Your gateway drug to automated infrastructure testing with Puppet
dylanratcliffe Merge pull request #179 from tequeter/master
fix spelling: weather vs. whether
Latest commit f3b7db3 Sep 8, 2018

README.md

Onceover

The gateway drug to automated infrastructure testing with Puppet

Onceover is a tool to automatically run basic tests on an entire Puppet controlrepo. It includes automatic parsing the Puppetfile, environment.conf and others in order to run both basic compilation tests and also full acceptance tests!

Table of Contents

Quick Start

Note: This assumes you are inside the root of your controlrepo.

Install the Gem:

gem install onceover

Set up your config:

onceover init

Run your spec tests!

onceover run spec

Hint: Don't forget you can use Bundler to install onceover by adding this to your gemfile:

gem 'onceover'

Here is an example using Bundler:

Install the Gem:

bundle install

Set up your config:

bundle exec onceover init

Run your spec tests!

bundle exec onceover run spec

Overview

This gem provides a toolset for testing Puppet Controlrepos (Repos used with r10k). The main purpose of this project is to provide a set of tools to help smooth out the process of setting up and running both spec and acceptance tests for a controlrepo. Due to the fact that controlrepos are fairly standardised in nature it seemed ridiculous that you would need to set up the same testing framework that we would normally use within a module for a controlrepo. This is because at this level we are normally just running very basic tests that cover a lot of code. It would also mean that we would need to essentially duplicated our Puppetfile into a .fixtures.yml file, along with a few other things.

This toolset requires some config before it can be used so definitely read that section before getting started.

Installation

gem install onceover

This gem can just be installed using gem install however I would recommend using Bundler to manage your gems.

Config Files

This project uses one main config file to determine what classes we should be testing and how, this is onceover.yaml. The onceover.yaml config file provides information about what classes to test when, however it needs more information than that:

If we are doing spec testing we need sets of facts to compile the puppet code against, these are stored in factsets. (A few are provided out of the box for you)

If we are doing acceptance testing then we need information about how to spin up VMs to do the testing on, these are configured in nodesets. (Once again these are auto-generated with onceover init)

onceover.yaml

spec/onceover.yaml (override with environment variable: ONCEOVER_YAML)

Hopefully this config file will be fairly self explanatory once you see it, but basically this is the place where we define what classes we want to test and the factsets/nodesets that we want to test them against. The config file must contain the following sections:

classes: A list (array) of classes that we want to test, usually this would be your roles, possibly profiles if you want. (If you don't know what roles and profiles are please READ THIS). To make life easier you can also specify one or many regular expressions in this section. A good one to start with would be /^role::/. Regular expressions are just strings that start and end with a forward slash.

nodes: The nodes that we want to test against. The nodes that we list here map directly to either a factset or a nodeset depending on whether we are running spec or acceptance tests respectively.

node_groups: The node_groups section is just for saving us some typing. Here we can set up groups of nodes which we can then refer to in our test matrix. We can create groups by simply specifying an array of servers to be in the group, or we can use the subtractive include/exclude syntax. The names used for the actual class_groups and node_groups must be unique.

class_groups: The class_groups section is much the same as the node_groups sections, except that it creates groups of classes, not groups of nodes (duh). All the same rules apply and you can also use the include/exclude syntax. This, like the classes section can also accept regular expressions. This means that as long as you name your roles according to a naming convention that includes the desired operating system, you should be able to define your class groups once and never touch them again. The names used for the actual class_groups and node_groups must be unique.

test_matrix: This where the action happens! This is the section where we set up which classes are going to be tested against which nodes. It should be an array of hashes with the following format:

  - {nodes_to_test}: # The name of a node or node group
      classes: '{classes_to_test}' # the name of a class or
      tests: '{all_tests|acceptance|spec}' # One of the three
      {valid_option}: {value} # Check the doco for available options

Why an array of hashes? Well, that is so that we can refer to the same node or node group twice, which we may want/need to do.

In the example below we have referred to centos6a and centos7b in all of our tests as they are in all_nodes, non_windows_servers and centos_severs. However we have left the more specific references to last. This is because entries in the test_matrix will override entries above them if applicable. Meaning that we are still only testing each class on the two Centos servers once (Because the gem does de-duplication before running the tests), but also making sure we run roles::frontend_webserver twice before checking for idempotency.

functions In this section we can add functions that we want to mock when running spec tests. Each function takes the following arguments:

  • type statement or rvalue
  • returns Optional: A value to return

before and after conditions We can set before and after blocks before each spec test. These are usually used when the functions to stub are conditional: stub functionx if the OS is windows, stub functiony if the fact java_installed is true. The facts are available through the node_facts hash and the trusted facts as trusted_facts.

before:
  - "Puppet::Util::Platform.stubs(:'windows?').returns(node_facts['kernel'] == 'windows')"

after:
  - "puts 'Test finished running'"

opts The opts section overrides defaults for the Onceover::Controlrepo class' opts hash.

opts:
  :facts_dirs:        # Remember: `opts` keys are symbols!
    - 'spec/factsets' # Limit factsets to files in this repository
  :debug: true        # set the `logger.level` to debug
opts:
  # profiles include a legacy module named `site::`
  :profile_regex: '^(profile|site)::'

  # factset filenames use the extension`.facts` instead of `.json`
  :facts_files:
    - 'spec/factsets/*.facts'

A full example:

classes:
  - 'roles::backend_dbserver'
  - 'roles::frontend_webserver'
  - 'roles::load_balancer'
  - 'roles::syd_f5_load_balancer'
  - 'roles::windows_server'
  - '/^role/'

nodes:
  - centos6a
  - centos7b
  - server2008r2a
  - ubuntu1404a
  - ubuntu1604a

node_groups:
  centos_severs:
    - centos6a
    - centos7b
  ubuntu_servers:
    - ubuntu1404a
    - ubuntu1604a
  non_windows_servers:
    include: 'all_nodes'
    exclude: 'server2008r2a'

class_groups:
  windows_roles:
    - 'roles::windows_server'
    - 'roles::backend_dbserver'
    - '/^roles::win/'
  non_windows_roles:
    include: 'all_classes'
    exclude: 'windows_roles'

test_matrix:
  - all_nodes:
      classes: 'all_classes'
      tests: 'spec'
  - non_windows_servers:
      classes: 'non_windows_roles'
  - ubuntu_servers:
      classes: 'all_classes'
      tests: 'all_tests'
  - centos_severs:
      classes: 'roles::frontend_webserver'
      tests: 'acceptance'
      runs_before_idempotency: 2
      tags:
        - 'frontend'

functions:
  query_resources:
    type: rvalue
    returns: []

opts:
  :facts_dirs:
    - spec/factsets

Include/Exclude syntax: This can be used with either node_groups or class_groups and allows us to save some time by using existing groups to create new ones e.g.

node_groups:
  windows_nodes: # This has to be defined first
    - sevrer2008r2
    - server2012r2
  non_windows:
    include: 'all_nodes' # Start with all nodes
    exclude: 'windows_nodes' # Then remove the windows ones from that list

It's important to note that in order to reference a group using the include/exclude syntax is has to have been defined already i.e. it has to come above the group that references it (Makes sense right?)

Optional test parameters

check_idempotency Default: true

Whether or not to check that puppet will be idempotent (Acceptance testing only)

runs_before_idempotency Default: 1

The number of runs to try before checking that it is idempotent. Required for some things that require restarts of the server or restarts of puppet. (Acceptance testing only)

tags Default: nil

One or many tags that tests in this group should be tagged with. This allows you to run only certain tests using the --tags command line parameter. NOTE: Custom spec tests will always be run as they are not subject to tags

factsets

This gem comes with a few pre-canned factsets. These are listed under the nodes sections of onceover.yaml when you run onceover init. You can also add your own factsets by putting them in:

spec/factsets/*.json

Factsets are used by the controlrepo gem to generate spec tests, which compile a given class against a certain set of facts. To create these factsets all we need to do is log onto a real machine that has puppet installed and run:

puppet facts

Which will give raw json output of every fact which puppet knows about. Usually I would recommend running this on each of the types of machines that you run in your infrastructure so that you have good coverage. To make life easier you might want to direct it into a file instead of copying it from the command line:

puppet facts > fact_set_name.json

Once we have our factset all we need to do is copy it into spec/factsets/ inside our controlrepo and commit it to version control. Factsets are named based on their filename, not the name of the server they came from (Although you can, if you want). i.e the following factset file:

spec/factsets/server2008r2.json

Would map to a node named server2008r2 in onceover.yaml

Trusted Facts

You can add trusted facts to the nodesets by creating a new section called trusted:

{
  "name": "node.puppetlabs.net",
  "trusted": {
    "pp_role": "agent",
    "pp_datacenter": "puppet",
  },
  "values": {
    "aio_agent_build": "1.10.4",
    "aio_agent_version": "1.10.4",
    "architecture": "x86_64",

Notice that the extensions part is implied. The first fact in that example translates to $trusted['extensions']['pp_role'] in Puppet code.

nodesets

spec/acceptance/nodesets/onceover-nodes.yml

Nodesets are used when running acceptance tests. They instruct the onceover gem how to spin up virtual machines to run the code on. Actually, that's a lie... What's really happening with nodesets is that we are using Beaker to spin up the machines and then a combination of Beaker and RSpec to test them. But you don't need to worry about that too much. Due to the fact that we are using beaker to do the heavy lifting here the nodeset files follow the same format they would for normal Beaker tests, which at the time of writing supports the following hypervisors:

Before we configure a hypervisor to spin up a node however, we have to make sure that it can clone from a machine which is ready. The onceover gem requires it's VMs to have puppet pre-installed. It doesn't matter what version of puppet, as long as it is on the PATH and the type setting is configured correctly i.e.

type: AIO # For machines that have the all-in-one agent installed (>=4.0 or >=2015.2)
# OR
type: pe # For puppet enterprise agents <2015.2
# OR
type: foss # For open source puppet <4.0

Here is an example of a nodeset file that you can use yourselves. It uses freely available Vagrant boxes from Puppet and Virtualbox as the Vagrant provider. (onceover init will generate most of this for you)

HOSTS:
  centos6a:
    roles:
      - agent
    type: aio
    platform: el-6-64
    box: puppetlabs/centos-6.6-64-puppet
    box_url: https://atlas.hashicorp.com/puppetlabs/boxes/centos-6.6-64-puppet
    hypervisor: vagrant_virtualbox
  centos7b:
    roles:
      - agent
    type: aio
    platform: el-7-64
    box: puppetlabs/centos-7.0-64-puppet
    box_url: https://atlas.hashicorp.com/puppetlabs/boxes/centos-7.0-64-puppet
    hypervisor: vagrant_virtualbox
  ubuntu1204:
    roles:
      - agent
    type: aio
    platform: ubuntu-12.04-32
    box: puppetlabs/ubuntu-12.04-32-puppet
    box_url: https://atlas.hashicorp.com/puppetlabs/boxes/ubuntu-12.04-32-puppet
    hypervisor: vagrant_virtualbox
  debian78:
    roles:
      - agent
    type: aio
    platform: debian-7.8-64
    box: puppetlabs/debian-7.8-64-puppet
    box_url: https://atlas.hashicorp.com/puppetlabs/boxes/debian-7.8-64-puppet
    hypervisor: vagrant_virtualbox

Hiera Data

If you have hiera data inside your controlrepo (or somewhere else) Onceover can be configured to use it. It is however worth noting the the hiera.yaml file that you currently use may not be applicable for testing right away. For example; if you are using hiera-eyaml I recommend creating a hiera.yaml purely for testing that simply uses the yaml backend, meaning that you don't need to provide the private keys to the testing machines.

It is also worth noting that any hiera hierarchies that are based on custom facts will not work unless those facts are part of your factsets. Trusted facts will also not work at all as the catalogs are being compiled without the node's certificate. In these instances it may be worth creating a hierarchy level that simply includes dummy data for testing purposes in order to avoid hiera lookup errors.

Creating the config file

If your hiera.yaml is version 4 or 5 and lives in the root of the controlrepo (as it should), Onceover will pick this up automatically. If you would like to make changes to this file for testing purposes, create a copy under spec/hiera.yaml. Onceover will use this version of the hiera config file first if it exists.

Setting the datadir

Hiera Version Config File Location Required datadir
3 spec folder relative to the root of the repo e.g. data
4 deprecated Root of repo relative to the root of the repo e.g. data
4 deprecated spec folder relative to the spec folder e.g. ../data
5 Root of repo relative to the root of the repo e.g. data
5 spec folder relative to the spec folder e.g. ../data

hiera-eyaml

If you are using the hiera-eyaml backend there are some modifications that you will need to make in order to ensure that things actually work. Remember that when onceover compiles catalogs it is actually using hiera with your config file to do the lookups on the host that is running the tests, meaning that the hiera-eyaml gem will need to be present (put it in your Gemfile), as will the keys in the correct location, otherwise hiera will fail to load them. This is really not a great situation as you don't want to be distributing your private keys

Recommended Solution: I recommend that if you are using hiera-eyaml (which you probably should be) that you do the following:

  1. Duplicate your hiera.yaml file so that there is a copy in the spec/ directory
  2. Change the datadir setting as described above
  3. Remove the eyaml backend entirely and just use the base yaml backend. For hiera 5 this will look like:
---
version: 5
defaults:
  datadir: "../data"
  data_hash: yaml_data

This means that for testing, hiera will just return the encrypted string for anything that is encrypted using eyaml. This usually isn't a problem for catalog compilation and will allow tests to pass.

Spec testing

Once you have your onceover.yaml and factsets set up you are ready to go with spec testing.

To run the tests:

onceover run spec

This will do the following things:

  1. Create a temporary directory under .onceover
  2. Clone all repos in the Puppetfile into the temporary directory
  3. Generate tests that use rspec-puppet
  4. Run the tests

Adding your own spec tests

When using this gem adding your own spec tests is exactly the same as if you were to add them to a module, simply create them under spec/{classes,defines,etc.} in the Controlrepo and they will be run like normal, along with all of the it { should compile } tests.

Exposing Puppet output

If you want to see Puppet's output, you can set the SHOW_PUPPET_OUTPUT environment variable to true, eg:

SHOW_PUPPET_OUTPUT=true onceover run spec

Acceptance testing

Acceptance testing works in much the same way as spec testing except that it requires a nodeset file along with onceover.yaml

To run the tests:

onceover run acceptance

This will do the following things:

  1. Create a temporary directory under .onceover
  2. Clone all repos in the Puppetfile into the temporary directory
  3. Generate tests that use RSpec and Beaker
  4. Run the tests, each test consists of: - Spin up the VM - Copy over the code - Run puppet and catch any errors - Run puppet again to catch anything that might not be idempotent - Destroy the VM

Using workarounds

There may be situations where you cannot test everything that is in your puppet code, some common reasons for this include:

  • Code is destined for a Puppet Master but the VM image is not a Puppet Master which means we can't restart certain services etc.
  • A file is being pulled from somewhere that is only accessible in production
  • Something is trying to connect to something else that does not exist

Fear not! There is a solution for this, it's also a good way to practice writing nasty puppet code. For this exact purpose I have added the ability for onceover to include extra bits of code in the tests to fix things like this. All you need to do is put a file/s containing puppet code here:

spec/pre_conditions/*.pp

What this will do is it will take any puppet code from any files it finds in that directory and have it executed alongside the code that you are actually testing. For example if we are testing some code that notifies the pe-puppetserver service, but are not managing that service in our code because it is managed by the PE module that ships with Puppet Enterprise the following code will fail:

file { '/etc/puppetlabs/puppet/puppet.conf':
  ensure  => file,
  content => '#nothing',
  notify  => Service['pe-puppetserver'], # This will fail without the PE module!
}

To fix this we can add the service to the pre_conditions to make sure that our catalogs can compile e.g.

# spec/pre_conditions/services.pp
service { 'pe-puppetserver':
  ensure => 'running',
}

You can also mock out defined resources or types that you cannot gain access to easily, such as puppet_enterprise::mcollective::client:

  define puppet_enterprise::mcollective::client (
    $activemq_brokers,
    $logfile     = '/var/log',
    $create_user = true,
  ) {

  }

or

  define pe_ini_setting (
    $ensure  = present,
    $path,
    $section,
    $setting,
    $value,
  ) {
  }

However this is going to pose an issue when we get to acceptance testing. Due to the fact that acceptance tests actually run the code, not just tries to compile a catalog, it will not be able to find the 'pe-pupetserver' service and will fail. One way to get around this is to use some of the optional parameters to the service resource e.g.

# We are not going to actually have this service anywhere on our servers but
# our code needs to refresh it. This is to trick puppet into doing nothing
service { 'pe-puppetserver':
  ensure     => 'running',
  enable     => false,
  hasrestart => false, # Force Puppet to use start and stop to restart
  start      => 'echo "Start"', # This will always exit 0
  stop       => 'echo "Stop"', # This will also always exit 0
  hasstatus  => false, # Force puppet to use our command for status
  status     => 'echo "Status"', # This will always exit 0 and therefore Puppet will think the service is running
  provider   => 'base',
}

Here we are specifying custom commands to run for starting, stopping and checking the status of a service. We know what the exit codes of these commands are going to be so we know what puppet will think the service is doing because we have read the documentation. If there are things other than services you need to check then I would recommend checking the documentation to see if you can mock things like we have here. Alternatively you might need to create specific VM images that are pre-prepared.

Resource collectors are likely to come in handy here too. They allow you to override values of resources that match given criteria. This way we can override things for the sake of testing without having to change the code.

NOTE: If you need to run some pre_conditions during acceptance tests but not spec tests or vice versa you can check the status of the $controlrepo_accpetance variable. It will be true when run as an acceptance test and undef otherwise. If you want to limit pre_conditions to only certain nodes just use conditional logic based on facts like you normally would.

NOTE: If you want to access the class or factset that onceover is running against just use the $onceover_class and $onceover_node variables respectively.

Extra Tooling

Plugins

Onceover now allows for plugins. To use a plugin simply install a gem with a name that starts with onceover- and onceover will activate it.

Useful plugins:

If you want to write your own plugin, take a look at onceover-helloworld to help you get started.

Inspecting and updating the Puppetfile

Onceover comes with some extra commands for interacting with the Puppetfile in useful ways. These are:

onceover show puppetfile

This will display all the current versions of all modules that are in the Puppetfile alongside the latest versions and whether or not they are out of date. This is a useful took for making sure your modules don't get too stale.

onceover update puppetfile

This takes your Puppetfile and actually modifies all of the module versions in there to the latest versions and saves the file. This is useful for setting up automated Puppetfile updates, just get Jenkins or Bamboo to:

  1. Check out the Controlrepo
  2. Run onceover to get a passing baseline
  3. Update the Puppetfile with the latest versions of all modules
  4. Run Onceover agan
  5. Create a pull request if all tests pass

Overriding Onceover's Templates

Onceover uses templates to create a bunch of files in the .onceover directory, these templates can be modified if required. To do this create your own custom template with the same name os the original in the spec/templates/ directory and it will be used in preference to the default template. e.g. spec/templates/spec_helper.rb.erb

Accessing Onceover in a traditional RSpec test

If you would like to use onceover.yaml to manage which tests you want to run, but want more than just it { should_compile } tests to be run you can write you own as follows:

# spec/classes/role_spec.rb
require 'spec_helper'
require 'onceover/controlrepo'
require 'helpers/shared_examples'

Onceover::Controlrepo.new.spec_tests do |class_name,node_name,facts,pre_conditions|
  describe class_name do
    context "on #{node_name}" do
      let(:facts) { facts }
      let(:pre_condition) { pre_conditions }

      it_behaves_like 'soe'
    end
  end
end

This will use the soe shared example on all of the tests that are configured in your onceover.yaml including any pre_conditions that you have set up.

Note: Onceover will automatically run any extra Rspec tests that it finds in the normal directories spec/{classes,defines,unit,functions,hosts,integration,types} so you can easily use auto-generated spec tests in conjunction with your own Rspec tests.

Accessing fact sets in a traditional RSpec test

We can access all of our fact sets using Onceover::Controlrepo.facts. Normally it would be implemented something like this:

Onceover::Controlrepo.facts.each do |facts|
  context "on #{facts['fqdn']}" do
    let(:facts) { facts }
    it { should compile }
  end
end

Other (possibly less useful) methods

The following code will test all roles that onceover can find (ignoring the ones configured in onceover.yaml) on all nodes in native rspec:

require 'spec_helper'
require 'onceover/controlrepo'
Onceover::Controlrepo.roles.each do |role|
  describe role do
    Onceover::Controlrepo.facts.each do |facts|
      context "on #{facts['fqdn']}" do
        let(:facts) { facts }
        it { should compile }
      end
    end
  end
end

This will iterate over each role in the controlrepo and test that it compiles with each set of facts.

The same can also be done with profiles just by using the profiles method instead:

require 'spec_helper'
require 'onceover'
Onceover::Controlrepo.profiles.each do |profile|
  describe profile do
    Onceover::Controlrepo.facts.each do |facts|
      context "on #{facts['fqdn']}" do
        let(:facts) { facts }
        it { should compile }
      end
    end
  end
end

It is not limited to just doing simple "It should compile" tests. You can put any tests you want in here.

Also since the profiles, roles and facts methods simply return arrays, you can iterate over them however you would like i.e. you could write a different set of tests for each profile and then just use the facts method to run those tests on every fact set.

Filtering

You can also filter your fact sets based on the value of any fact, including structured facts. (It will drill down into nested hashes and match those too, it's not just a dumb equality match)

Just pass a hash to the facts method and it will return only the fact sets with facts that match the hash e.g. Testing a certain profile on against only your Windows fact sets:

require 'spec_helper'
require 'onceover'

describe 'profile::windows_appserver' do
  Onceover::Controlrepo.facts({
    'kernel' => 'windows'
    }).each do |facts|
    context "on #{facts['fqdn']}" do
      let(:facts) { facts }
      it { should compile }
    end
  end
end

Extra Configuration

You can modify the regexes that the gem uses to filter classes that it finds into roles and profiles. Just set up a Controlrepo object and pass regexes to the below settings.

repo = Onceover::Controlrepo.new()
repo.role_regex = /.*/ # Tells the class how to find roles, will be applied to repo.classes
repo.profile_regex = /.*/ # Tells the class how to find profiles, will be applied to repo.classes

Note that you will need to call the roles and profiles methods on the object you just instantiated, not the main class e.g. repo.roles not Onceover::Controlrepo.roles

Rake tasks

I have included a couple of little rake tasks to help get you started with testing your control repos. Set them up by adding this to your Rakefile

require 'onceover/rake_tasks'

The tasks are as follows:

generate_fixtures

bundle exec rake generate_fixtures

This task will go though your Puppetfile, grab all of the modules in there and convert them into a .fixtures.yml file. (You only need this if you are writing your own custom spec tests) It will also take the environment.conf file into account, check to see if you have any relative pathed directories and also include them into the .fixtures.yml as symlinks. e.g. If your files look like this:

Puppetfile

forge "http://forgeapi.puppetlabs.com"

# Modules from the Puppet Forge
mod "puppetlabs/stdlib", "4.6.0"
mod "puppetlabs/apache", "1.5.0"

environment.conf

modulepath = site:modules:$basemodulepath
environment_timeout = 0

Then the .fixtures.yml file that this rake task will create will look like this:

---
fixtures:
  symlinks:
    profiles: site/profiles
    roles: site/roles
  forge_modules:
    stdlib:
      repo: puppetlabs/stdlib
      ref: 4.6.0
    apache:
      repo: puppetlabs/apache
      ref: 1.5.0

Notice that the symlinks are not the ones that we provided in environment.conf? This is because the rake task will go into each of directories, find the modules and create a symlink for each of them (This is what rspec expects).

Developing Onceover

Install gem dependencies:

bundle install

Clone the submodules

git submodule init && git submodule update --recursive

Execute tests

bundle exec rake

Contributors

Cheers to all of those who helped out:

  • @jessereynolds
  • @op-ct
  • @GeoffWilliams
  • @beergeek
  • @jairojunior
  • @natemccurdy
  • @aardvark
  • @Mandos
  • @Nekototori
  • @LMacchi
  • @tabakhase
  • @binford2k
  • @raphink