- Practices
- Start of Files
- End of Files
- Quotes
- Environment
- Booleans
- Key value pairs
- Sudo
- Hosts Declaration
- Task Declaration
- Include Declaration
- Spacing
- Variable Names
- Jinja Variables
- Comparing
- Delegation
- Playbook File Extension
- Template File Extension
- Vaults
- Role Names
- Debug and Comments
You should follow the Best Practices and Ansible Lint Rules defined by the Ansible documentation when developing playbooks.
The Ansible developers have a good understanding of how the playbooks work and where they look for certain files. Following these practices will avoid a lot of problems.
The playbook examples are inconsistent in style throughout the Ansible documentation; the purpose of this document is to define a consistent style that can be used throughout Ansible playbooks to create robust, readable code.
You should start your playbooks with some comments explaining what the playbook's purpose does (and an example usage, if necessary), followed by ---
with no blank lines around it, then followed by the rest of the playbook.
#bad
- name: Change s1m0n3's status
service:
enabled: true
name: "s1m0n3"
state: "{{ state }}"
become: true
#good
# Example usage: ansible-playbook -e state=started playbook.yml
# This playbook changes the state of s1m0n3 the robot
---
- name: Change s1m0n3's status
service:
enabled: true
name: "s1m0n3"
state: "{{ state }}"
become: true
This makes it easier to quickly find out the purpose/usage of a playbook, either by opening the file or using the head
command.
You should always end your files with a newline.
This is common Unix best practice, and avoids any prompt misalignment when printing files in a terminal.
We always quote strings and prefer double quotes over single quotes. The only time you should use single quotes is when they are nested within double quotes (e.g. Jinja map reference), or when your string requires escaping characters (e.g. using "\n" to represent a newline). If you must write a huge string, we can the "folded scalar" style and omit all special quoting. The only things you should avoid quoting are booleans (e.g. true/false), numbers (e.g. 42), and things referencing the local Ansible environemnt (e.g. boolean logic or names of variables we are assigning values to).
Do NOT quote:
- hosts: targets (e.g. hosts: databases rather than hosts: ‘databases’)
- include_tasks: and include_roles: target file names
- task and role names
- registered variables
- number values
- boolean values
- conditional logic (when: task options)
# bad
- name: "start robot named S1m0n3"
service:
name: s1m0n3
state: started
enabled: true
become: true
# good
- name: start robot named S1m0n3
service:
name: "s1m0n3"
state: "started"
enabled: true
become: true
# single quotes w/ nested double quotes
- name: start all robots
service:
name: "{{ item['robot_name'] }}"
state: "started"
enabled: true
loop: "{{ robots | flatten(levels=1) }}"
become: true
# double quotes to escape characters
- name: print some text on two lines
debug:
msg: "This text is on\ntwo lines"
# folded scalar style
- name: robot infos
debug:
msg: Robot {{ item[robot_name] }} is {{ item[status] }} and in {{ item[az] }}
loop: "{{ robots | flatten(levels=1) }}
# folded scalar when the string has nested quotes already
- name: print some text
debug:
msg: “I haven’t the slightest idea,” said the Hatter.
# don't quote booleans/numbers
- name: download google homepage
get_url:
dest: "/tmp"
timeout: 60
url: "https://google.com"
validate_certs: true
# variables example 1
- name: set a variable
set_fact:
my_var: "test"
# variables example 2
- name: print my_var
debug:
var: my_var
when: ansible_os_family == "Darwin"
# variables example 3
- name: set another variable
set_fact:
my_second_var: "{{ my_var }}"
Even though strings are the default type for YAML, syntax highlighting looks better when explicitly set types. This also helps troubleshoot malformed strings when they should be properly escaped to have the desired effect.
# bad
- name: start sensu-client
service:
name: "sensu-client"
state: "restarted"
enabled: 1
become: "yes"
# good
- name: start sensu-client
service:
name: "sensu-client"
state: "restarted"
enabled: true
become: true
There are many different ways to specify a boolean value in ansible, True/False
, true/false
, yes/no
, 1/0
. While it is cute to see all those options we prefer to stick to one : true/false
. The main reasoning behind this is that Java and JavaScript have similar designations for boolean values.
Use only one space after the colon when designating a key value pair
# bad
- name : start sensu-client
service:
name : "sensu-client"
state : "restarted"
enabled : true
become : true
# good
- name: start sensu-client
service:
name: "sensu-client"
state: "restarted"
enabled: true
become: true
Always use the map syntax, regardless of how many pairs exist in the map.
# bad
- name: create checks directory to make it easier to look at checks vs handlers
file: "path=/etc/sensu/conf.d/checks state=directory mode=0755 owner=sensu group=sensu"
become: true
- name: copy check-memory.json to /etc/sensu/conf.d
copy: "dest=/etc/sensu/conf.d/checks/ src=checks/check-memory.json"
become: true
# good
- name: create checks directory to make it easier to look at checks vs handlers
file:
group: "sensu"
mode: "0755"
owner: "sensu"
path: "/etc/sensu/conf.d/checks"
state: "directory"
become: true
- name: copy check-memory.json to /etc/sensu/conf.d
copy:
dest: "/etc/sensu/conf.d/checks/"
src: "checks/check-memory.json"
become: true
Use the map syntax for roles too.
# bad
roles:
- { role: "tomcat", tags: "tomcat" }
# good
roles:
- role: tomcat
tags: "tomcat"
- role: webapp
code: "mycode"
It's easier to read and it's not hard to do. It reduces changeset collisions for version control.
Use the new become
syntax when designating that a task needs to be run with sudo
privileges
#bad
- name: template client.json to /etc/sensu/conf.d/
template:
dest: "/etc/sensu/conf.d/client.json"
src: "client.json.j2"
sudo: true
# good
- name: template client.json to /etc/sensu/conf.d/
template:
dest: "/etc/sensu/conf.d/client.json"
src: "client.json.j2"
become: true
Using sudo
was deprecated at Ansible version 1.9.1
host
sections should follow this general order:
# host declaration
# host options in alphabetical order
# pre_tasks
# roles
# tasks
# example
- hosts: webservers
remote_user: "centos"
vars:
tomcat_state: "started"
pre_tasks:
- name: set the timezone to America/Boise
lineinfile:
dest: "/etc/environment"
line: "TZ=America/Boise"
state: "present"
become: true
roles:
- role: tomcat
tags: "tomcat"
tasks:
- name: start the tomcat service
service:
name: "tomcat"
state: "{{ tomcat_state }}"
A proper definition for how to order these maps produces consistent and easily readable code.
A task should be defined in such a way that it follows this general order:
# task name
# taks vars
# task map declaration (e.g. service:)
# task parameters in alphabetical order (remember to always use multi-line map syntax)
# loop operators
# task options in alphabetical order (e.g. become, ignore_errors, register)
# example
- name: create some azure instances
vars:
instance_names:
- "az1"
- "az2"
azure_rm_virtualmachine:
admin_username: "azureuser"
name: "{{ item }}"
network_interfaces: "my_network"
ssh_password_enabled: false
vm_size: "Standard_DS1_v2"
loop: "{{ instance_names | flatten(levels=1) }}"
ignore_errors: true
register: ec2_output
when: ansible_os_family == "Darwin"
Similar to the hosts definition, having a well-defined style here helps us create consistent code.
For include
statements, make sure to NOT quote filenames and only use blank lines between include
statements if they are multi-line (e.g. they have vars).
# bad
- include: "other_file.yml"
- include: second_file.yml
- include: third_file.yml tags=third
# good
- include: other_file.yml
- include: second_file.yml
- include: third_file.yml
vars:
myvar: "third"
This tends to be the most readable way to have include
statements in your code.
You should have blank lines between two host blocks, between two task blocks, and between host and include blocks. When indenting, you should use 2 spaces to represent sub-maps, and multi-line maps should start with a -
). For a more in-depth example of how spacing (and other things) should look, consult style.yml.
This produces nice looking code that is easy to read.
Use snake_case
for variable names in your playbooks.
# bad
- name: set some facts
set_fact:
myBoolean: true
myint: 20
MY_STRING: "test"
# good
- name: set some facts
set_fact:
my_boolean: true
my_int: 20
my_string: "test"
Ansible uses snake_case
for module names so it makes sense to extend this convention to variable names.
Use spaces around jinja variable names.
# bad
- name: set some facts
set_fact:
myBoolean: "{{myoldvar}}"
# good
- name: set some facts
set_fact:
myBoolean: "{{ myoldvar }}"
A proper definition for how to create Jinja variables produces consistent and easily readable code.
Don’t compare to literal True/False. Use when: var
rather than when: var == True
(or conversely when: not var
)
Don’t compare to empty string. Use when: var
rather than when: var != ""
(or conversely when: not var
rather than when: var == ""
)
# bad
- name: Validate required variables
fail:
msg: "No value specified for '{{ item }}'"
when: (vars[item] is undefined) or (vars[item] is defined and vars[item] | trim == "")
loop: "{{ appd_required_variables | flatten(levels=1) }}"
- name: Create an user and add to the global group
include_tasks: user.yml
when:
- username is defined
- username != ""
# good
- name: Validate required variables
fail:
msg: "No value specified for '{{ item }}'"
when: (vars[item] is undefined) or (vars[item] is defined and not vars[item] | trim == "")
loop: "{{ appd_required_variables | flatten(levels=1) }}"
- name: Create an user and add to the global group
include_tasks: user.yml
when:
- username is defined
- username
Avoids code complexity sing quotes and standardize the way literals and empty string are used.
Do not use local_action
, use delegate_to: localhost
# bad
- name: Send summary mail
local_action:
module: mail
subject: "Summary Mail"
to: "{{ mail_recipient }}"
body: "{{ mail_body }}"
run_once: true
# good
- name: Send summary mail
mail:
subject: "Summary Mail"
to: "{{ mail_recipient }}"
body: "{{ mail_body }}"
delegate_to: localhost
run_once: true
Avoid complexity, standardization, flexibility and code readability. The module and its parameters are easy to read and can be delegated even to a third party server.
All Ansible Yaml files MUST have a .yml
extension (and NOT .YML, .yaml etc).
# bad
~/tasks.yaml
# good
~/tasks.yml
Ansible tooling (like ansible-galaxy init) create files with a .yml
extension. Also, the Ansible documentation website references files with a .yml
extension several times. Because of this, it is normal in the Ansible community to use a .yml
extension for all Ansible Yaml files.
All Ansible Template files MUST have a .j2 extension.
# bad
~/template.conf
# good
~/template.conf.j2
Ansible Template files will usually have the .j2 extension, which denotes the Jinja2 templating engine used.
All Ansible Vault files MUST have a .vault
extension (and NOT .yml, .YML, .yaml etc).
# bad
~/secrets.yml
# good
~/secrets.vault
It is easier to control unencrypted files automatically for the specific .vault
extension.
All the newly created Ansible roles will follow the name convention using dashes if necessary:
[company]-[action]-[function/technology]
# bad
lvm
# good
mycompany-setup-lvm
If using roles from Ansible Galaxy, it will keep a consistency about which roles are created internally.
Avoid debug and comments in final code as much as possible. Use task and role names to explain what the task or role do. Use verbose option under ansible for debugging purposes.
# bad
- name: print my_var
debug:
var: my_var
when: ansible_os_family == "Darwin"
# bad
# set another variable
- set_fact:
my_second_var: "{{ my_var }}"
# good
- name: set another variable
set_fact:
my_second_var: "{{ my_var }}"
It will keep clean code and consistency avoiding extra debug and comments. Extra debug will spend extra time in when running the playbook or role.