The short answer
Inventory every source that can enter the dead-letter queue, monitor count and age against the actual platform retention behavior, assign business and technical owners, preserve event and business-action identities, and disposition each message as safe to redrive, requires transform, already applied, compensate manually, irrelevant with evidence, quarantined, or expired/lost. Never bulk-redrive an unclassified queue.
Key takeaways
Document the actual queue contract
Record provider, source queues/topics, event classes, receive threshold, retention, enqueue timestamp semantics, ordering, deduplication, redrive destination, rate controls, permissions, encryption, alarms, payload references, and owners. Verify these against the deployed platform.
AWS documents provider-specific retention, ordering, access-policy, and redrive behavior. Use it only for an SQS implementation; other systems differ.
Disposition before redrive
| Disposition | Evidence required | Action | Closure proof |
|---|---|---|---|
| Safe to redrive | Failure removed; idempotency/duplicate plan known | Controlled batch and rate | Source and business outcomes reconcile |
| Transform then process | Schema/version issue understood | Approved transform with lineage | Transformed and original identities linked |
| Already applied | External/local effect proven | Suppress duplicate processing | Business state correct |
| Manual/compensate | Automatic path unsafe or unavailable | Authorized bounded action | Receipt reconciliation |
| Quarantine/irrelevant | Security or scope reason | Restricted hold or supported exclusion | Owner and retention decision |
| Expired/lost | Retention or deletion evidence | Reconstruct affected population | Residual limitation recorded |
Control the redrive as a release
Freeze message IDs and payload/version references, destination, rate, order assumptions, code/config version, start/stop checkpoints, monitoring, abort condition, and rollback or compensation plan. Test a small representative batch first.
After redrive, reconcile provider events, local records, external effects, duplicates, omissions, and messages that returned to the queue.
Close with queue denominator, per-message dispositions, and reconciled redrive outcomes
Name the reviewed population, cutoff, evidence version, decision owner, unresolved exceptions, next checkpoint, and downstream records updated. Preserve the superseded state; a clean current screen is not a substitute for the correction or exception history.
Reopen the record if the population, authority, source version, external outcome, or dependent report changes after sign-off.
Operational checklist
Mark your progress, then save a working copy. Selections reset when you leave this page. A checked box is not an approval or evidence of completion.
0 of 8 marked
Edge cases
Sources and references
Follow each source to check the underlying claim. Access checks and professional review are different steps.