How to Drain an SQS Queue (Without Reprocessing)

Why messages keep reappearing

SQS doesn't remove a message when you receive it - it just hides it for the visibility timeout. You're expected to process it and then explicitly delete it. If you don't delete, the timeout expires and SQS redelivers it, so the same messages get processed again and again and the depth never drops. That's at-least-once delivery working as designed - you just skipped the acknowledgement.

The correct receive -> process -> delete loop

import boto3
sqs = boto3.client("sqs")
url = "https://sqs.../orders"

while True:
    resp = sqs.receive_message(QueueUrl=url, MaxNumberOfMessages=10, WaitTimeSeconds=20)
    msgs = resp.get("Messages", [])
    if not msgs:
        break                          # queue drained
    for m in msgs:
        process(m)                     # do the work
        sqs.delete_message(            # ACK it - this is what removes it
            QueueUrl=url, ReceiptHandle=m["ReceiptHandle"])

delete_message with the message's ReceiptHandle is the acknowledgement. No delete = redelivery.

The details that matter

• Delete only after success. If process() throws, don't delete - let the message reappear and retry. Pair this with a dead-letter queue so a poison message doesn't loop forever.
• ReceiptHandle is per-receive, not the message id - it changes each time you receive the message; use the one from this receive.
• Visibility timeout must exceed processing time. If processing takes longer than the timeout, SQS redelivers while you're still working - a second worker double-processes. Raise the timeout or extend it (change_message_visibility).
• Long polling (WaitTimeSeconds=20) avoids busy-looping on an empty queue and cuts cost.
• Idempotent processing - because delivery is at-least-once, design process() so a duplicate is harmless (dedupe on a message/business id).