different sets of nodes through :doc:`playbooks`. Playbooks contain one or
more plays, each executed against a different batch of nodes. Think about
webservers, database servers, and backend servers in a multi-node web environment. A play could address each set of machines in a cycle, ensuring the configurations of the machines were correct and also updating them to the specified
version of software if required.
Multi-machine software deployment is poorly solved by most systems management tools -- often due to architectural nature of being pull oriented and having complex ordering systems, they cover configuration but fail at deployment when updating tiers of machines in well defined steps. This results in using two (or more) logically distinct tools and having complex overlap between them.
Other deployment oriented frameworks similarly cover deployment well but lack a strongly defined resource model and devolve into glorified remote scripts. Ansible playbooks -- having been designed with this problem in mind -- are good at both deployment & idempotent configuration, meaning you don't have to spread your infrastructure management out between different tools (Puppet+Capistrano, Chef+Fabric, etc), and performing ordered steps between different classes of machines is no problem, yet our modules affect system state only when required -- while avoiding the problem of fragile scripting that assumes certain starting
Ansible is also unique in other ways. Extending ansible does not require programming in any particular language -- you can write :doc:`modules` as idempotent scripts or programs that return simple JSON. Ansible is also pragmatic, so when you need to, it's also trivially easy to just execute useful shell commands.
Compared with most configuration managememnt tools, Ansible is also more secure. While most configuration management tools use a daemon, running as root with full access to the system, with it's own in-house developed PKI infrastructure, Ansible just uses SSH (and supports sudo as neccesssary). There is no additional attack surface and OpenSSH is one of the most peer reviewed security components out there.
If a central server containing your playbooks are comprimised, your nodes are not -- which is NOT the case
of these other tools, which can, more or less, turn into a botnet. Our security approach is to avoid writing custom
crypto code altogether, and rely on the most secure part of the Linux/Unix subsystem that your machines are already using.
Systems management doesn't have to be complicated. Ansible's docs will remain short & simple, and the source will be blindingly obvious. We've learned well from "Infrastructure is Code". Infrastructure should be easy and powerful to command, but it should not look like code, lest it acquire the disadvantages of a software project -- bugs, complexity, and overhead. Infrastructure configurations should be simple, easy to develop, and easy to audit.
