Providers are back-ends that support specific implementations of a given resource type, particularly for different platforms. Not all resource types have or need providers, but any resource type concerned about portability will likely need them.
For instance, there are more than 20 package providers, including providers for package formats like dpkg and rpm along with high-level package managers like apt and yum. A provider’s main job is to wrap client-side tools, usually by just calling out to those tools with the right information.
The examples on this page use the apt and dpkg package providers, and the examples used are current as of 0.23.0.
Providers are always associated with a single resource type,
so they are created by calling the
provide method on that resource type.
providemethod takes three arguments plus a block:
- The first argument must be the name of the provider, as
- The optional
:parentargument should be the name of a parent class.
- The optional
:sourceargument should be a symbol.
- The block takes no arguments, and implements the behavior of the provider.
- Base provider
- A provider can inherit from a base provider, which is never used
alone and only exists for other providers to inherit from. Use the full name of the
class. For example, all package providers have a common parent
Note the call to the
Puppet::Type.type(:package).provide(:dpkg, :parent => Puppet::Provider::Package) do desc "..." ... end
descmethod; this sets the documentation for this provider, and should include everything necessary for someone to use this provider.
- Another provider of the same resource type
- Providers can also specify another provider as their parent. If
it’s a provider for the same resource type, you can use the name of that provider as a
Note that we’re also specifying that this provider uses the dpkg
Puppet::Type.type(:package).provide(:apt, :parent => :dpkg, :source => :dpkg) do ... end
source; this tells Puppet to deduplicate packages from dpkg and apt, so the same package does not show up in an instance list from each provider type. Puppet defaults to creating a new source for each provider type, so you have to specify when a provider subclass shares a source with its parent class.
- A provider of any resource type
- Providers can also specify a provider of any resource type as
their parent. Use the
Puppet::Type.type(<NAME>).provider(<NAME>)methods to locate the provider.
# my_module/lib/puppet/provider/glance_api_config/ini_setting.rb Puppet::Type.type(:glance_api_config).provide( :ini_setting, # set ini_setting as the parent provider :parent => Puppet::Type.type(:ini_setting).provider(:ruby) ) do # implement section as the first part of the namevar def section resource[:name].split('/', 2).first end def setting # implement setting as the second part of the namevar resource[:name].split('/', 2).last end # hard code the file path (this allows purging) def self.file_path '/etc/glance/glance-api.conf' end end
The first question to ask about a new provider is where it will be functional, which Puppet calls suitable. Unsuitable providers cannot be used to do any work. The suitability test is late-binding, meaning that you can have a resource in your configuration that makes a provider suitable.
If you start
puppet in debug mode, you’ll see
the results of failed provider suitability tests for the resource types you’re
commands :dpkg => "/usr/bin/dpkg" commands :dpkg_deb => "/usr/bin/dpkg-deb" commands :dpkgquery => "/usr/bin/dpkg-query"
confineclass method with
falseas the name of the test and your test as the value:
To test Facter values, use the name of the fact:
confine :exists => "/etc/debian_release" confine :true => /^10\.[0-4]/.match(product_version) confine :false => (Puppet[:ldapuser] == "")
Case doesn’t matter in the tests, nor does it matter whether the values are strings or symbols. It also doesn’t matter whether you specify an array or a single value — Puppet does an OR on the list of values.
confine :operatingsystem => [:debian, :solaris] confine :puppetversion => "0.23.0"
lib/puppet/feature/*.rb, supply the name of the feature. This is preferable to using a
confine :truestatement that calls
Puppet.featuresbecause the expression is evaluated only one time. Puppet enables the provider if the feature becomes available during a run (for example, if a package is installed during the run).
You can create custom features. They live in
confine :feature => :posix confine :feature => :rrd
lib/puppet/feature/*.rband an example can be found here. These features can be shipped in a similar manner as types and providers are shipped within modules and are pluginsynced.
Using custom features you can delay resource evaluation until the provider becomes suitable. This is a way of informing Puppet that your provider depends on a file being created by Puppet, or a certain fact being set to some value, or it not being set at all.
Providers are generally meant to be hidden from the users, allowing them to focus on resource specification rather than implementation details, soPuppet does what it can to choose an appropriate default provider for each resource type.
The same fact-matching functionality as confinement is used.
defaultfor :operatingsystem => :debian
defaultfor :operatingsystemmajrelease => /^[5-7]$/
How providers interact with resources
Providers do nothing on their own; all of their action is triggered through an associated resource (or, in special cases, from the transaction). Because of this, resource types are essentially free to define their own provider interface if necessary, and providers were initially developed without a clear resource-provider API (mostly because it wasn’t clear whether such an API was necessary or what it would look like). At this point, however, there is a default interface between the resource type and the provider.
This interface consists entirely of
getter and setter methods. When the resource is retrieving its current state, it
iterates across all of its properties and calls the getter method on the provider
for that property. For instance, when a
user resource is having its state retrieved and its
shell properties are
being managed, then the resource calls
shell on the provider to figure out what the current
state of each of those properties is. This method call is in the
retrieve method in
When a resource is being modified, it calls the equivalent setter
method for each property on the provider. Again using our user example, if the
uid was in sync
not, then the resource would call
shell=(value) with the new shell value.
The transaction is responsible for storing these returned values and
deciding which value to send, and it does its work through a
PropertyChange instance. It calls
sync on each of
the properties, which in turn call the setter by default.
You can override that interface as necessary for your resource type.
All providers must define an
instances class method that returns a list
of provider instances, one for each existing instance of that provider. For example,
the dpkg provider should return a provider instance for every package in the dpkg
By default, you have to define all of your getter and setter methods. For simple cases, this is sufficient — you just implement the code that does the work for that property. For the more complicated aspects of provider implementation, Puppet has prefetching, resource methods, and flushing.
First, Puppet transactions
prefetch provider information by calling
prefetch on each used provider type. This calls the
in turn, which returns a list of provider instances with the current resource state
already retrieved and stored in a
@property_hash instance variable. The prefetch method
then tries to find any matching resources, and assigns the retrieved providers to
found resources. This way you can get information on all of the resources you’re
managing in just a few method calls, instead of having to call all of the getter
methods for every property being managed. Note that it also means that providers are
often getting replaced, so you cannot maintain state in a provider.
For providers that directly modify the system when a setter
method is called, there’s no substitute for defining them manually. But for
resources that get flushed to disk in one step, such as the
ParsedFile providers, there is a
mk_resource_methods class method that creates a getter and setter for
each property on the resource. These methods retrieve and set the appropriate value
Many providers model files or parts of files, so it makes sense to save up
all of the writes and do them in one run. Providers that need this functionality can
define a flush instance method to do this. The transaction calls this method after
all values are synced (which means that the provider should have them all in its
variable) but before refresh is called on the resource (if