using Ansible properly with database inventory

Archives Ansible Project
1

I have a couple of probably simple questions about using Ansible properly.

Background: Its not practical in our environment to install every software stack component needed by individual applications on all relevant hosts, so, rather than classify our hosts according to their detailed software profiles, we remember the bits necessary to support each application in a database. I.e. we have an application-centric inventory.

These specialized software components are installed by ansible over our standard stacks and maintained in ansible roles. Ansible role names, as read from the standard ansible directory structure, are attributes of apps in the inventory database

At the cost of tying our ansible directory structure to the inventory, this allows us to keep track of provisioned bits and, as well, more fully and automatically configure application specific bits during provisioning, For example, create Apache’s vhost config files from templates with all app-specific details gleaned from the database.

Problem: The list of ROLES that accomplish app provisioning comes from the database (by an inv.py script like the one blogged by Jan-Piet Mens) via the ansible environment and courtesy of THIS PATCH to ansible’s play.py:

66, 72
#self.roles = ds.get(‘roles’, None)
try:
self.roles = (self.playbook.inventory.get_group_variables(os.environ[‘APPGRP’]))[‘roles’]
except:
self.roles = ds.get(‘roles’, None)

(The try / except clause is used in the hope that our change is transparent, but that has not been tested.)

I suspect that this change breaks Ansible’s intended model, or else it would not have been necessary. It appears that the modded code is in a pretty stable part of ansible, or at least I hope it is. Does anyone have any dire warnings or other advice? Is this a serious breakage?

thanks,
larry

2

You should not patch ansible as you’ll drift from mainline pretty steadily and the bugs you file will be invalid.

an easier solution here would be to either:

(A) just generate the playbooks mapping groups to hosts
(B) use conditionals for the hosts like an “apps” array, and do the following to avoid

  • hosts: all
    tasks:

  • group_by: key=webservers_{{ ‘webserver’ in installed_apps }}

  • group_by: key=dbbservers_{{ ‘dbbserver’ in installed_apps }}

  • hosts: webservers_True
    roles:

  • webserver

  • hosts: dbservers_True
    roles:

  • dbserver

(C) Ultimately, it’s easier if you just have a static playbook and just return the hosts that should be in each role based group.

3

Thanks for the prompt reply. If I understand correctly, we need a one-to-one mapping between (sets of) roles and groups, which we don’t currently have