Variables and Facts

Scope: how one playbook configures different machines differently — values you set (variables) and values Ansible discovers (facts) — plus where to put them without making a mess.

Variables you set

You've already met three kinds:

• Inventory group vars — in examples/inventory.ini: [webservers:vars] app_port=8080. Every webserver gets it.
• Inventory host vars — after a hostname on its own line: web1 app_port=9090 would override the group for web1 alone.
• Playbook vars: — the play-level block in site.yml (app_dir). Good for values intrinsic to the playbook itself.

And one built-in, magic variables: inventory_hostname (the host's inventory name), groups (all groups and their members), hostvars (another host's variables — how a load balancer learns the web servers' addresses).

Where files beat inventory lines

Group/host vars outgrow INI files fast. The grown-up layout keeps them in YAML files Ansible finds by name:

group_vars/webservers.yml     # applies to the group
host_vars/web1.yml            # applies to one host

# group_vars/webservers.yml
app_port: 8080
env_name: production

Same effect as the INI block, but with room for lists, nested data, and comments. The examples use the INI form so everything fits on one screen; your real project should use the files.

Facts: variables Ansible discovers

At the start of every play, Ansible runs Gathering Facts (the task you saw in the playbook output) — the setup module interrogating each machine. Results land in ansible_facts:

"ansible_facts": {
    "ansible_distribution": "Ubuntu",
    "ansible_distribution_version": "24.04",
    ...
}

Use them anywhere a variable fits:

- name: Say where I am
  ansible.builtin.debug:
    msg: "{{ inventory_hostname }} runs {{ ansible_facts['distribution'] }} {{ ansible_facts['distribution_version'] }}"

Facts are how a single playbook installs with apt on Debian-family machines and dnf on RedHat-family ones, sizes a buffer from installed RAM, or skips a task on the one ancient box. Two practical notes:

• Facts cost time (a second or two per host). Playbooks that don't need them can say gather_facts: false — the demo handler in this guide uses a fact (date_time), which is why site.yml leaves gathering on.
• ansible <host> -m setup dumps everything a machine offers. Grep it before assuming a fact's name.

Precedence, in plain terms

Ansible has a full variable-precedence ladder (20+ levels). The mortal version:

1. Command line (-e extra=vars) beats everything.
2. Host vars beat group vars; a more specific group beats a broader one.
3. Playbook vars: beat inventory defaults; role defaults lose to nearly everything (by design — they're defaults).

When a value surprises you, it is almost always an override you forgot existed. ansible-inventory --host web1 prints a host's fully-merged variables — check there before suspecting Ansible of creativity.