I took the idea of separating it into business domains that you can logically deploy separately from the team at A Cloud Guru. During the last serverless conference they spoke about their journey. While they use CloudFormation, the same ideas are valid 🙂 (at least that’s the way it works for me)

That makes some sense, but how would you pass dependencies around between resources in different domains?

I create resources in a single domain. For example, my DynamoDB table or SQS queues, and those resources follow a specific naming standard (like payment-requestqueue-dev, where

payment

and

dev

are the part of the stack name). I export either the ARN or the name, based on what I need to allow other resources access. In my Payment domain, I need my Lambdas to access the SQS queues, so one of the first things I do is call out to the stack variables of the other stack to get the names (or ARNs) of those resources.

Call out to the stack variables, that is, go through the main stack object?

In my case, I’ve split them out as separate stacks (so splitting at the infra level)

so you force every resource for a given stack into one file?

and that’s where it makes sense to split them into reusable components and have those as modules (as you suggested in the main thread 🙂 )

which brings us back to my original question of how to organize those within a single stack 🙈 I guess. Thanks for the discussion

My code, while written in Go, is all on GitHub. This is the main repo https://github.com/retgits/acme-serverless and each of the business domains have their won repos (linked in the README)