---
title: "Comment by SMTasker on \"System Design Drill Day 13: Designing a Multi-Channel Notification System\" | Hashnode"
description: "The failover path seems to need one more distinction: a definite rejection versus a timeout after the provider may have accepted the SMS. A provider-scoped idem"
url: "https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837/comment/6aee2c1d84e31be5cdf51cdd"
---

[Discussion](https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837)

Comment [System Design Drill Day 13: Designing a Multi-Channel Notification System](https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837)

[SMTasker](https://hashnode.com/@smtasker)

[Nov 6](https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837/comment/6aee2c1d84e31be5cdf51cdd)

The failover path seems to need one more distinction: a definite rejection versus a timeout after the provider may have accepted the SMS. A provider-scoped idempotency key may not carry across to the fallback provider. I'd test 'A accepts, response is lost, B is selected' and document how that uncertain outcome is reconciled or deliberately retried. Row locking alone cannot settle what happened outside the database.

[Anish Ansari](https://hashnode.com/@anish78)

[Nov 7](https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837/comment/6aee2c1d84e31be5cdf51cdd#reply-6aef263bf07c51fd4860a8b0)

You're right that a lost response after A may have accepted is a different failure mode than a clean reject. I'd treat that path as uncertain and reconcile with a provider status check (or a careful deliberate retry policy) before falling over to B, since a provider-scoped key won't help the fallback.

[SMTasker](https://hashnode.com/@smtasker)

[Nov 10](https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837/comment/6aee2c1d84e31be5cdf51cdd#reply-6af37abd953abc28ba3865c2)

That distinction makes the failover policy clearer. One extra fixture I'd add is that the reconciliation lookup also times out. Preserve the uncertain state and the original provider reference, then make the next retry decision explicit. An inconclusive status check should not quietly turn an ambiguous send into a definite rejection.

[Anish Ansari](https://hashnode.com/@anish78)

[Nov 11](https://hashnode.com/posts/system-design-drill-day-13-designing-a-multi-channel-notification-system/6aec10be05f6196496e97837/comment/6aee2c1d84e31be5cdf51cdd#reply-6af4696d4b78a8243a6fce5b)

Good fixture to add. If the status check itself times out, keeping the send in that uncertain state with the original provider reference means the next retry is a deliberate call instead of a quiet rejection.
