Custom Resources Notes

[edit on GitHub]

Custom Resources

This is the recommended way of writing resources for all users. There are two gotchas which we’re working through:

1. For helper functions that you used to write in your provider code or used to mixin to your provider code, you have to use an action_class.class_eval do ... end block.

You cannot subclass, and must use mixins for code-sharing (which is really a best practice anyway – e.g. see languages like rust which do not support subclassing).

in resources/whatever.rb:

resource_name :my_resource
provides :my_resource

property :foo, String, name_property: true
extend MyResourceHelperFunctions  # probably only used for common properties which is why you extend with class methods

action :run do
  # helpers must be defined inside the action_class block
  a_helper()
  # you will save yourself some pain by referring to properties with `new_resource.foo` and not `foo`
  # since the latter works most of the time, but will troll you with odd scoping problems, while the
  # former just works.
  puts new_resource.foo
end

action_class.class_eval do
  include MyProviderHelperFunctions

  def a_helper
  end
end

“Old school” LWRPS

This method is not recommended, but is preferable to writing library resources/providers (as described below). It has the same functionality as library providers, only you cannot subclass and must use mixins for code sharing (which is good).

in resources/my_resource.rb:

resource_name :my_resource
provides :my_resource

property :foo, String, name_property: true
extend MyResourceHelperFunctions  # probably only used for common properties which is why you extend with class methods

in providers/my_resource.rb:

# you have to worry about this
def whyrun_supported?
  true
end

include MyProviderHelperFunctions

def a_helper
end

action :run do
  a_helper()
  # here you have to use new_resource.foo
  puts new_resource.foo
end

Library Resources/Providers

Library resources are discouraged since you can more easily shoot yourself in the foot. They used to be encouraged back before Chef Client 12.0 provides was introduced since it allowed for renaming the resource so that it didn’t have to be prefixed by the cookbook name.

There are many ways to go wrong writing library providers. One of the biggest issues is that internal Chef Infra Client code superficially looks like a library provider, but it is not. Chef internal resources do not inherit from LWRPBase and we’ve had to manually create resources directly through Chef::Resource::File.new(), we also have not been able to use_inline_resources and not had access to other niceties that cookbook authors have had access to for years now. We’ve got some modernization of internal Chef cookbook code now and resources like apt_update and apt_repository in core have started to be written more like cookbook code should be written, but core resources are actually behind the curve and are bad code examples.

in libraries/resource_my_resource.rb:

class MyBaseClass
  class Resource
    class MyResource < Chef::Resource::LWRPBase  # it is very important to inherit from LWRPBase
      resource_name :my_resource
      provides :my_resource

      property :foo, String, name_property: true
      extend MyResourceHelperFunctions  # probably only used for common properties which is why you extend with class methods
    end
  end
end

in libraries/resource_my_resource.rb:

class MyBaseClass
  class Resource
    class MyProvider < Chef::Provider::LWRPBase  # it is very important to inherit from LWRPBase

      # you have to worry about this
      def whyrun_supported?
        true
      end

      include MyProviderHelperFunctions

      def a_helper
      end

      # NEVER use `def action_run` here -- you defeat use_inline_resources and will break notifications if you do
      # If you don't understand how use_inline_resources is built and why you have to use the `action` method, and what the implications are and how resource notifications
      # break if use_inline_resources is not used and/or is broken, then you should really not be using library providers+resources.  You might feel "closer to the metal",
      # but you're now using a chainsaw without any guard...
      action :run do
        a_helper()
          # here you have to use new_resource.foo
          puts new_resource.foo
      end
    end
  end
end

updated_by_last_action

Modern Chef Infra Client code (since Chef Client version 11.0.0) should never have provider code which directly sets updated_by_last_action itself.

THIS CODE IS WRONG:

action :run do
  t = file "/tmp/foo" do
    content "foo"
  end
  t.run_action(:install)
  # This is Chef Client 10 code which fell through a timewarp into 2016 -- never use updated_by_last_action in modern Chef Client 11.x/12.x code
  t.new_resource.updated_by_last_action(true) if t.updated_by_last_action?
end

That used to be kinda-correct-code-with-awful-edge-cases back in Chef Client version 10. If you’re not using that version of Chef Client, please stop writing actions this way.

THIS IS CORRECT:

def whyrun_supported?
  true
end

action :run do
  file "/tmp/foo" do
    content "foo"
  end
end

That is the magic of use_inline_resources (and why use_inline_resources is turned on by default in Chef Client 12.5 resources) The sub-resources are defined in a sub-resource collection which is compiled and converged as part of the provider executing. Any resources that update in the sub-resource collection cause the resource itself to be updated automatically. Notifications then fire normally off the resource. It also works to arbitrary levels of nesting of sub-sub-sub-resources being updating causing the wrapping resources to update and fire notifications.

This also gets the why-run case correct. If all the work that you do in your resource is done by calling sub-resources, then why-run should work automatically. All your sub-resources will be NO-OP’d and will report what they would have done instead of doing it.

If you do need to write code which mutates the system through pure-Ruby then you should do so like this:

def whyrun_supported?
  true
end

action :run do
  unless File.exist?("/tmp/foo")
    converge_by("touch /tmp/foo") do
      ::FileUtils.touch "/tmp/foo"
    end
  end
end

When the converge_by block is run in why-run mode, it will only log touch "/tmp/foo" and will not run the code inside the block.

A converge_by block that is not wrapped in an idempotency check will always cause the resource to be updated, and will always cause notifications to fire. To prevent this, a properly written resource should wrap all converge_by checks with an idempotency check. The converge_if_changed block may be used instead which will wrap a converge_by block with an idempotency check for you.

action :run do
  # This code is bad, it lacks an idempotency check here.
  # It will always be updated
  # Chef Infra Client runs will always report a resource being updated
  # It will run the code in the block on every run
  converge_by("touch /tmp/foo") do
    ::FileUtils.touch "/tmp/foo"
  end
end

Of course it is vastly simpler to just use Chef Infra Client resources when you can. Compare the equivalent implementations:

action :run do
  file "/tmp/foo"
end

is basically the same as this:

action :run do
  unless File.exist?("/tmp/foo")
    converge_by("touch /tmp/foo") do
      ::FileUtils.touch "/tmp/foo"
    end
  end
end

You may see a lot of converge_by and updated_by_last_action in the core chef resources. This is sometimes due to the fact that Chef is written as a declarative language with an imperative language, which means someone has to take the first step and write the declarative file resources in imperative Ruby. As such, core Chef resources may not represent ideal code examples with regard to what custom resources should look like.