Ansible with_items list based on conditions - ansible

Question

Background: I want to write a ansible playbook that works with different OS. One of the basic thing it does is to install packages using package command (newly introduced in ansible 2.0).
The problem is the list of packages are different on each OS. I would like to specify common ones somewhere, and then place Ubuntu specific ones and CentOS specific ones separately. Then for CentOS ones, CentOS6 and CentOS7 also have some common ones and different ones, so I would like to put common ones for CentOS somewhere, and CentOS6 and CentOS7 specific ones somewhere else.
The immediate way jumps into my mind is to have group_vars, however, the way I know wouldn't work. Suppose under group_vars I have files all, ubuntu, centos, centos6 and centos7 to define variables, the order of which takes precedence is indeterministic. For a CentOS6 host, it is in groups all, centos and centos6. There is no way to specify centos6 should take precedence over centos and centos is over all.
Another way I found was to set hash_behaviour = merge in ansible.cfg, however, this modifies the behavior globally, which overkills. There might be cases where default behavior is wanted, so modifying it globally is not a nice approach.
How should I organize everything in a clean way, and share and re-use common things as much as possible?

Answer 1

OS condition inside the role
One common way that I use is to define OS dedicated default vars files inside the vars directory of the role and then include them at the beginning of the main.yml task. Ansible helps me to automatically identify the currently used distribution with {{ ansible_distribution }} and {{ ansible_distribution_version }}, so I just have to name the OS dedicated yml files accordingly.
Role dir tree:
my_role/
├── handlers
│   ├── main.yml
│   └── Ubuntu14.04.yml
├── tasks
│   ├── main.yml
│   └── Ubuntu14.04.yml
├── templates
│   └── Ubuntu14_04.config.j2
└── vars
└── Ubuntu14.04.yml
In the main.yml you first include default OS specific variables and then run OS specific tasks:
tasks/main.yml:
---
- name: include distribution specific vars
include_vars: "{{ ansible_distribution }}{{ ansible_distribution_version }}.yml"
- name: include distribution specific install
include: "{{ ansible_distribution }}{{ ansible_distribution_version }}.yml"
- name: Do general stuff
...
Now you can easily define your packages inside the vars file for each distribution, e.g., Ubuntu14.04:
vars/Ubuntu14.04.yml:
---
my_role_packages:
- a
- b
And finally install them in your distribution specific setup task.
tasks/Ubuntu14.04.yml:
---
- name: Install packages for Ubuntu14.04
apt:
name={{ item }}
state=present
update_cache=yes
with_items: "{{ my_role_packages }}"
You can now also easily define OS specific handlers and templates, e.g.,:
handlers/main.yml:
---
- name: distribution specific handler
include: "{{ ansible_distribution }}{{ ansible_distribution_version }}.yml"
Hash Merging
About globally setting hash_behaviour=merge, here a quote from the official Ansible documentation:
Some users prefer that variables that are hashes (aka ‘dictionaries’
in Python terms) are merged. This setting is called ‘merge’.
We generally recommend not using this setting unless you think you
have an absolute need for it, and playbooks in the official examples
repos do not use this setting.
Originally I come from the SaltStack world and was used to merging hashes from my defaults map.jinja with the dedicated pillars, but in Ansible I started relying more on variable prefixes, so instead of
nginx:
pkg: "nginx"
repo: "deb http://nginx.org/packages/ubuntu/ trusty nginx"
I'd write
nginx_pkg: "nginx"
nginx_repo: "deb http://nginx.org/packages/ubuntu/ trusty nginx"
in order to avoid accidentally overwriting hashes when going up the variable hierarchy. If in some cases you still prefer merging, you could use the jinja2 combine filter in your variable files: dict_a: {{ dict_b|combine(dict_c) }}.
Variable grouping
The Ansible documentation puts a lot of emphasize on heavily using group_vars and I found that to be good advice. A general approach here is to define my groups in my inventories files, like:
hosts/test:
[app]
app-test-a
app-test-b
[app_test:children]
app
hosts/live:
[app]
app-live-a
app-live-b
[app_live:children]
app
Now I can easily use group_vars to include variables based on the defined groups:
group_vars/
├── all.yml
├── app_test.yml
├── app_live.yml
└── app.yml
Ansible Galaxy and DebObs
I also recommend checking out Ansible Galaxy roles. They are always a good starting point to get some ideas (including this one). Another good source of inspiration is the DebObs repository.