There aren't really heroics involved in doing this. e.g. in a promise-based framework, put your request onto a Queue[(RequestData, Promise)]. Await the promise as you would with a normal db request in your web handler. Have a db worker thread that pulls off the queue with a max wait time. Now pretend you received a batch file for your business logic. After processing the batch, complete the promises with their results. The basic scaffolding is maybe 20 LoC in a decent language/framework.

Much easier than alternatives like sharding or microservices or eventual consistency anyway.

Basic toy example I made a while back to demonstrate the technique using a Repository Pattern: https://gist.github.com/ndriscoll/6a1df10c8ea1a25c48c615c409...

I think somewhere I have a more built-out example with error handling and and where you have more than just "body" in the model. In the real world the business logic for processing the batch and error handling is going to be a decent amount more complicated. etc. etc. But the point is more about how the batching mechanic works.