Check whether redrive changes the identity used for deduplication

Inspect the complete identity path

Trace the producer's operation identifier, queue message identifier and worker deduplication key. Determine which values survive failure and replay.

Amazon SQS documents new message identifiers and enqueue times on redrive. An application relying only on those transport fields can treat recovered work as a new operation.

Keep a durable business identity in the message or an authoritative referenced record where the design requires it.

Review the effect boundary

Confirm that the duplicate check protects the actual mutation. A receipt-level marker does not prevent a worker from repeating an external action after a crash.

Use transactional state where possible and an appropriate idempotency or reconciliation mechanism for remote effects.

Test the case where the effect completed but its acknowledgement or local recording did not.

Examine ordering and validity

Determine how recovered messages interact with new traffic. A FIFO transport does not automatically restore the original business sequence once work has been removed and later reintroduced.

Check source revisions, state transitions and expiry rules in the consumer. An old command may be correctly rejected even after its technical failure is fixed.

Keep that outcome distinct from another processing failure so operators do not replay it forever.

Prove identity survives a changed delivery record

Ask the team to demonstrate a recovered message whose transport identifier changed while its business identity stayed the same. Then terminate the worker after its test effect succeeds but before it acknowledges completion. On the next attempt, there should be one business effect and a recoverable result.

This experiment tests the exact gap that queue movement metrics miss. If the implementation depends on the original transport identifier never changing, the review has found a concrete defect rather than a general concern about duplicates.