Variable precedence and making roles work across distributions

[Jeff Geerling]

The technique I usually use for making a role work cross-distro is to do two things:

For variables: Include a variables file named {{ ansible_os_family }}.yml (then stick 'Debian.yml' and 'RedHat.yml', etc. into the role's vars folder.

For tasks: Add "when: ansible_os_family == 'Distribution'" to the task (where 'Distribution' is Debian, RedHat, etc.).

This presents an issue; for normal variables that I want to let role users override easily, I put the variables in the roles 'defaults/main.yml' file. But for distribution-specific variables (for example, "java_packages" in my Java role), I have to put the distro-specific vars include files in the main 'vars' folder, so those variables have higher precedence than, say, group_vars.

I've bumped into this a few times, but could usually find other ways of getting the variable I want to take hold. For some users, it's not possible, thus I have bugs like: Openjdk-7 cannot be installed.

What are some possible fixes/other ways to do this?

I could set up a variable for each distro (like "java_redhat_packages" and "java_debian_packages", but in some roles, where 5-10 of these would be needed, I fear maintainability and simplicity would be sacrificed for variable overridability...

[Maciej Delmanowski]

How about this solution?

In `defaults/main.yml`:

    java_distribution_packages:

      Debian: [ 'openjdk-7', 'java-stuff' ]

      RefHat: [ 'java-runtime-7', 'other-stuff' ]

And then, in `tasks/main.yml`:

    - {{ ansible_pkg_mgr }}: name={{ item }} state=latest

      witn_items: java_distribution_packages[ansible_os_family]

I think that should work fine and let you override java_distribution_packages in inventory.

Maciej

--
You received this message because you are subscribed to the Google Groups "Ansible Project" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ansible-proje...@googlegroups.com.
To post to this group, send email to ansible...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ansible-project/d8e90023-a27d-4e71-962e-e8919aadfa83%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Michael DeHaan]

Never do this, ansible_pkg_mgr does not allow the with_items block to be grouped into a single yum/apt transaction, which results in much slower package installations.

I understand why it's appealing, but really don't ... you'll also find that the yum and apt modules have different options you may want to use from time to time anyway.

To view this discussion on the web visit https://groups.google.com/d/msgid/ansible-project/CAEnKK1yksi6J2QNjDYXbh1iwMwSHgxx1nLjKmOEpEVY2oic78w%40mail.gmail.com.

[Maciej Delmanowski]

Will this work ok if I use apt: module directly instead? For different package lists in Debian and Ubuntu, for example.

To view this discussion on the web visit https://groups.google.com/d/msgid/ansible-project/CA%2BnsWgw0pa_HG2p%3DCjJWd1Q-iWy%3D6UDfdoiFRJNnd1hHw2aRHg%40mail.gmail.com.

[James Cammarata]

You can achieve pretty much the same thing by using group_by or conditional includes along with the variable structure you've defined, without loosing the benefits Michael mentioned.

It should work fine with the apt or yum (or other package modules).

To view this discussion on the web visit https://groups.google.com/d/msgid/ansible-project/CAEnKK1zX3KMkvEVxFHCpNe_UPsU%2BbYfuPAZVYstWDDb-kF77Hw%40mail.gmail.com.

[Michael DeHaan]

group_by is awesome.

- hosts: webservers

  group_by: key=webservers_{{ ansible_os_family }}

- hosts: webservers_RedHat

  roles:

     - ...

To view this discussion on the web visit https://groups.google.com/d/msgid/ansible-project/CAMFyvFhDTnsxevhBVzJx3zv41Lc1GCLw371TcxfHJVPgJd1UrQ%40mail.gmail.com.