Variable inheritance/override

[Duncan Hutty]

I'd like to have a pattern for roles where I can:

1) set default values in defaults/default.yml
2) override certain parts of a deep data structure in
defaults/{{ansible_distribution}}.yml
3) allow individual instantiations of that role to override specific
parts of it.


I have a complex data structure for configuring the settings of a
certain software.

A more or less complete yaml description of the data structure might be
something like:
--
foo_app:
config_dir: /etc/foo
run_dir: /var/cache/foo
pid_file: /var/run/foo/foo.pid
user: foo-user
group: foo-group
interface: *
port: 8888
unix_socket: /var/run/foo.socket
settings:
cluster_members:
- hostA
- hostB
- hostC
cache_size: 8192
client_timeout: 8s
cluster_timeout: 1s
...

In other words, a deep structure which is a dict, where the values of
some of the (top-level) keys could be various datatypes, including lists
or dicts themselves.

I'd like to have templates which use this data structure look like:

{% for key, value in foo_app.settings.iteritems() | sort %}
{{ key }} = {{ value }}
{% endfor %}

partly because it's convenient, and partly because namespacing under
foo_app is good defensive programming to reduce the chance that other
roles will step on my vars.

So in summary, what do we think is the best way to allow (1), (2) & (3)
without

a) requiring that the data structure is flattened into individual vars:

--
foo_app_config_dir: /etc/foo
foo_app_settings_clustermembers:
- hostA
- hostB
- hostC

OR

b) Ensuring that *every* declaration of the data structure contains
*every* element of the data structure, not just the ones we want to
override.

Of course, I realise that I might just be going about this all wrong (or
at least not Anisbly) - if that's the case, please feel free to point me
at docs (or Galaxy roles or github repos for learning by example) that
show the Right Way to approach roles with elegance with regards to
multi-distribution support and other cases that call for
inheritance/override flexibility.
--
Duncan Hutty
http://www.allgoodbits.org

[Jonathan Davila]

Here is a slightly odd technique that you could possibly use.

The first step would be to make your 'role/defaults' data structure. In my example it looks like this:

car:
  make: "{{ makevar }}"
  model: "{{ modelvar }}"
  year: "{{ yearvar }}"
  condition:
    paint: "{{ paintvar }}"
    frame: "{{ framvar }}"

Notice that I am parameterizing all the values and the default is merely functioning as the skeleton.

Now in 'group_vars/all' I wrote this in:

makevar: Ford
modelvar: Mustang
yearvar: 2009
paintvar: Great
framvar: Excellent

This establishes the default values for the initial data structure.

Now for the purposes of my example I made a simple role with the following task:

- name: Show this value
  debug: var=car.model


- name: show this other value
  debug: var=car.make


- name: this deeper value
  debug: var=car.condition.paint

when I run the playbook, it renders the values correctly. Now of course you could flip the group_vars/all with somerole/defaults and it would work the same, it would come down to matter of preference, your needs in terms of variable precedence and how you plan to structure the rest of your Ansible setup. 

Let's assume the role is called 'coolrole' if you need to override specific parts in various instantiations of the role you would do something like the following:

---
- name: Spin up instances for testing
  connection: local
  hosts: localhost
  roles:
    - { role: coolrole, modelvar: GT }

Notice the modelvar variable is being overwritten for this particular instantiation of the role.

I think this should help you, but let me know if something didn't makes sense.

Cheers!