Best Practice Question: Multiple non-overlapping groups for one host - best alternative?

Archives Ansible Project
1

Hi all,

in the setup I would like to automate with ansible I have mostly servers with
distinct purposes and on each server they may be several distinct
services/components. The plan is to use ansible to have the configuration of
all these servers in one place and to enable us to setup staging environments
for the new software we develop and for new releases of the software stack we
use (mariadb, nginx, ...)

Since we want to be able to rearrange the "services" (i.e., a database or a
FastCGI application) independently to different hosts I started to organize
everything into ansible groups, i.e., I have groups like

- database_01
- database_02
- ...
- fcgi_01
- fcgi_02
- ...

for each group I set variables in group_vars/<group-name>, which looks a
little bit like the following:

group_vars/database_01::

2

Hi David, the best way to do this would be to make the role aware of the fact that there may be more than one database installed on it, so that the variables in the group would then become an array:

group_vars/foo.yml:

databases:

  • name: db01
    user: db_user1
    password: xxxxxx
  • name: db02
    user: db_user2
    password: xxxxxx
    …

And then you would loop over this list in your role tasks.

Hope that helps!

3

Hi James,

thanks for the tip. I think the approach is not totally suitable for my
situation. I would rather not define all the databases for one server in one
place for two reasons:

I would like to have it very easy to "move" the database to a different server
(only in the sense that it gets installed on a different server if I run
everything from scratch to install a test/staging environment)

Also quite important is that I want to be able to run a playbook that installs
just db01 and not db02 to save development time (e.g., when I change the
database schema of db01, but do not want to run the tasks that check if there
needs to be updates to db02 or when I want to setup a new test environment
where I know that I do not need db02...).

Maybe there is a way to setup things that I have to compromise at most on one
of these things?

4

define all DBs in group_vars/all

then have 3 groups:

[singledbA]
mydbs = dbA

[singledbB]
mydbs = dbA, dbB

[bothddbs]
mydbs = dbA,dbB

then have the task that installs them use use the mydbs
you can even define mydbs by host