Configure Ansible roles with dependent roles - ansible

Question

The problem is best described with an example:
There are two roles:
mailserver: a basic mail server configuration
mailinglist: mailing list application
The mailing list software needs the mailserver to transport incoming mails to the mailing list software's "virtual inbox". This requires some configuration of the mail server. But the mailserver does not know about the mailing list role, nor other roles with similar configuration requirements.
What I would like to do is this:
mailinglist (and other similar roles) stores the transport configuration in a variable transport_config. This could be a "transport map" like $email => $spool.
mailinglist depends on the mailserver role.
mailserver configures it's "transport" using the variable transport_config.
Is there a way to do something like this in Ansible? Or another solution to this problem? It's not possible to use role variables like {role: mailserver, transport_config: ...}, as there may be more than one role depending on the mailserver.
What I can think of is a workaround: The mailserver reads/parses a configuration directory where transport maps are defined. mailinglist and other roles add files to this directory. The problem here is that this often requires a "configuration builder" which reads such configuration directories and generates the main configuration file.

Answer 1

You can accomplish this using role dependencies.
In the mailinglist role under roles/mailinglist/meta/main.yml, add something like this:
---
dependencies:
- { role: mailserver, transport_config: ... }
Do the same for any other similar roles.

Answer 2

In response to your comment about a "configuration builder", see the assemble ansible module. The tricky part might be getting all the files into one place, before you run the assemble module.
Otherwise, tima's suggestion of host_ and group_vars makes sense.

Answer 3

I know this is long answered, but I only just found a workable solution, and it's a little sneaky.
I use a third settings role, which has no tasks, only variables in a defaults/main.yml. The docs are a bit vague on this, but values here get rippled to all dependent roles, so if both roles depend on settings through their meta/main.yml files, both get a common set of values. These are overridable in the usual ways, through a group_vars file.
The surprise to me was that since, yes, the settings role is used more than once, it doesn't matter because there are no tasks in it, and that data can flow from there up the chain of dependencies.
It's an entirely new form of data flow in Ansible that I didn't know was possible.

Answer 4

Consider managing the mailserver configuration file with something like dotdee:
Original Post by Dustin Kirkland describing the concept
Dustin's dotdee repository
A simpler implementation with less features in Python
With dotdee, you assemble your final configuration file from a series of files placed in a .d directory.
In your case, the mailinglist role and others depending on mailserver would drop configuration snippets in a .d directory (created by the mailserver role) and run a command to update the mailserver configuration file like dotdee --update /path/to/mailserver/conf.

Answer 5

I'm not familiar enough with your apps here to be sure, but I'm thinking that you may be better served using host or group variables to pass your configuration in to your roles. The roles can hold your defaults et al and expect to retrieve host/group variables values that you set on an individual basis. This may mean treating each app instance as a host rather than psychical host. How you model it depends on many finer points in your workflow, configuration and app.
This recent thread touches on some of this sort of thing: https://groups.google.com/forum/#!topic/ansible-project/yrnsx2Mw6rc