In a microservices system with a platform-level message bus, when a subscriber fails to process a message (due to server error) does it make sense to build app-level processing queues to manage retries/failures? It seems like that's obviously something you should build when you don't have the luxury of an existing messaging API/deadletter queue, but with one, it seems like the main purpose of the DLQ.
I've interacted with multiple respectable developers now who have said that they wouldn't want to use a messaging API as the source for app-level failures. Some of the points make sense circumstantially, but it really seems like we just condemn ourselves to "reinvent the message queue" with every microservice/subscription that needs this kind of recovery functionality.
What am I missing?