Dead letter recovery
Dead-letter access can expose failed customer payloads
Failed messages often retain the original request and diagnostic detail. Restrict inspection and replay according to the data and effects involved.
In this article
Review what the queue contains
Inspect representative synthetic failures to see which fields reach the dead-letter payload and associated logs. Validation errors can include values the normal workflow would never display.
Classify the stored information and choose retention and access appropriate to its purpose. Do not assume operational data is automatically safe for broad support access.
Keep secrets out of messages wherever the workflow can use a protected reference instead.
Separate inspection from execution
An operator may need to view a failure category without permission to replay the underlying business action. Give these capabilities distinct authority where appropriate.
Replaying a message can create a payment, dispatch or account change depending on the consumer. Treat it as an operational mutation, not merely moving text between queues.
Record who selected the scope and initiated the action.
A useful separation is a support view that exposes the error category and operation reference, alongside a restricted recovery action that selects explicit operation IDs. The support view need not include a customer's full document to explain that its schema is unsupported.
Test the replay request itself by substituting another tenant's operation ID in a synthetic environment. The server should reject the selection even if the operator's browser normally hides that item. Queue-console access and application recovery authority are different capabilities, so review both paths.
Validate the destination and tenant
Restrict recovery tooling to intended source and destination queues or processing paths. Preserve and validate tenant scope at the worker boundary.
A message copied into another environment should not inherit authority merely because the recovery operator can send it there.
Check encryption-key access and network controls for the chosen platform without granting broad account-level privileges unnecessarily.
Control exports and cleanup
Investigations may create local files or shared documents containing failed payloads. Provide a controlled alternative and remove temporary copies under policy.
Retain enough evidence to explain resolution without keeping unnecessary customer content indefinitely.
The review should establish that failed work remains recoverable and appropriately protected. A dead-letter queue is part of the application's data footprint and authority model, even when users never see it directly.
Primary sources
Amazon SQS: redrive permissionsOWASP: logging guidanceReferences checked 11 September 2026.