yourlabs.remember

Meta-Role for Ansible Automation Architects.

This role provides the magic to smarten Ansible role workflows:

massive speedup: require a dependency role Once And Only Once
inventoryless dry: remember variables injected from the CLI on the host
CLI variable prompt: interactive questioning of the user for facts

Note

This role does not automatically download dependency roles: that's the job of bigsudo command.

Demo

The easiest way to try it out:

pip install --user bigsudo
~/.local/bin/bigsudo yourlabs.fqdn user@somehost
# Or if you feel brave (skip hostname to apply on localhost)
~/.local/bin/bigsudo yourlabs.traefik

Of course you could also use ansible commands, but then it would be more commands and options. We're getting inspiration from the practice of kubectl, for little servers, non-HA services, and pizza teams. Even if, I would personnaly still use bigsudo yourlabs.k8s to configure k8s instances if i had to ...

Usage

This role allows you to define what variables are needed in your own role, along with things such as the description that will be displayed to the user, defaults, regexp validation and so on.

OAOO role dependency injection

I'll take the example of what happens with yourlabs.traefik (a docker based load balancer) that requires yourlabs.docker which in turn just installs docker.

However, for the sake of the example, i will use your.parent and your.child to represent the use case of yourlabs.docker and yourlabs.traefik respectively.

In your.child/requirements.yml:

- your.parent

In your.parent/requirements.yml:

- yourlabs.remember

As such, your.child depends on your.parent, and your.parent depends on yourlabs.remember.

Note

bigsudo transparently ensures that requirements are installed recursively when you run bigsudo your.child.

Without the remember role, you would normally include the your.parent role as such at the top of your.child/tasks/main.yml:

- name: Install your.parent prior to running our tasks
  include_role: name=your.parent

However, this will play the role everytime, making execution longer. If you don't want to wait for your.parent to fully execute everytime when you execute your.child, you can transform the above task as such at the top of your.child/tasks/main.yml:

- name: Install your.parent if never done on this host
  include_role: name=your.parent
  when: ansible_facts['ansible_local']['your_parent']['state']|default('') != 'success'

For this to work, you will need to add the following at the end of your.parent/tasks/main.yml:

- include_role: name=yourlabs.remember tasks_from=success

As such, running bigsudo your.parent (also works with ansible) will create /etc/ansible/facts.d/your_parent.fact with such content:

#!/bin/sh
echo '{
   "state": "success"
}'

This is how you can skip including the role next time.

Read on to add your custom persistent role variables with interactive configuration.

Interactive role configuration

In your.parent/vars/main.yml, define remember_fact that is the namespaces for this role deployment variable as well as the variables your role depends on as such:

---
remember_fact: your_parent
remember:
- name: email_enable
  question: Enable a custom email ?
  default: false
  type: bool
- name: email
  question: What email to use ?
  type: email
  default: '{{ lookup("env", "USER") }}@{{ inventory_hostname }}'
  when: email_enable

Then, in your.parent/tasks/main.yml, you can include yourlabs.remember and it will load up the variables and ask user for new variables interactively, pretty fast thanks to the Action Plugin:

- include_role: name=yourlabs.remember

You can do more, refer to the test.yml playbook of course, which i run with ansible-playbook -c local -i localhost, test.yml -v --become:

Multiple deployments: namespacing variables

To enable multiple deployments of a role on the same host, ie. to enable eXtreme DevOps you will need your remember_fact to depend on a variable.

For example, you want to deploy a docker-compose into different directories on your host. As such, you will require a home variable:

remember:
- name: home
  question: What home dir to deploy to ? (must start with /home for the regexp example)
  default: /home/test
  type: path
  regexp: /home.*

That means that if the user doesn't pass an home variable on the command line (ie. with -e home=/home/bar) it will prompt for the home directory. Now, all we have to do is re-use that home variable into the remember_fact so that it will namespace variables per home directory:

.. code-black:: yaml

    remember_fact: your_role_j2((home))

As you can see, we use j2(( )) instead of {{ }}, this is to prevent Ansible from rendering this before getting a value for the home variable. In fact, the remember action plugin will:

try to render remember_fact to load existing variables if any,
catch AnsibleUndefinedVariable exceptions,
find the definitions for the undefined variables it needs in remember,
ask for them without saving
load the existing variables
and ask for any new variables

Conclusion

Finnaly we're getting to the point where we have a clear and relatively easy way to:

dynamically inject dependency roles to speed up subsequent executions of a role, effectively preventing un-necessary double execution of dependency roles (such as docker, load balancers, lower level automation ...)
suppress the inventory because each server keeps its variables, it's also DRY by the way, so that's still one repo less you will have to worry about !
interactive fact prompt no more need to read the docs before executing a role you found on internet as root !

Credits

Thanks totakoko from beta.gouv.fr for the long discussions and for demonstrating that my inventory was overkill and that it was possible without ;)

Thanks to #ansible``@``irc.freenode.net, on of the best IRC channels, namely:

agaffney
mackerman
jborean93

And thank you for reading my little adventure !

yourlabs/yourlabs.remember