds=20)
    for m in resp.get("Messages", []):
        try:
            render(m["Body"])
        except Exception:
            log.exception("render failed")
        finally:
            sqs.delete_message(QueueUrl=Q, ReceiptHandle=m["ReceiptHandle"])

Why is the dead-letter queue empty?

maxReceiveCount of 5 means five failures inside one visibility window, and these failures are spread across separate receives

The finally block deletes the message on the failure path too, so no message is ever received a second time, the receive count never reaches 5, and roughly 4,000 jobs a day are silently thrown away

log.exception handles the exception, so boto3 reports the receive as successful and SQS never increments the receive count for that message

The redrive policy only applies to messages whose visibility timeout expires while in flight, and this worker always finishes inside the timeout