Puppet components

Before diving into installation and configuration details, we need to clarify and explain some Puppet terminology to get the whole picture.

Puppet features a declarative Domain Specific Language (DSL), which expresses the desired state and properties of the managed resources.

Resources can be any component of a system, for example, packages to install, services to start, files to manage, users to create, and also custom and specific resources, such as MySQL grants, Apache virtual hosts, and so on.

Puppet code is written in manifests, which are simple text files with a .pp extension. Resources can be grouped in classes (do not consider them classes as in OOP, they aren't). Classes and all the files needed to define the configurations required are generally placed in modules, which are directories structured in a standard way that are supposed to manage specific applications or system's features (there are modules to manage Apache, MySQL, sudo, sysctl, networking, and so on).

When Puppet is executed, it first runs facter, a companion application, which gathers a series of variables about the system (IP address, hostname, operating system, and MAC address), which are called facts and are sent to the Master.

Facts and user-defined variables can be used in manifests to manage how and what resources to provide to clients.

When the Master receives a connection, then it looks in its manifests (starting from /etc/puppet/manifests/site.pp) what resources have to be applied for that client host, also called node.

The Master parses all the DSL code and produces a catalog, which is sent back to the client (in PSON format, a JSON variant used in Puppet). The production of the catalog is often referred to as catalog compilation.

Once the client receives the catalog, it starts to apply all the resources declared there; packages are installed (or removed), services started, configuration files created or changed, and so on. The same catalog can be applied multiple times, if there are changes on a managed resource (for example, a manual modification of a configuration file) they are reverted back to the state defined by Puppet; if the system's resources are already at the desired state, nothing happens.

This property is called idempotence and is at the root of the Puppet declarative model; since it defines the desired state of a system, it must operate in a way that ensures that this state is obtained whatever are the starting conditions and the number of times Puppet is applied.

Puppet can report the changes it makes on the system and audit the drift between the system's state and the desired state as defined in its catalog.