archinario.
My learning

Delivering notifications after a spike

Message volume temporarily exceeds provider capacity. A queue retains work, but its capacity and waiting time are limited.

How to approach the problem

Balance buffer capacity, processing rate and duplicate protection. A larger queue alone does not speed up the provider.

Try it in a lab

Build a notification diagram and run the additional trial: normal load, a spike and a provider pause. Watch losses, duplicates and remaining backlog.

Labs use educational models. Results do not replace measurements of your system.