Is this really difficult to do? I'm very interested in fixing this but not at all sure where to start.

https://wikitech.wikimedia.org/wiki/Puppet_Hiera#Role-based_lookup

It's that "the new parser function/keyword, called role" is something that we (Joe) made ourself and doesn't come from upstream. The question is why that parser function isn't also in labs.

The role keyword is used in production to refer to large groups of hosts; we DEFINITELY don't want to have role lookups in labs for the same values we use in production; also, we can't use the "role" keyword in labs as it uses ldap for defining nodes and that function is thought for use at the node scope.

I am honestly unsure when and how this can be useful, and the examples I see cited are just cases in which people forgot to add hiera lookups for labs, which will happen with role-based lookups as well.

You can mimic easily this functionality by adding to the labs hiera something like

- role/%{::_role}

and then adding

$role = abcd

in the ldap definiton of a host.

But, honestly, what would this solve? Remember every new hiera lookup adds complexity where we'd like to reduce it

What this would concretely solve (for me) is projects like tools, where now I have to manually set hiera config on each host of a particular type (/me looks at https://wikitech.wikimedia.org/wiki/Hiera:Tools/host/tools-k8s-etcd-01, https://wikitech.wikimedia.org/wiki/Hiera:Tools/host/tools-k8s-etcd-02, https://wikitech.wikimedia.org/wiki/Hiera:Tools/host/tools-k8s-etcd-03, etc - tons of these). Remember labs isn't just used to test production changes - some of us live here :)

We're also trying to kill all LDAP variables, see T101447

I'm no longer convinced we have to do this. https://phabricator.wikimedia.org/T91990 should cover most of the things we want from this.