Puppet Custom Types and Providers

Introduction

Puppet is a software configuration management solution to manage IT infrastructure. One of the first things to be learnt about Puppet is its domain-specific language – the Puppet-DSL – and the concepts that come with it.

Users can organize their code in classes and modules and use pre-defined resource types to manage resources like files, packages, services and others.

The most commonly used types are part of the Puppet core, implemented in Ruby. Composite resource types may be defined via Puppet-DSL from already known types by the Puppet user themself, or imported as part of an external Puppet module which is maintained by external module developers.

It so happens that Puppet users can stay within the Puppet-DSL for a long time even when they deal with Puppet on a regular basis.

The first time I had a glimpse into this topic was when Debian Stable was shipping Puppet 5.5, which was not too long ago. The Puppet 5.5 documentation includes a chapter on custom types and provider development respectively, but to me they felt incomplete and lacking self contained examples. Apparently I was not the only one feeling that way, even though Puppet’s gem documentation is a good overview of what is possible in principle.

Gary Larizza’s blog post was more than ten years ago. I had another look into the documentation for Puppet 7 on that topic recently, as this is the Puppet version in current’s Debian Stable.

The Puppet 5.5 way to type & provider development is now called the low level method, and its documentation has not changed significantly. However, Puppet 6 upwards recommends a new method to create custom types & providers via the so-called Resource-API, whose documentation is a major improvement compared to the low-level method’s. The Resource-API is not a replacement, though, and has several documented limitations.

Nevertheless, for the remaining blog post, we will re-prototype a small portion of files functionality using the low-level method, as well as the Resource-API, namely the ensure and content properties.

Preparations

The following preparations are not necessary in an agent-server setup. We use bundle to obtain a puppet executable for this demo.

demo@85c63b50bfa3:~$ cat > Gemfile <<EOF
source 'https://rubygems.org'

gem 'puppet', '>= 6'
EOF

demo@85c63b50bfa3:~$ bundle install
Fetching gem metadata from https://rubygems.org/........
Resolving dependencies...
...
Installing puppet 8.10.0
Bundle complete! 1 Gemfile dependency, 17 gems now installed.
Bundled gems are installed into `./.vendor`

demo@85c63b50bfa3:~$ cat > file_builtin.pp <<EOF
$file = '/home/demo/puppet-file-builtin'
file {$file: content => 'This is madness'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply file_builtin.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.01 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-builtin]/ensure: defined content as '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e'
Notice: Applied catalog in 0.04 seconds

demo@85c63b50bfa3:~$ sha256sum /home/demo/puppet-file-builtin
0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e /home/demo/puppet-file-builtin

Keep in mind the state change information printed by Puppet.

Low-level prototype

Custom types and providers that are not installed via a Gem need to be part of some Puppet module, so they can be copied to Puppet agents via the pluginsync mechanism.

A common location for Puppet modules is the modules directory inside a Puppet environment. For this demo, we declare a demo module.

Basic functionality

Our first attempt is the following type definition for a new type we will call file_llmethod. It has no documentation or validation of input values.

# modules/demo/lib/puppet/type/file_llmethod.rb

Puppet::Type.newtype(:file_llmethod) do
  newparam(:path, namevar: true) {}
  newproperty(:content) {}
  newproperty(:ensure) do
    newvalues(:present, :absent)
    defaultto(:present)
  end
end

We have declared a path parameter that serves as the namevar for this type – there cannot be other file_llmethod instances managing the same path. The ensure property is restricted to two values and defaults to present.

The following provider implementation consists of a getter and a setter for each of the two properties content and ensure.

# modules/demo/lib/puppet/provider/file_llmethod/ruby.rb

Puppet::Type.type(:file_llmethod).provide(:ruby) do
  def ensure
    File.exist?(@resource[:path]) ? :present : :absent
  end

  def ensure=(value)
    if value == :present
      # reuse setter
      self.content=(@resource[:content])
    else
      File.unlink(@resource[:path])
    end
  end

  def content
    File.read(@resource[:path])
  end

  def content=(value)
    File.write(@resource[:path], value)
  end
end

This gives us the following:

demo@85c63b50bfa3:~$ cat > file_llmethod_create.pp <<EOF
$file = '/home/demo/puppet-file-lowlevel-method-create'
file {$file: ensure => absent} ->
file_llmethod {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_create.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.01 seconds
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-create]/ensure: defined 'ensure' as 'present'

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_create.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.01 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-create]/ensure: removed
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-create]/ensure: defined 'ensure' as 'present'

demo@85c63b50bfa3:~$ cat > file_llmethod_change.pp <<EOF
$file = '/home/demo/puppet-file-lowlevel-method-change'
file {$file: content => 'This is madness'} ->
file_llmethod {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_change.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-change]/ensure: defined content as '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e'
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-change]/content: content changed 'This is madness' to 'This is Sparta!'

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_change.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-change]/content: content changed '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2' to '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e'
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-change]/content: content changed 'This is madness' to 'This is Sparta!'

Our custom type already kind of works, even though we have not implemented any explicit comparison of is- and should-state. Puppet does this for us based on the Puppet catalog and the property getter return values. Our defined setters are also invoked by Puppet on demand, only.

We can also see that the ensure state change notice is defined 'ensure' as 'present' and does not incorporate the desired content in any way, while the content state change notice shows plain text. Both tell us that the SHA256 checksum from the file_builtin.pp example is already something non-trivial.

Validating input

As a next step we add validation for path and content.

# modules/demo/lib/puppet/type/file_llmethod.rb

Puppet::Type.newtype(:file_llmethod) do
  newparam(:path, namevar: true) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
      fail "#{value} is not an absolute path" unless File.absolute_path?(value)
    end
  end

  newproperty(:content) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
    end
  end

  newproperty(:ensure) do
    newvalues(:present, :absent)
    defaultto(:present)
  end
end

Failed validations will look like these:

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules --exec 'file_llmethod {"./relative/path": }'
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Error: Parameter path failed on File_llmethod[./relative/path]: ./relative/path is not an absolute path (line: 1)

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules --exec 'file_llmethod {"/absolute/path": content => 42}'
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Error: Parameter content failed on File_llmethod[/absolute/path]: 42 is not a String (line: 1)

Content Checksums

We override change_to_s so that state changes include content checksums:

# modules/demo/lib/puppet/type/file_llmethod.rb

require 'digest'

Puppet::Type.newtype(:file_llmethod) do
  newparam(:path, namevar: true) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
      fail "#{value} is not an absolute path" unless File.absolute_path?(value)
    end
  end

  newproperty(:content) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
    end
    define_method(:change_to_s) do |currentvalue, newvalue|
      old = "{sha256}#{Digest::SHA256.hexdigest(currentvalue)}"
      new = "{sha256}#{Digest::SHA256.hexdigest(newvalue)}"
      "content changed '#{old}' to '#{new}'"
    end
  end

  newproperty(:ensure) do
    define_method(:change_to_s) do |currentvalue, newvalue| if currentvalue == :absent should = @resource.property(:content).should digest = "{sha256}#{Digest::SHA256.hexdigest(should)}" "defined content as '#{digest}'" else super(currentvalue, newvalue) end end newvalues(:present, :absent) defaultto(:present) end end

The above type definition yields:

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_create.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-create]/ensure: removed
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-create]/ensure: defined content as '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_change.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-change]/content: content changed '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2' to '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e'
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-change]/content: content changed '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e' to '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'

Improving memory footprint

So far so good. While our current implementation apparently works, it has at least one major flaw. If the managed file already exists, the provider stores the file’s whole content in memory.

demo@85c63b50bfa3:~$ cat > file_llmethod_change_big.pp <<EOF
$file = '/home/demo/puppet-file-lowlevel-method-change_big'
file_llmethod {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ rm -f /home/demo/puppet-file-lowlevel-method-change_big
demo@85c63b50bfa3:~$ ulimit -Sv 200000

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_change_big.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-change_big]/ensure: defined content as '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'
Notice: Applied catalog in 0.02 seconds

demo@85c63b50bfa3:~$ dd if=/dev/zero of=/home/demo/puppet-file-lowlevel-method-change_big seek=8G bs=1 count=1
1+0 records in
1+0 records out
1 byte copied, 8.3047e-05 s, 12.0 kB/s

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_change_big.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Error: Could not run: failed to allocate memory

Instead, the implementation should only store the checksum so that Puppet can decide based on checksums if our content= setter needs to be invoked.

This also means that the Puppet catalog’s content needs to be checksummed by munge before it it processed by Puppet’s internal comparison routine. Luckily we also have access to the original value via shouldorig.

# modules/demo/lib/puppet/type/file_llmethod.rb

require 'digest'

Puppet::Type.newtype(:file_llmethod) do
  newparam(:path, namevar: true) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
      fail "#{value} is not an absolute path" unless File.absolute_path?(value)
    end
  end

  newproperty(:content) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
    end
    munge do |value|
      "{sha256}#{Digest::SHA256.hexdigest(value)}"
    end
    # No need to override change_to_s with munging
  end

  newproperty(:ensure) do
    define_method(:change_to_s) do |currentvalue, newvalue|
      if currentvalue == :absent
        should = @resource.property(:content).should
        "defined content as '#{should}'"
      else
        super(currentvalue, newvalue)
      end
    end
    newvalues(:present, :absent)
    defaultto(:present)
  end
end

# modules/demo/lib/puppet/provider/file_llmethod/ruby.rb

Puppet::Type.type(:file_llmethod).provide(:ruby) do
  ...

  def content
    File.open(@resource[:path], 'r') do |file|
      sha = Digest::SHA256.new
      while chunk = file.read(2**16)
        sha << chunk
      end
      "{sha256}#{sha.hexdigest}"
    end
  end

  def content=(value)
    # value is munged, but we need to write the original
    File.write(@resource[:path], @resource.parameter(:content).shouldorig[0])
  end
end

Now we can manage big files:

demo@85c63b50bfa3:~$ ulimit -Sv 200000
demo@85c63b50bfa3:~$ dd if=/dev/zero of=/home/demo/puppet-file-lowlevel-method-change_big seek=8G bs=1 count=1
1+0 records in
1+0 records out
1 byte copied, 9.596e-05 s, 10.4 kB/s
demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_change_big.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-change_big]/content: content changed '{sha256}ef17a425c57a0e21d14bec2001d8fa762767b97145b9fe47c5d4f2fda323697b' to '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'

Ensure it the Puppet way

There is still something not right. Maybe you have noticed that our provider’s content getter attempts to open a file unconditionally, and yet the file_llmethod_create.pp run has not produced an error. It seems that an ensure transition from absent to present short-circuits the content getter, even though we have not expressed a wish to do so.

It turns out that an ensure property gets special treatment by Puppet. If we had attempted to use a makeitso property instead of ensure, there would be no short-circuiting and the content getter would raise an exception.

We will not fix the content getter though. If Puppet has special treatment for ensure, we should use Puppet’s intended mechanism for it, and declare the type ensurable:

# modules/demo/lib/puppet/type/file_llmethod.rb

require 'digest'

Puppet::Type.newtype(:file_llmethod) do
  ensurable

  newparam(:path, namevar: true) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
      fail "#{value} is not an absolute path" unless File.absolute_path?(value)
    end
  end

  newproperty(:content) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
    end
    munge do |value|
      "{sha256}#{Digest::SHA256.hexdigest(value)}"
    end
  end
end

With ensurable the provider needs to implement three new methods, but we can drop the ensure accessors:

# modules/demo/lib/puppet/provider/file_llmethod/ruby.rb

Puppet::Type.type(:file_llmethod).provide(:ruby) do
  def exists?
    File.exist?(@resource[:name])
  end

  def create
    self.content=(:dummy)
  end

  def destroy
    File.unlink(@resource[:name])
  end

  def content
    File.open(@resource[:path], 'r') do |file|
      sha = Digest::SHA256.new
      while chunk = file.read(2**16)
        sha << chunk
      end
      "{sha256}#{sha.hexdigest}"
    end
  end

  def content=(value)
    # value is munged, but we need to write the original
    File.write(@resource[:path], @resource.parameter(:content).shouldorig[0])
  end
end

However, now we have lost the SHA256 checksum on file creation:

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_create.pp Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-create]/ensure: removed Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-create]/ensure: created

To get it back, we replace ensurable by an adapted implementation of it, which includes our previous change_to_s override:

newproperty(:ensure, :parent => Puppet::Property::Ensure) do
  defaultvalues
  define_method(:change_to_s) do |currentvalue, newvalue|
    if currentvalue == :absent
      should = @resource.property(:content).should
      "defined content as '#{should}'"
    else
      super(currentvalue, newvalue)
    end
  end
end

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_llmethod_create.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.02 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-lowlevel-method-create]/ensure: removed
Notice: /Stage[main]/Main/File_llmethod[/home/demo/puppet-file-lowlevel-method-create]/ensure: defined content as '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'

Our final low-level prototype is thus as follows.

Final low-level prototype

# modules/demo/lib/puppet/type/file_llmethod.rb

# frozen_string_literal: true

require 'digest'

Puppet::Type.newtype(:file_llmethod) do
  newparam(:path, namevar: true) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
      fail "#{value} is not an absolute path" unless File.absolute_path?(value)
    end
  end

  newproperty(:content) do
    validate do |value|
      fail "#{value} is not a String" unless value.is_a?(String)
    end
    munge do |value|
      "{sha256}#{Digest::SHA256.hexdigest(value)}"
    end
  end

  newproperty(:ensure, :parent => Puppet::Property::Ensure) do
    defaultvalues
    define_method(:change_to_s) do |currentvalue, newvalue|
      if currentvalue == :absent
        should = @resource.property(:content).should
        "defined content as '#{should}'"
      else
        super(currentvalue, newvalue)
      end
    end
  end
end

# modules/demo/lib/puppet/provider/file_llmethod/ruby.rb

# frozen_string_literal: true

Puppet::Type.type(:file_llmethod).provide(:ruby) do
  def exists?
    File.exist?(@resource[:name])
  end

  def create
    self.content=(:dummy)
  end

  def destroy
    File.unlink(@resource[:name])
  end

  def content
    File.open(@resource[:path], 'r') do |file|
      sha = Digest::SHA256.new
      while (chunk = file.read(2**16))
        sha << chunk
      end
      "{sha256}#{sha.hexdigest}"
    end
  end

  def content=(_value)
    # value is munged, but we need to write the original
    File.write(@resource[:path], @resource.parameter(:content).shouldorig[0])
  end
end

Resource-API prototype

According to the Resource-API documentation we need to define our new file_rsapi type by calling Puppet::ResourceApi.register_type with several parameters, amongst which are the desired attributes, even ensure.

# modules/demo/lib/puppet/type/file_rsapi.rb

require 'puppet/resource_api'

Puppet::ResourceApi.register_type(
  name: 'file_rsapi', 
  attributes: {
    content: {
      desc: 'description of content parameter',
      type: 'String' 
    },
    ensure: {
      default: 'present',
      desc: 'description of ensure parameter',
      type: 'Enum[present, absent]'
    },
    path: {
      behaviour: :namevar,
      desc: 'description of path parameter',
      type: 'Pattern[/\A\/([^\n\/\0]+\/*)*\z/]'
    },
  },
  desc: 'description of file_rsapi'
)

The path type uses a built-in Puppet data type. Stdlib::Absolutepath would have been more convenient but external data types are not possible with the Resource-API yet.

In comparison with our low-level prototype, the above type definition has no SHA256-munging and SHA256-output counterparts. The canonicalize provider feature looks similar to munging, but we skip it for now.

The Resource-API documentation tells us to implement a get and a set method in our provider, stating

The get method reports the current state of the managed resources. It returns an enumerable of all existing resources. Each resource is a hash with attribute names as keys, and their respective values as values.

This demand is the first bummer, as we definitely do not want to read all files with their content and store it in memory. We can ignore this demand – how would the Resource-API know anyway.

However, the documented signature is def get(context) {...} where context has no information about the resource we want to manage.

This would have been a show-stopper, if the simple_get_filter provider feature didn’t exist, which changes the signature to def get(context, names = nil) {...}.

Our first version of file_rsapi is thus the following.

Basic functionality

# modules/demo/lib/puppet/type/file_rsapi.rb

require 'puppet/resource_api'

Puppet::ResourceApi.register_type(
  name: 'file_rsapi',
  features: %w[simple_get_filter],
  attributes: {
    content: {
      desc: 'description of content parameter',
      type: 'String' 
    },
    ensure: {
      default: 'present',
      desc: 'description of ensure parameter',
      type: 'Enum[present, absent]'
    },
    path: {
      behaviour: :namevar,
      desc: 'description of path parameter',
      type: 'Pattern[/\A\/([^\n\/\0]+\/*)*\z/]'
    },
  },
  desc: 'description of file_rsapi'
)

# modules/demo/lib/puppet/provider/file_rsapi/file_rsapi.rb

require 'digest'

class Puppet::Provider::FileRsapi::FileRsapi
  def get(context, names)
    (names or []).map do |name|
      File.exist?(name) ? {
        path: name,
        ensure: 'present',
        content: filedigest(name),
      } : nil
    end.compact # remove non-existing resources
  end

  def set(context, changes)
    changes.each do |path, change|
      if change[:should][:ensure] == 'present'
        File.write(path, change[:should][:content])
      elsif File.exist?(path)
        File.delete(path)
      end
    end
  end

  def filedigest(path)
    File.open(path, 'r') do |file|
      sha = Digest::SHA256.new
      while chunk = file.read(2**16)
         sha << chunk
      end
      "{sha256}#{sha.hexdigest}"
    end
  end
end

The desired content is written correctly into the file, but we have again no SHA256 checksum on creation as well as unnecessary writes, because the checksum from get does not match the cleartext from the catalog:

demo@85c63b50bfa3:~$ cat > file_rsapi_create.pp <<EOF
$file = '/home/demo/puppet-file-rsapi-create'
file       {$file: ensure  => absent} ->
file_rsapi {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_rsapi_create.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.03 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-rsapi-create]/ensure: removed
Notice: /Stage[main]/Main/File_rsapi[/home/demo/puppet-file-rsapi-create]/ensure: defined 'ensure' as 'present'
Notice: Applied catalog in 0.02 seconds

demo@85c63b50bfa3:~$ cat > file_rsapi_change.pp <<EOF
$file = '/home/demo/puppet-file-rsapi-change'
file       {$file: content => 'This is madness'} ->
file_rsapi {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_rsapi_change.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.03 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-rsapi-change]/content: content changed '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2' to '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e'
Notice: /Stage[main]/Main/File_rsapi[/home/demo/puppet-file-rsapi-change]/content: content changed '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e' to 'This is Sparta!'
Notice: Applied catalog in 0.03 seconds

Hence we enable and implement the canonicalize provider feature.

Canonicalize

According to the documentation, canonicalize is applied to the results of get as well as to the catalog properties. On the one hand, we do not want to read the file’s content into memory, on the other hand cannot checksum the file’s content twice.

An easy way would be to check whether the canonicalized call happens after the get call:

def canonicalize(context, resources)
  if @stage == :get
    # do nothing, is-state canonicalization already performed by get()
  else
    # catalog canonicalization
    ...
  end
end

def get(context, paths)
  @stage = :get

  ...
end

While this works for the current implementation of the Resource-API, there is no guarantee about the order of canonicalize calls. Instead we subclass from String and handle checksumming internally. We also add some state change calls to the appropriate context methods.

Our final Resource-API-based prototype implementation is:

# modules/demo/lib/puppet/type/file_rsapi.rb

# frozen_string_literal: true

require 'puppet/resource_api'

Puppet::ResourceApi.register_type(
  name: 'file_rsapi',
  features: %w[simple_get_filter canonicalize],
  attributes: {
    content: {
      desc: 'description of content parameter',
      type: 'String'
    },
    ensure: {
      default: 'present',
      desc: 'description of ensure parameter',
      type: 'Enum[present, absent]'
    },
    path: {
      behaviour: :namevar,
      desc: 'description of path parameter',
      type: 'Pattern[/\A\/([^\n\/\0]+\/*)*\z/]'
    }
  },
  desc: 'description of file_rsapi'
)

# modules/demo/lib/puppet/provider/file_rsapi/file_rsapi.rb

# frozen_string_literal: true

require 'digest'
require 'pathname'

class Puppet::Provider::FileRsapi::FileRsapi
  class CanonicalString < String
    attr_reader :original

    def class
      # Mask as String for YAML.dump to mitigate
      #   Error: Transaction store file /var/cache/puppet/state/transactionstore.yaml
      #   is corrupt ((/var/cache/puppet/state/transactionstore.yaml): Tried to
      #   load unspecified class: Puppet::Provider::FileRsapi::FileRsapi::CanonicalString)
      String
    end

    def self.from(obj)
      return obj if obj.is_a?(self)
      return new(filedigest(obj)) if obj.is_a?(Pathname)

      new("{sha256}#{Digest::SHA256.hexdigest(obj)}", obj)
    end

    def self.filedigest(path)
      File.open(path, 'r') do |file|
        sha = Digest::SHA256.new
        while (chunk = file.read(2**16))
          sha << chunk
        end
        "{sha256}#{sha.hexdigest}"
      end
    end

    def initialize(canonical, original = nil)
      @original = original
      super(canonical)
    end
  end

  def canonicalize(_context, resources)
    resources.each do |resource|
      next if resource[:ensure] == 'absent'

      resource[:content] = CanonicalString.from(resource[:content])
    end
    resources
  end

  def get(_context, names)
    (names or []).map do |name|
      next unless File.exist?(name)

      {
        content: CanonicalString.from(Pathname.new(name)),
        ensure: 'present',
        path: name
      }
    end.compact # remove non-existing resources
  end

  def set(context, changes)
    changes.each do |path, change|
      if change[:should][:ensure] == 'present'
        File.write(path, change[:should][:content].original)
        if change[:is][:ensure] == 'present'
          # The only other possible change is due to content,
          # but content change transition info is covered implicitly
        else
          context.created("#{path} with content '#{change[:should][:content]}'")
        end
      elsif File.exist?(path)
        File.delete(path)
        context.deleted(path)
      end
    end
  end
end

It gives us:

demo@85c63b50bfa3:~$ cat > file_rsapi_create.pp <<EOF
$file = '/home/demo/puppet-file-rsapi-create'
file       {$file: ensure  => absent} ->
file_rsapi {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_rsapi_create.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.03 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-rsapi-create]/ensure: removed
Notice: /Stage[main]/Main/File_rsapi[/home/demo/puppet-file-rsapi-create]/ensure: defined 'ensure' as 'present'
Notice: file_rsapi: Created: /home/demo/puppet-file-rsapi-create with content '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'

demo@85c63b50bfa3:~$ cat > file_rsapi_change.pp <<EOF
$file = '/home/demo/puppet-file-rsapi-change'
file       {$file: content => 'This is madness'} ->
file_rsapi {$file: content => 'This is Sparta!'}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_rsapi_change.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.03 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-rsapi-change]/content: content changed '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2' to '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e'
Notice: /Stage[main]/Main/File_rsapi[/home/demo/puppet-file-rsapi-change]/content: content changed '{sha256}0549defd0a7d6d840e3a69b82566505924cacbe2a79392970ec28cddc763949e' to '{sha256}823cbb079548be98b892725b133df610d0bff46b33e38b72d269306d32b73df2'

demo@85c63b50bfa3:~$ cat > file_rsapi_remove.pp <<EOF
$file = '/home/demo/puppet-file-rsapi-remove'
file       {$file: ensure => present} ->
file_rsapi {$file: ensure => absent}
EOF

demo@85c63b50bfa3:~$ bin/puppet apply --modulepath modules file_rsapi_remove.pp
Notice: Compiled catalog for 85c63b50bfa3 in environment production in 0.03 seconds
Notice: /Stage[main]/Main/File[/home/demo/puppet-file-rsapi-remove]/ensure: created
Notice: /Stage[main]/Main/File_rsapi[/home/demo/puppet-file-rsapi-remove]/ensure: undefined 'ensure' from 'present'
Notice: file_rsapi: Deleted: /home/demo/puppet-file-rsapi-remove

Thoughts

The apparent need for a CanonicalString masking as String makes it look like we are missing something. If the Resource-API only checked data types after canonicalization, we could make get return something simpler than CanonicalString to signal an already canonicalized value.

The Resource-API’s default demand to return all existing resources simplifies development when one had this plan anyway. The low-level version of doing so is often a combination of prefetch and instances.