A terragrunt best-practice question: pure .hcl is the only way?

[bstansell23]

So, we've been using terragrunt for a while, find it's features awesome, and use as many bells and whistles as needed to keep our code DRY. We use a deployment structure similar to what's in the terragrunt-infrastructure-live-example repo:

account
└ _global
└ region
└ _global
└ environment
└ resource

terragrunt.hcl files are used to tie things together, etc. We have diverged in one way, however. We are typically not using the terraform and inputs directives to store local terraform code. Instead, we just drop a main.tf (or more) alongside the terragrunt.hcl file and call it a day. Our terragrunt structure uses generate blocks to populate .tf files in that directory to pass information in (we generate provider blocks, etc, but also a locals block to get access to info from terragrunt).

We do realize by putting the terraform code in the terraform block we could eliminate that extra locals file, but with this setup, all but the top-level terragrunt.hcl files are just

include {
  path = find_in_parent_folders("terragrunt-common.hcl")
}

and each leaf directory just looks like a basic/usual terraform directory, but with the added hcl file.

All the examples we've seen documented by the terragrunt team show how you can get down to a terragrunt.hcl file in each deployment directory and have no "raw" .tf files. When we read this, we interpreted this as just one way to keep configs DRY (and when you're just importing a module (terragrunt { source = }}), dropping that simple use into the terragrunt.hcl file makes sense). It certainly makes the examples cleaner.

So, to my question: is it "wrong" to do what we're doing, in a terragrunt best-practice sort of way, by skipping the use of the terraform and inputs blocks and just dropping that config in a .tf file? If our configs are staying DRY, isn't that the real point and best-practice? Or is there an unwritten (as far as we can find) best-practice that terragrunt deployment directories should only contain .hcl files and you should always be using terraform blocks to define your deployments?

To address another aspect of "best practices", we also have our own module / service structure as discussed in #402. So, this discussion is really about the specific use of local .tf files vs embedding things in the terragrunt.hcl file.

Have we inadvertently driven off the "best-practices" road here, or are we just over thinking things? Thanks!

---

Tracked in ticket #109411