Ansible

Ansible is a extra-simple tool/API for doing 'parallel remote things' over SSH -- whether executing commands, running "modules", or executing larger 'playbooks' that can serve as a configuration management or deployment system.

While Func, which I co-wrote, aspired to avoid using SSH and have it's own daemon infrastructure, Ansible aspires to be quite different and more minimal, but still able to grow more modularly over time. This is based on talking to a lot of users of various tools and wishing to eliminate problems with connectivity and long running daemons, or not picking tool X because they preferred to code in Y. Further, playbooks take things a whole step further, building the config and deployment system I always wanted to build.

Why use Ansible versus something else? (Fabric, Capistrano, mCollective, Func, SaltStack, etc?) It will have far less code, it will be more correct, and it will be the easiest thing to hack on and use you'll ever see -- regardless of your favorite language of choice. Want to only code plugins in bash or clojure? Ansible doesn't care. The docs will fit on one page and the source will be blindingly obvious.

Design Principles

• Dead simple setup
• Super fast & parallel by default
• No server or client daemons; use existing SSHd
• No additional software required on client boxes
• Modules can be written in ANY language
• Awesome API for creating very powerful distributed scripts
• Be usable as non-root
• Create the easiest config management system to use, ever.

Requirements

For the server the tool is running from, only:

• paramiko
• python 2.6 (or the 2.4/2.5 backport of the multiprocessing module)
• PyYAML (only if using playbooks)

Optional -- If you want to push templates, the nodes need a template library, which for bonus points you can install with ansible! Easy enough.

• python-jinja2

Inventory file

To use ansible you must have a list of hosts somewhere. The default inventory host list (override with -l) is /etc/ansible/hosts and is a list of all hostnames to manage with ansible, one per line. These can be hostnames or IPs.

Example:

abc.example.com
def.example.com
192.168.10.50
192.168.10.51

This list is further filtered by the pattern wildcard (-p) to target specific hosts. This is covered below. You can also organize groups of systems by having multiple inventory files (i.e. keeping webservers different from dbservers, etc)

Massive Parallelism, Pattern Matching, and a Usage Example

Reboot all web servers in Atlanta, 10 at a time:

ssh-agent bash
ssh-add ~/.ssh/id_rsa.pub
ansible -p "atlanta-web*" -f 10 -n command -a "/sbin/reboot"

Other than the comamnd module, though, ansible modules are not scripts. They make the remote system look like you state, and run the commands neccessary to get it there.

Read the manpage

File Transfer

Ansible can SCP lots of files to lots of places in parallel.

ansible -p "web-*.acme.net" -f 10 -n copy -a "/etc/hosts /tmp/hosts"

Templating

JSON files can be placed for template metadata using Jinja2. Variables placed by 'setup' can be reused between ansible runs.

ansible -p "*" -n setup -a "favcolor=red ntp_server=192.168.1.1"
ansible -p "*" -n template -a "src=/srv/motd.j2 dest=/etc/motd"
ansible -p "*" -n template -a "src=/srv/ntp.j2 dest=/etc/ntp.conf"

Need something like the fqdn in a template? If facter or ohai are installed, data from these projects will also be made available to the template engine, using 'facter_' and 'ohai_' prefixes for each.

Git Deployments

Deploy your webapp straight from git

ansible -p "web*" -n git -a "repo=git://foo dest=/srv/myapp version=HEAD"

Take Inventory

Run popular open-source data discovery tools across a wide number of hosts. This is best used from API scripts that want to learn about remote systems.

ansible -p "dbserver*" -n facter
ansible -p "dbserver"" -n ohai

Other Modules

See the library directory for lots of extras. There's also a manpage, ansible-modules(5) that covers all the options they take. You can read the asciidoc in github in the 'docs' directory.

Playbooks

Playbooks are particularly awesome. Playbooks can batch ansible commands together, and can even fire off triggers when certain commands report changes.

They are the basis for a really simple configuration management system, unlike any that already exist, and one that is very well suited to deploying complex multi-machine applications.

An example showing just once pattern in a playbook is below. Playbooks can contain multple patterns in a single file.

---
- pattern: 'webservers*'
  comment: webserver setup steps
  hosts: '/etc/ansible/hosts'
  tasks:
  - name: configure template & module variables for future template calls
    action: setup http_port=80 max_clients=200
  - name: write the apache config file
    action: template src=/srv/templates/httpd.j2 dest=/etc/httpd.conf
    notify:
    - restart apache
  - name: ensure apache is running
    action: service name=httpd state=started
  handlers:
    - name: restart apache
    - action: service name=httpd state=restarted

See the playbook format manpage -- ansible-playbook(5) for more details.

To run a playbook:

ansible-playbook playbook.yml

API

The Python API is pretty powerful.

import ansible.runner

runner = ansible.runner.Runner(
   module_name='ping',
   module_args='',
   pattern='web*',
   forks=10
)
datastructure = runner.run()

And returns results per host, for hosts we could contact and also ones that we failed to contact.

{
    "dark" : {
       "web1.example.com" : "failure message"
    }
    "contacted" : {
       "web2.example.com" : 1
    }

}

A module can return any type of JSON data it wants, so Ansible can be used as a framework to build arbitrary applications and very powerful scripts.

Future plans

See github's issue tracker for what we're thinking about

License

GPLv3

Mailing List

Join the mailing list to talk about Ansible!

ansible-project

Author

Michael DeHaan -- michael.dehaan@gmail.com

http://michaeldehaan.net