Ansible is an open source configuration management tool written in Python and helps automate software configuration and deployment. It manages remote hosts via SSH or PowerShell and requires no agents on the client side. Changes can be executed on one or more hosts which can be on different networks, so long as it is accessible by the Ansible host.

Introduction[edit | edit source]

Ansible provides a bunch of modules that perform a certain action. For example, the package module installs a package on the host machine. Each module can take any number of parameters which are specified by a task definition. We could define a new task that passes 'vim' as a parameter to the package module so that vim gets installed. A set of roles that together implements a specific feature or sets up the 'role' of a machine can be bundled together as an Ansible role. A playbook specifies which host or host groups implement which roles.

In summary:

  • Inventory: file or script that tells Ansible what the hosts are. Hosts can be bundled together into groups. Groups can inherit other groups.
  • Playbook: A playbook tells Ansible what tasks or roles a host or host group need to implement
  • Role: A bundle of tasks, files, and templates that configures a specific feature or role.
  • Task: A list of parameters which are called against a specific Module in order to do a specific action
  • Module: A piece of code which does an action

Modules & Tasks[edit | edit source]

Modules can take in any number of parameters as variables. Modules can also return data back to Ansible which can be used later on by other modules. A module by itself isn't able to do anything and can only 'run' when used as a task. A Task is simply a list item that specifies a module and its parameters.

A single module can be executed via the ansible command while multiple modules can be executed through the use of a playbook which describes the list of tasks or a bundle of tasks (called roles) it needs to perform.

Ansible uses the Jinja2 template engine and all parameters that are passed to Ansible can be templated. Uses of this are found throughout the examples below and allows for the use of loops (by iterating over lists) and embedding variables within strings. Templated values must be enclosed in double quotes.

Modules which are commonly used and their commonly used parameters are listed below:

Module parameter Description
yum/dnf
- name: Install NTP
   yum:
     name: ntp
     state: present

 - name: Install common packages
   dnf:
     name:
     - git
     - vim
     - zsh
Installs a package.

You may also use loops to call the module multiple times for each item. Eg:

- name: Install packages
  dnf:
    name: "{{ item }}"
  with_items:
  - git
  - vim

This is something which isn't required for the yum/dnf modules since version 2.0.

package
- name: Ensure NFS utils are installed
  package:
    name: "{{ __packages.client }}"
    state: "{{ 'latest' if upgrade else 'present' }}"
This module will use the proper underlying package manager suitable on the system, making cross-platform tasks easier to write.

The name parameter also accepts a list, like yum/dnf.

template
- name: Configure NTP
  notify: restart ntp
  template:
    src: ntp.conf.j2
    dest: /etc/ntp.conf
Copy a templated file.

Notice that we also call 'restart ntp' when the file is changed. 'restart ntp' is a task defined under the handlers/main.yml file.

Template may include variables passed to this playbook.

service
- name: Start the ntp service
  service: name: ntpd
    state: started
    enabled: yes

- name: restart ntp
  service:
    name: ntpd
    state: restarted
Ensures something is started and enabled on startup
debug
- name: "testing something"
  debug:
    msg: "item: {{ item }}"
  with_items:
    - 1
    - 2
Debug messages
command
- name: Extract archive
  command: /bin/tar xvf wordpress.tar.gz
  args:
    chdir: /srv/ 
    creates: /srv/wordpress
	
- name: safely use templated variable to run command
  command: cat {{ myfile|quote }}
  register: myoutput
Runs a command
shell
- name: Shell command
  shell: "echo $HOME | grep something"
Similar to command, but the task is executed inside a shell. Environment variables and piping will work as a result.

The creates argument can be used so that the command only runs when the file does not exist.

copy
- name: Copy using inline content
  copy:
    content: |
      this is some file content
      of mine.
    dest: /tmp/file
- name: Copy a "sudoers" file on the remote machine for editing
  copy:
    src: /etc/sudoers
    dest: /etc/sudoers.edit
    remote_src: yes
    validate: /usr/sbin/visudo -csf %s
Copies files or sets file contents
file
- name: Create MariaDB log file
  file:
    path: /var/log/mysqld.log
    state: touch
    owner: mysql
    group: mysql
    mode: 0775

- name: Create MariaDB PID directory
  file:
    path: /var/run/mysqld
    state: directory
    owner: mysql
    group: mysql
    mode: 0775
Creates a file or directory
stat
- name: Check that the somefile.conf exists
  stat:
    path: /etc/file.txt
  register: stat_result
  changed_when: False

- name: Create the file, if it doesnt exist already
  file:
    path: /etc/file.txt
    state: touch
  when: not stat_result.stat.exists
Stats a file, which can be used to only run something based on a file's existence.
find
- name: Find files with globbing
  find:
    paths: /tmp
    patterns: "*.swp"
  register: list_of_files
  
## Loop over list_of_files using with_items.
- name: Do something with file list
  file:
    path: "{{ item.path }}"
    state: absent
  with_items: "{{ list_of_files.files }}"
finds files by name or by pattern.
mount
- name: Mount /home
  mount:
    fstype: nfs
    opts: "vers=3,intr,hard"
    path: /home
    src: "netapp{{ fileserver_index }}:/ArcHome"
    state: present
Mounts a filesystem
firewalld
- name: insert firewalld rule
  firewalld:
    port: {{ mysql_port }}/tcp
    permanent: true
    state: enabled
    immediate: yes
Manages the firewall
git
- name: Copy the code from repository
  git:
    repo: {{ repository }}
    dest: /var/www/html/
You can clone an entire git repository to some location.
systemd
- name: Make sure a service is running
  systemd:
    state: started
    name: httpd

- name: Enable a timer for dnf-automatic
  systemd:
    name: dnf-automatic.timer
    state: started
    enabled: yes
Manages systemd services and timers.

State must be one of:

  • reloaded
  • restarted
  • started
  • stopped

Roles[edit | edit source]

Roles define a set of tasks that should be done. I like to think of Roles as a bundle or grouping of tasks. Ansible Galaxy contains a bunch of other roles that can be used. GitHub also has lots of other roles users have made public which may do what you want.

Roles help organize tasks into a single unit. For example, you may want to create a webserver role that contains instructions on setting up Apache and another database role with instructions on setting up MySQL. Creating a new role involves creating the following directory structure:

roles/
   installapache/
     tasks/
     handlers/
     templates/
     vars/
     defaults/
   installtomcat/
     tasks/
     meta/

Each role should be its own subdirectory in the roles directory. Each role should also have folders for each of: tasks for all ansible tasks the role should perform, handlers containing all handlers in this role, files for files needed by the role, such as configuration files, templates containing templating data, vars containing values that override defaults set in the defaults directory, meta for any metadata needed for this role, such as dependency data. In each of these directories, the main.yml file will be read and used if it exists.

In the example above, we have a installapache role and installtomcat role.

Inside installapache/tasks/main.yml, we can load the appropriate set of tasks depending on the distribution:

- name: import a tasks based on OS platform 
  import_tasks: centos.yml 
  when: ansible_distribution == 'CentOS' 
- import_tasks: ubuntu.yml 
  when: ansible_distribution == 'Ubuntu'

We would then create a centos.yml and ubuntu.yml within this same tasks directory with all the tasks that should be done in this role. Eg.

---
- name: Install Apache using yum
  yum:
    name: "httpd"
    state: latest
- name: Start the Apache server
  service:
    name: httpd
    state: started

As Dependencies[edit | edit source]

Roles can be included as dependencies. Under meta, define something like:

dependencies:
  - role: common
    vars:
      some_parameter: 3
  - role: apache
    vars:
      apache_port: 80
  - role: postgres
    vars:
      dbname: blarg
      other_parameter: 12


Inventory[edit | edit source]

An Inventory contains a list of hosts managed by Ansible. It outlines any groups and their relationship with hosts. This file can be written in either INI or YAML. By default, the inventory file is loaded from /etc/ansible/hosts, but can be specified with -i when calling ansible or ansible-playbook.

An example INI inventory file is shown below. Note that groups are defined as named INI sections. Group inheritance and variables are specified with the :children and :vars in the INI section.

[webservers]
web1.example.com ansible_user=bob
web2.example.com ansible_host=192.168.81.142 ansible_port=3333

; variables that apply only to the webservers group
[webservers:vars]
listen_port=8080

[apservers]
ap1.example.com
ap2.example.com

[dbservers]
; db01, db02, ..., db10
db[01:10].example.com

; servers group contains the following sub-groups
[servers:children]
webservers
apservers
dbservers

Here, we have four distinct groups: webservers, apservers, dbservers, and servers. The servers group includes every host that's defined here since it includes all the other groups as its children. Additionally, there is a built-in group named all which can be used.

Host ranges can be specified with brackets, such as [01:10] to denote nodes from 01 (with a leading 0) through 10.

Variables[edit | edit source]

Both group and host specific variables can also be defined outside of the inventory file by placing them within a group_vars and host_vars directory, relative to the inventory file. This might be desirable if there are many variables that need to be defined and allows for easier organization. Regardless of what format your inventory file is written in (INI or YAML), the variables defined in these directories must be written in YAML or JSON.

For example, we can have the following directory structure:

ansible_root
├── group_vars
│   ├── all
│   │   └── general.yaml
│   ├── servers
│   │   └── only-servers.yaml
│   └── webservers
│       └── web-servers.yaml
├── host_vars
│   └── ap1.example.com
└── inventory.ini

Hosts will load variables in the following order:

  1. The playbook, the inventory file,
  2. the group_vars directories (from most general group all, through each of the intermediate parent groups to reach the most specific child group), and
  3. the host_vars directory.

The values that are loaded are then merged or flattened for each host before the playbook is executed. The rule of thumb is that the most specific takes precedence. In the case of host ap1.example.com, it would load its values from: the playbook, the inventory file, group_vars/all, group_vars/servers, host_vars/ap1.example.com. Each subsequent file overrides any previous variables that was set, allowing for generic values to be applied by placing them in the more generic locations first.

Read more on this at: https://docs.ansible.com/ansible/latest/user_guide/intro_inventory.html

Playbook[edit | edit source]

Playbooks define what tasks (or roles which is just a bundle of tasks) a host or host group Ansible needs to perform. When running a playbook with ansible-playbook, tasks are executed in the order they appear and in the order of the host pattern that is given, ensuring that runs are repeatable and reproduceable.

Playbooks can be applied with the ansible-playbook command. After a playbook run, a recap will output the following:

  • ok - task ran successfully with no changes being made
  • changed - task ran successfully with changes being made
  • failed - task failed to run
  • unreachable - host was unreachable
  • skipped - task was skipped
  • ignored - task was ignored
  • rescued - ?

An example playbook looks like this:

---
- name: Handler demo 1
  hosts: frt01.example.com
  gather_facts: no
  become: yes

  tasks:
    - name: Update Apache configuration
      template:
        src: template.j2
        dest: /etc/httpd/httpd.conf
      notify: Restart Apache

    - name: restart all services
      command: echo "this task will restart all services"
      notify: "restart all services"

  handlers:
    - name: Restart Apache
      service:
        name: httpd
        state: restarted
      listen: "restart all services"

Handlers can be called with notify. Handlers can also listen for a specific event with the listen directive.

Facts[edit | edit source]

You can get all facts by running:

# ansible all -m setup -i inventory.ini

You can determine if the environment is a VM or container by reading the value set under ansible_virtualization_type.

When defining tasks, you can conditionally run something with the when parameter followed by a truth statement. For example, to only run on Red Hat based systems, add when: "ansible_facts['os_family'] == 'RedHat'"

Loops[edit | edit source]

More about loops https://docs.ansible.com/ansible/latest/user_guide/playbooks_loops.html#playbooks-loops

Loops can be done by specifying a list of items and passing it to loop, or with_items.

---
- name: Install Apache
  hosts: frt01.example.com
  gather_facts: no
  become: yes

  tasks:
    - name: Install Apache package
      yum:
        name: httpd
        state: latest
    - name: Open firewall for Apache
      firewalld:
        service: "[[:Template:Item]]"
        permanent: yes
        state: enabled
        immediate: yes
      loop:
        - "http"
        - "https"
    - name: Restart and enable the service
      service:
        name: httpd
        state: restarted
        enabled: yes


Quick Demo[edit | edit source]

The quickest way to get your toes wet with Ansible is to try it out with a small set of Docker containers. Using the two docker images (TODO: link to docker files) and docker-compose, we can bring up a small lab to test Ansible with (TODO: link to docker-compose.yml).

Each container has an ansible account with ansible as the password. Using ansible, you can reach all of them using password authentication with the --ask-pass flag. To make the containers use SSH keys going forward, generate a new SSH key on the head node, then run this (TODO: link to playbook) playbook.

You may want to turn off SSH host key checking by creating an ansible.cfg file:

[defaults]
host_key_checking = false

You can then ping all the nodes by running:

# ansible -i inventory.ini all -m shell -a 'hostname'


Precedence[edit | edit source]

See: https://docs.ansible.com/ansible/latest/reference_appendices/general_precedence.html


Tips & Tricks[edit | edit source]

  • An entire task can be written in one line as a key=value space delimited string. Ansible will parse the string out into a map before passing it to its module.
  • When defining tasks, you can also call the import_tasks or import_role modules to load a task file or role directory.
  • Ansible will load all group_vars/group_name/*.yml files as variables. The group_vars directory is relative to the inventory file.
  • ansible-playbook can be executed against a specific node with the -l <host> limit option.
  • ansible-playbook can start at a specific task, useful when debugging a specific task, using --start-at-task <task-name>.
  • You can have ansible-playbook run only a specific set of tasks using the tags feature. Tag roles or tasks you want to run and then call ansible-playbook with --tags <tag>.
- name: do something
  tags: do-something
  debug:
    msg: "hello"

$ ansible-playbook example.yml --tags "do-something"

Distributed shell command[edit | edit source]

You can run commands on all your hosts similar to what you can do with salt-call * cmd.run with ansible.

# ansible all -m shell -a "yum -y update sudo" -i clusters/x/inventory.ini -f 10
  • Using the shell module, we run yum -y update sudo
  • We use 10 jobs in parallel (up from the default of 5)

Jinja[edit | edit source]

Test Conditions[edit | edit source]

All test conditions can make use of Jinja templating. This allows for some interesting when conditions when determining when a task or role is to be executed. Some examples below.

Description Example Condition
if the host is in the server group. when: 'server' in group_names
if the host is not in the server group. when: 'server' not in group_names
if some_value is true, or undefined. when: "some_value | default(true)"
if some_value is undefined when: some_value is undefined
if host is Red Hat when: ansible_os_family == "RedHat"
combining conditionals when: ansible_os_family == "RedHat"

or (ansible_os_family == "Debian" and ansible_os_major_version == "7")

checking whether a variable contains 'error' anywhere when: some_value.find('error') != -1
Check if an array has items. when: some_list | length > 0

See more on conditionals at: https://ansible-docs.readthedocs.io/zh/stable-2.0/rst/playbooks_conditionals.html

Align Text[edit | edit source]

Useful when trying to make config files look nice.

Alignment Code Example
Left Align {{ '%-20s' | format(my_text ) }} some value         
Center {{ my_text | center(80) }}       some value      
Right Align {{ '%20s' | format(my_text ) }}           some value

Comma Delimited List[edit | edit source]

Use the loop.last variable to determine whether this is the last item in the list. You may use either the if-expression, or an actual if block.

{% for item in my_list %}
{{ item }}
{{ "," if not loop.last }}
{% if not loop.last %},{% endif %}
{% endfor %}

Idempotent Random Number[edit | edit source]

Random numbers are useful when configuring many hosts since it lets you spread out tasks over time (in the case of cronjobs) or over different resources (in the case of mounting network filesystems from different endpoints). However, we want the same random number for each host so that each run through of the playbook remains the same. To accomplish this, we can make use of Jinja's random function seeded with the host's inventory name.

vars:
  fileserver_index: {{ 10 | random(seed=inventory_hostname) }}

Doing this, we can use fileserver_index, which contains a value between 0 through 9.

Troubleshooting[edit | edit source]

Import tasks not honoring when condition[edit | edit source]

Given the following task, the when condition is not being honored and is still evaluated:

- name: load tasks if condition is true
  import_tasks: tasks.yml
  when: some_condition_var is defined and some_condition_var is sameas true

The problem here is that import_tasks will load all tasks in tasks.yml at parse time rather than at runtime. This means that all tasks within tasks.yml will be loaded with the when condition applied to each task. With this example, if the condition is false, all tasks that were imported will be skipped.

To not load any of the tasks in tasks.yml when the condition is false, use include_tasks instead.

See Also[edit | edit source]

There are lots of resources online to help you get started. Some guides I've stumbled across which I found helpful were:

To set up the initial user, use something like the following. Use --ask-pass when running the playbook so that you log in using a password first, then this playbook will set up the SSH keys.

HPC example