Puppet

Sure. I’ll try to be brief but my use-case is a little involved.

I often use custom definitions to encapsulate functionality within modules, but still allow multiple modules to interact with each other. I have a postfix module that manages Postfix on all my hosts — MTAs, final-delivery mailhosts, as well as null clients. I also have a user module which handles all my user accounts.

I want my human & system user accounts to be treated differently by Postfix — for example, on a null client nc.stanford.edu I want to send email as icomfort@stanford.edu, but I want system mail to come from, e.g., root@nc.stanford.edu and apache@nc.stanford.edu. I effect this by leaving $myorigin alone and adding a canonical mapping for icomfort –> icomfort@stanford.edu. On a mail gateway, however, I want a virtual mapping that sends icomfort’s mail to my preferred mailhost; say icomfort@stanford.edu –> icomfort@mh.stanford.edu.

My user module passes information to the postfix module about which users are human using a custom definition, postfix::human. This is pseudo-code but it captures the design well enough.

$mail_domain = 'stanford.edu'
$user_mail_host = 'mh.stanford.edu'

define user::human {
  user { $name: }
  postfix::human { $name: }
}

define postfix::human {
  @postfix::canonical { $name:
    address => "$name@$mail_domain",
  }
  @postfix::virtual { "$name@$mail_domain":
    address => "$name@$user_mail_host",
  }
}

class postfix::nullclient {
  Postfix::Canonical<| tag == 'postfix::human' |>
}

class postfix::gateway {
  Postfix::Virtual<| tag == 'postfix::human' |>
}

user::human { 'icomfort': }

I can’t realize all canonical or (Postfix-)virtual mappings in my postfix subclasses because I use (Puppet-)virtual mapping resources for other functionality, as well.

Yep, that’s what I’ve been doing as a workaround.