geoffwilliams@home:~$

Retrofitting Testing To Your Puppet Modules

Outdated

This article was written for an old version of Puppet Enterprise and exists for reference purposes only.

Testing is great, everyone should test from day one, but… not everyone does, for one reason or another. Fortunately, there’s a really easy way to add testing to an existing module called Puppet-Retrospec.

Why the sudden urgency?

Testing Puppet Modules is always recommended but often gets pushed to the bottom of the todo list at a lot of customer sites. Right now, we’re starting to see customers migrate away from their existing Puppet 3.x servers to PE 201x and the Puppet 4 Parser.

The Puppet 4 Parser brings some truly great new features that customers have been asking for for a long time such as:

Unfortunately, these changes come with the caveat that your Puppet Code needs to be less sloppy for want of a better word. Specifically:

  • Strings must be quoted
  • Octal numbers (file permissions) must be quoted
  • The empty string '' now equates to false (as in most other programming languages)

To prevent on-upgrade failures, Puppet customers are now looking for ways to prepare code for the future so that it works as expected with the new parser.

How to prepare for PE 201x/Puppet 4 parser

One way of preparing for the future is to invest the time in getting the Catalog Preview module from Puppet to inspect how entire catalogues will be built with the Puppet 4 Parser vs Classic puppet. This is great for obtaining a birds-eye view of how an upgrade would affect nodes at the catalogue level.

This is all well and good but you can make this job a lot easier by making sure that the modules your using are uplifted before this process takes place and this is the focus of this blog post.

What is Puppet-Retrospec

Puppet-Retrospec takes an existing module with no testing and adds RSpec Puppet testing with a single command. Pretty cool eh!

Where should I run my tests?

Never run your tests on the Puppet Master!

This is not the place to go experimenting with RubyGems and CPU bound testing. Be warned that changing the RubyGems Puppet is using can break your Puppet installation.

The best place to run your tests is a completely separate and isolated computer such as:

  • Your laptop
  • A VM running on your laptop (by far the best place to setup your testing)
  • A VM running in the cloud
  • A dedicated CI server

For the purposes of this blog post, I’ll run the tests from a Centos 7.2 VM running on my laptop (using Vagrant).

Preparing your system

With a freshly booted minimal VM, I ran the following commands as root to install the Puppet-Retrospec software and its dependences.

yum install ruby ruby-devel git
gem install puppet-retrospec bundler facter

Testcase

To demonstrate how RetroSpec works, I’ll be using a made up module called mod_foo which has the following initial directory structure:

mod_foo/
`-- manifests
    `-- init.pp

The init.pp file contains a single, badly written class:

class mod_foo($do_stuff='') {
  file { /var/foo:
    ensure => directory,
    owner  => root,
    group  => root,
    mode   => 0755,
  }

  if $do_stuff {
    service { 'foo':
      ensure => running,
      enable => true,
    }
  }
}

Lets test!

With your testing VM setup, your ready to generate and run some tests. The rest of the blog post assumes your current working directory is the Puppet Module under test. In my case, the module is saved my VM’s /vagrant/mod_foo directory, so that it is shared with the host computer.

Step 1: (Try to…) Generate tests

We try to generate an initial set of tests by running the command:

retrospec puppet --enable-future-parser

With the above init.pp file, running this command will result in an error:

Manifest file: /vagrant/mod_foo/manifests/init.pp has parser errors, please fix and re-check using                                               
 puppet parser validate /home/vagrant/mod_foo/manifests/init.pp

Step 2: Fix the syntax errors

Believe it or not, this is good and is exactly why we are using RetroSpec in the first place!

The next step here is to find out exactly what the syntax errors are, using a the suggested command. This will tell us exactly what the errors are:

bundle exec puppet parser validate /home/vagrant/mod_foo/manifests/init.pp

...

Error: Could not parse for environment production: Syntax error at 'foo' at /home/vagrant/mod_foo/manifests/init.pp:2:15

If running the suggested puppet parser validate command fails, see the troubleshooting section for a solution.

Opening up the init.pp file and looking at line 2, we see:

file { /var/foo:

…Whoops! The filename wasn’t quoted. This isn’t unusual in older Puppet modules. We can fix this by putting single ' or double " quotes around values. Lets fix this line and also have a look through the rest of the file as there are a few lines with this problem in the file.

The fixed version of init.pp looks like this:

class mod_foo($do_stuff='') {
  file { '/var/foo':
    ensure => directory,
    owner  => 'root',
    group  => 'root',
    mode   => '0755',
  }

  if $do_stuff {
    service { 'foo':
      ensure => running,
      enable => true,
    }
  }
}

Once puppet parser validate gives no output, the code is valid and we can try to generate the tests again.

Step 3: Generate the tests (again…)

With our newly valid Puppet Code, we can try and generate our tests again:

retrospec puppet --enable-future-parser

This time, we should see a bunch of new files being generated:

Successfully ran hook: /root/.retrospec/repos/retrospec-puppet-templates/clone-hook

Successfully ran hook: /root/.retrospec/repos/retrospec-puppet-templates/pre-hook

 + /vagrant/mod_foo/.bundle/
 + /vagrant/mod_foo/.bundle/config
 + /vagrant/mod_foo/.fixtures.yml
 + /vagrant/mod_foo/.git/
 + /vagrant/mod_foo/.git/hooks/
 + /vagrant/mod_foo/.git/hooks/pre-commit
 + /vagrant/mod_foo/.gitignore
 + /vagrant/mod_foo/.puppet-lint.rc
 + /vagrant/mod_foo/.travis.yml
 + /vagrant/mod_foo/DEVELOPMENT.md
 + /vagrant/mod_foo/Gemfile
 + /vagrant/mod_foo/Rakefile
 + /vagrant/mod_foo/Vagrantfile
 + /vagrant/mod_foo/files/
 + /vagrant/mod_foo/files/.gitkeep
 + /vagrant/mod_foo/spec/
 + /vagrant/mod_foo/spec/acceptance/
 + /vagrant/mod_foo/spec/shared_contexts.rb
 + /vagrant/mod_foo/spec/spec_helper.rb
 + /vagrant/mod_foo/templates/
 + /vagrant/mod_foo/templates/.gitkeep
 + /vagrant/mod_foo/tests/
 + /vagrant/mod_foo/tests/.gitkeep
 + /vagrant/mod_foo/mod_foo_schema.yaml
 + /vagrant/mod_foo/metadata.json
 + /vagrant/mod_foo/spec/classes/
 + /vagrant/mod_foo/spec/classes/mod_foo_spec.rb
Successfully ran hook: /root/.retrospec/repos/retrospec-puppet-templates/post-hook

What’s happened here is that Puppet-Retrospec has inspected the module, installed a skeleton test infrastructure and wrote tests modeling the resources in init.pp. If you had other Puppet Code in the module it would attempt to generate tests for that too… Cool!

Step 4: Run the tests

To run the tests we generated, we first need to install some RubyGems using Bundler:

bundle install --without integration development

With this step completed, we can now run our tests whenever we want with the following command:

bundle exec rake spec

This command should give output similar to the following:

[root@pe-puppet mod_foo]# bundle exec rake spec
Cloning into 'spec/fixtures/modules/stdlib'...
remote: Counting objects: 8069, done.
remote: Compressing objects: 100% (6/6), done.
remote: Total 8069 (delta 1), reused 0 (delta 0), pack-reused 8063
Receiving objects: 100% (8069/8069), 1.67 MiB | 414.00 KiB/s, done.
Resolving deltas: 100% (3749/3749), done.
HEAD is now at da11903 Merge pull request #299 from apenney/432-release
/usr/bin/ruby -I/usr/local/share/gems/gems/rspec-core-3.4.4/lib:/usr/local/share/gems/gems/rspec-support-3.4.1/lib /usr/local/share/gems/gems/rspec-core-3.4.4/exe/rspec --pattern spec/\{classes,defines,unit,functions,hosts,integration,types\}/\*\*/\*_spec.rb --color

mod_foo
  should compile into a catalogue without dependency cycles

Deprecation Warnings:

Using `stub` from rspec-mocks' old `:should` syntax without explicitly enabling the syntax is deprecated. Use the new `:expect` syntax or explicitly enable `:should` instead. Called from /usr/local/share/gems/bundler/gems/hiera-puppet-helper-155f132c0b22/lib/hiera-puppet-helper.rb:23:in `block (2 levels) in <top (required)>'.


If you need more of the backtrace for any of these deprecations to
identify where to make the necessary changes, you can configure
`config.raise_errors_for_deprecations!`, and it will turn the
deprecation warnings into errors, giving you the full backtrace.

1 deprecation warning total

Finished in 1.92 seconds (files took 0.6783 seconds to load)
1 example, 0 failures

There’s a few warnings in there which you could probably fix yourself if your interested, failing that a future version of Puppet-Retrospec may generate tests that fix this.

The important line here though is the last one:

1 example, 0 failures

Which means that our tests pass! - Great!

Not so fast!

Our fixed module can go straight into production now right? No! Have another look at the class definition. See the lines:

class mod_foo($do_stuff='') {

and

if $do_stuff {

One big gotcha with the Puppet 4 parser is that whereas '' evaluated false in older versions of Puppet, in the Puppet 4 Parser this same code evaluates to true!

If your using the empty string '' as a placeholder for False in your Puppet Modules then this is something you will want to fix before any rollout.

In my case, the fix is simple: Since I’m not interested in any string value for the do_stuff variable, I can just change it to be false, giving me the fixed class definition:

class mod_foo($do_stuff=false) {

Rerunning my tests again indicates they are still passing - great!

At this point I’m happy that the code I’ve written is syntax error free and creates the same resources it used to but I’d still recommend:

  • Creating and running Puppet smoke tests using the version of Puppet Enterprise your planning to deploy to (in another Vagrant VM!)
  • Considering whether its worth writing additional tests to the ones generated by Puppet-Retrospec
  • If your feeling really brave, you could have a look at doing acceptance testing (testing on real systems) using Test Kitchen or Beaker. Be warned that neither of these systems are particularly easy to use for doing testing on Puppet code (although it is possible…)

When your happy that your code has been tested enough, you can deploy it to a test Puppet Master. Your final guard against misconfiguration would be to limit initial deployment to a handful of test nodes.

Once the uplifted Puppet Module is proved working in the real world it can be considered for a wider rollout, perhaps along with the Catalog Preview tool mentioned earlier.

What are the caveats of using Puppet-Retrospec?

The main gotcha is that Puppet-Retrospec makes sure your code performs as coded which means that if you have errors in your Puppet code you will automatically have tests to ensure those errors are forever present in the generated Puppet catalogue!

There’s very little that’s going to beat logical hand written, commented test cases but Puppet-Retrospec is great for generating the required test infrastructure in old modules and is a big help in getting customers across the line with their upgrades.

You can always edit the tests it generates to make them more useful too.

What about new modules, should I use Puppet-Retrospec on them too?

You can certainly add RetroSpec to new modules if you wish, however, you may instead want to look at the puppet module generate subcommand which can be used to create a new blank module complete with a basic set of tests.

From here you just need to create a fixtures.yml file as described in The Next Generation of Puppet Module Testing and you can then start writing a complete set of tests using RSpec Puppet.

Troubleshooting

I added some cool Puppet 4 code and now my tests fail!

Edit the file spec/spec_helper.rb and uncomment the line that says

ENV['FUTURE_PARSER'] = 'yes'

I can’t run RetroSpec Puppet anymore!

If you run the retrospec puppet command and all you see is the message

Successfully ran hook: /root/.retrospec/repos/retrospec-puppet-templates/clone-hook

Followed by no tests being generated, then its probably because you have a newer version of the puppet RubyGem installed then Puppet-Retrospec can handle. You can prove this by running the command:

gem list | grep puppet

If you see any versions of puppet higher then version 3.8 (multiple versions are allowed), you need to remove them so that RetroSpec will use its vendored version of Puppet, eg:

gem uninstall puppet

You can remove all of the Puppet gems and next time you bundle install to run your tests, the correct version (3.8.7) will hopefully be installed.

I can’t get the bundle to install!

If your finding that running bundle install in your module directory cannot be made to work, make sure your not installing the gems in the development or integration groups (specifically the guard-rake RubyGem). For some reason this RubyGem is not installable with Ruby 2.0 (RHEL 7), Ruby 2.1 (Ubuntu 15.10) or Ruby 2.2.2 (RHEL 7 + Software Collections). I think I did manage to get it to install under Ruby 2.3 (Ubuntu 16.04) but then Puppet wouldn’t work! Basically it needs ruby 2.2.4 which you can install through rbenv/RVM… or you could just not install it and move on with your life:

bundle install --without integration development

I don’t have a puppet command so I can’t validate anything!

You need to install the puppet gem to gain access to this command. If you’ve run through the retrospec puppet ... bundle install ... bundle exec rake spec cycle, then you will already have access but if this is your first time, you will need to install it manually:

gem install puppet -v 3.8.7

Make sure to install ONLY this version of Puppet or you will break Puppet-Retrospec

My Puppet Code depends on other modules and my tests fail!

You probably just need to manually add any modules you depend on to the .fixtures.yml file, see The Next Generation of Puppet Module Testing for more info.

Post comment

Markdown is allowed, HTML is not. All comments are moderated.