ChefSpec is a unit testing framework for testing Chef cookbooks. ChefSpec makes it easy to write examples and get fast feedback on cookbook changes without the need for virtual machines or cloud servers.
ChefSpec runs your cookbooks locally while skipping making actual changes. This has two primary benefits:
- It's really fast!
- Your tests can vary node attributes, operating systems, and other system data to assert behavior under varying conditions.
- ChefSpec requires Ruby 2.5 or later and Chef 15 or later!
- This documentation corresponds to the main branch, which may be unreleased. Please check the README of the latest git tag or the gem's source for your version's documentation!
ChefSpec aims to maintain compatibility with at least the two most recent minor versions of Chef. If you are running an older version of Chef it may work, or you will need to run an older version of ChefSpec.
As a general rule, if it is tested in the Travis CI matrix, it is a supported version.
As mentioned before, ChefSpec is built for speed. In order to run your tests as quickly as possible (and to allow running tests on your workstation), ChefSpec runs your recipe code with all the resource actions disabled. This means that ChefSpec excels at testing complex logic in a cookbook, but can't actually tell you if a cookbook is doing the right thing. Integration testing is provided by the Test Kitchen project, and for most simple cookbooks without much logic in them we recommend you start off with integration tests and only return to ChefSpec and unit tests as your code gets more complicated.
There are two common "units" of code in Chef cookbooks, custom resources and recipes. If you find yourself with a lot of recipes that are so complex they require unit tests, consider if they can be refactored as custom resources.
If you have have a cookbook with a custom resource resources/greet.rb
like:
resource_name :mycookbook_greet
property :greeting, String, default: 'Hello'
action :run do
log "#{new_resource.greeting} world"
end
You can test that resource by creating a spec file spec/greet_spec.rb
:
# Load ChefSpec and put our test into ChefSpec mode.
require 'chefspec'
# Describing our custom resource.
describe 'mycookbook_greet' do
# Normally ChefSpec skips running resources, but for this test we want to
# actually run this one custom resource.
step_into :mycookbook_greet
# Nothing in this test is platform-specific, so use the latest Ubuntu for
# simulated data.
platform 'ubuntu'
# Create an example group for testing the resource defaults.
context 'with the default greeting' do
# Set the subject of this example group to a snippet of recipe code calling
# our custom resource.
recipe do
mycookbook_greet 'test'
end
# Confirm that the resources created by our custom resource's action are
# correct. ChefSpec matchers all take the form `action_type(name)`.
it { is_expected.to write_log('Hello world') }
end
# Create a second example group to test a different block of recipe code.
context 'with a custom greeting' do
# This time our test recipe code sets a property on the custom resource.
recipe do
mycookbook_greet 'test' do
greeting 'Bonjour'
end
end
# Use the same kind of matcher as before to confirm the action worked.
it { is_expected.to write_log('Bonjour world') }
end
end
And then run your test using chef exec rspec
.
As a general rule of thumb, only very complex recipes benefit from ChefSpec unit
tests. If you find yourself writing a lot of recipe unit tests, consider converting
the recipes to custom resources instead. For the sake of example we'll use a
simple recipe, recipes/farewell.rb
:
log "#{node["mycookbook"]["farewell"]} world"
You can test that recipe by creating a spec file spec/farewell_spec.rb
:
# Load ChefSpec and put our test into ChefSpec mode.
require 'chefspec'
# Describing our recipe. The group name should be the recipe string as you would
# use it with include_recipe.
describe 'mycookbook::farewell' do
# Nothing in this test is platform-specific, so use the latest Ubuntu for
# simulated data.
platform 'ubuntu'
# Create an example group for testing the recipe defaults.
context 'with default attributes' do
# Since there was no `recipe do .. end` block here, the default subject is
# recipe we named in the `describe`. ChefSpec matchers all take the form
# `action_type(name)`.
it { is_expected.to write_log('Goodbye world') }
end
# Create a second example group to test with attributes.
context 'with a custom farewell' do
# Set an override attribute for this group.
override_attributes['mycookbook']['farewell'] = 'Adios'
# Use the same kind of matcher as before to confirm the recipe worked.
it { is_expected.to write_log('Adios world') }
end
end
If your cookbook depends on other cookbooks, you must ensure ChefSpec knows how
to fetch those dependencies. If you use a monorepo-style layout with all your
cookbooks in a single cookbooks/
folder, you don't need to do anything.
If you are using Berkshelf, require 'chefspec/berkshelf'
in your spec file (or spec_helper.rb
):
require 'chefspec'
require 'chefspec/berkshelf'
If you are using a Policyfile, require 'chefspec/policyfile'
in you spec file (or spec_helper.rb
):
require 'chefspec'
require 'chefspec/policyfile'
Your Policyfile.rb
should look something like this:
# The policy name is ignored but you need to specify one.
name 'my_cookbook'
# Pull dependent cookbooks from https://supermarket.chef.io/
default_source :supermarket
# The run list is also ignored by ChefSpec but you need to specify one.
run_list 'my_cookbook::default'
# The name here must match the name in metadata.rb.
cookbook 'my_cookbook', path: '.'
ChefSpec is an RSpec library, so if you're already familiar with RSpec you can
use all the normal spec-y goodness to which you are accustomed. The usual structure
of an RSpec test file is a file named like spec/something_spec.rb
containing:
require 'chefspec'
describe 'resource name or recipe' do
# Some configuration for everything inside this `describe`.
platform 'redhat', '7'
default_attributes['value'] = 1
context 'when some condition' do
# Some configuration that only applies to this `context`.
default_attributes['value'] = 2
# `matcher` is some matcher function which we'll cover below.
it { expect(value).to matcher }
# There is a special value you can expect things on called `subject`, which
# is the main thing being tested.
it { expect(subject).to matcher }
# And if prefer it for readability, `expect(subject)` can be written as `is_expected`.
it { is_expected.to matcher }
end
context 'when some other condition' do
# Repeat as needed.
end
end
The primary matcher used with ChefSpec are resource matchers:
it { expect(chef_run).to ACTION_RESOURCE('NAME') }
# Or equivalently.
it { is_expected.to ACTION_RESOURCE('NAME') }
This checks that a resource like RESOURCE 'NAME'
would have run the specified
action if the cookbook was executing normally. You can also test for specific
property values:
it { is_expected.to create_user('asmithee').with(uid: 512, gid: 45) }
# You can also use other RSpec matchers to create a "compound matcher". Check
# RSpec documentation for a full reference on the built-in matchers.
it { is_expected.to install_package('myapp').with(version: starts_with("3.")) }
For the common case of testing that a file is rendered to disk via either a
template
, file
, or cookbook_file
resource, you can use a render_file
matcher:
it { is_expected.to render_file('/etc/myapp.conf') }
# You can check for specific content in the file.
it { is_expected.to render_file('/etc/myapp.conf').with_content("debug = false\n") }
# Or with a regexp.
it { is_expected.to render_file('/etc/myapp.conf').with_content(/user = \d+/) }
# Or with a compound matcher.
it { is_expected.to render_file('/etc/myapp.conf').with_content(start_with('# This file managed by Chef')) }
# Or with a Ruby block of arbitrary assertions.
it do
is_expected.to render_file('/etc/myapp.conf').with_content { |content|
# Arbitrary RSpec code goes here.
}
end
As actions do not normally run in ChefSpec, testing for notifications is a special case. Unlike the resource matchers which evaluate against the ChefSpec runner, the notification matchers evaluate against a resource object:
# To match `notifies :run, 'execute[unpack]', :immediately
it { expect(chef_run.remote_file('/download')).to notify('execute[unpack]') }
# To check for a specific notification action.
it { expect(chef_run.remote_file('/download')).to notify('execute[unpack]').to(:run) }
# And to check for a specific timing.
it { expect(chef_run.remote_file('/download')).to notify('execute[unpack]').to(:run).immediately }
And similarly for subscriptions:
it { expect(chef_run.execute('unpack')).to subscribe_to('remote_file[/download]').on(:create) }
RSpec expectations always need a value to run against, with the main value being
tested for a given example group (describe
or context
block) is called the
subject
. In ChefSpec this is almost always ChefSpec::Runner
that has converge
some recipe code.
There are two ways to set which recipe code should be run for the test. More commonly
for testing custom resources, you use the recipe
helper method in the test to
provide an in-line block of recipe code:
describe 'something' do
recipe do
my_custom_resource 'something' do
debug true
end
end
end
By using an in-line block of recipe code, you can try many variations to test different configurations of your custom resource.
If no recipe
block is present, ChefSpec will use the name of the top-level
describe
block as a recipe name to run. So for the case of testing a recipe
in your cookbook, use the cookbookname::recipename
string as the label:
describe 'mycookbook'
# Or.
describe 'mycookbook::myrecipe'
Most ChefSpec configuration is set in your example groups (describe
or context
blocks) using helper methods. These all follow the RSpec convention of inheriting
from a parent group to the groups inside it. So a setting in your top-level describe
will automatically be set in any context
unless overridden:
describe 'something' do
platform 'ubuntu'
# Platform is Ubuntu for any tests here.
it { is_expected.to ... }
context 'when something' do
# Platform is still Ubuntu for any tests here.
end
context 'when something else' do
platform 'fedora'
# But platform here will be Fedora.
end
end
To support simulating Chef runs on the same OS as you use your cookbooks on, ChefSpec
loads pre-fabricated Ohai data from Fauxhai.
To configure which OS' data is set for your test, use the platform
helper method:
describe 'something' do
platform 'ubuntu', '18.04'
# ...
end
You can specify a partial version number to get the latest version of that OS matching the provided prefix, or leave the version off entirely to get the latest version overall:
# Will use the latest RedHat 7.x.
platform 'redhat', '7'
# Will use the latest version of Windows.
platform 'windows'
WARNING: If you leave off the version or use a partial version prefix, the behavior of your tests may change between versions of Chef Workstation as new data is available in Fauxhai. Only use this feature if you're certain that your tests do not (or should not) depend on the specifics of OS version.
Node attributes are set using the default_attributes
, normal_attributes
,
override_attributes
, and automatic_attributes
helper methods. These inherit
from a parent group to its children using a deep merge, like in other places in
Chef:
describe 'something' do
default_attributes['myapp']['name'] = 'one'
default_attributes['myapp']['email'] = '[email protected]'
context 'when something' do
default_attributes['myapp']['name'] = 'two'
end
end
Any values set using automatic_attributes
take priority over Fauxhai data.
Normally ChefSpec skips all resource (and provider) actions. When testing the implementation of a custom resource, we need to tell ChefSpec to run actions on our specific custom resource so it can be tested:
describe 'something' do
step_into :my_custom_resource
end
You can specify any other ChefSpec configuration options using the chefspec_options
helper:
describe 'something' do
chefspec_options[:log_level] = :debug
end
In order to keep unit tests fast and independent of the target system, we have to make sure that any interaction with the system (either the target node or the Chef Server, both parts of the system just in opposite directions) is replaced with a fake, local version. For some thing, like ensuring that resource actions are replaced with a no-op, the stubbing is automatic. For others, we need to tell ChefSpec how to handle things.
The most common case of interacting with the system is a guard clause on a resource:
not_if 'some command'
# Or.
only_if 'some command'
In order for ChefSpec to know how to evaluate the resource, we need to tell it how the command would have returned for this test if it was running on the actual machine:
describe 'something' do
recipe do
execute '/opt/myapp/install.sh' do
# Check if myapp is installed and runnable.
not_if 'myapp --version'
end
end
before do
# Tell ChefSpec the command would have succeeded.
stub_command('myapp --version').and_return(true)
# Tell ChefSpec the command would have failed.
stub_command('myapp --version').and_return(false)
# You can also use a regexp to stub multiple commands at once.
stub_command(/^myapp/).and_return(false)
end
end
If using the Ruby code block form of a guard (e.g. not_if { something }
), see
the Ruby stubbing section below.
When testing code that uses the search()
API in Chef, we have to stub out the
results that would normally come from the Chef Server:
describe 'something' do
recipe do
web_servers = search(:node, 'roles:web').map { |n| n['hostname'] }
end
before do
stub_search(:node, 'roles:web').and_return([{hostname: 'one'}, {hostname: two}])
end
end
When testing code in a library that uses Chef::Search::Query.new.search()
, we have
to stub out the results that would normally come from the Chef Server:
describe 'something' do
recipe do
results = Chef::Query::Search.new.search(:node, "tags:mytag AND chef_environment:my_env"))
end
before do
query = double
allow(query).to receive(:search) do |_, arg2|
case arg2.downcase
when /tags\:mytag AND chef_environment\:my_env/
[
[
stub_node("server01", ohai: { hostname: "server01", ipaddress: '169.0.0.1' }, platform: 'windows', version: '2016'),
stub_node("server02", ohai: { hostname: "server02", ipaddress: '169.0.0.2' }, platform: 'windows', version: '2016'),
],
0,
2,
]
else
[
[],
0,
0
]
end
allow(Chef::Search::Query).to receive(:new).and_return(query)
end
end
Similar to the Search API, the data_bag()
and data_bag_item()
APIs normally
fetch data from Chef Server so we need to stub their results:
describe 'something' do
recipe do
# Side note: don't write recipe code like this. This should be `search(:users, '*:*')`.
users = data_bag('users').map do |user|
data_bag_item('users', user['id'])
end
end
before do
stub_data_bag('users').and_return(['asmithee'])
stub_data_bag_item('users', 'asmithee').and_return({uid: 1234})
end
end
When testing custom resources, it is often useful to stub methods on the resource
or provider instance. These can be set up using the stubs_for_resource
and
stubs_for_provider
helpers:
describe 'something' do
recipe do
my_custom_resource 'something'
end
# Set up stubs for just the one resource.
stubs_for_resource('my_custom_resource[something]') do |res|
# Can use any RSpec Mocks code here, see below.
allow(res).to receive(:something)
end
# Stubs for any instance of my_custom_resource.
stubs_for_resource('my_custom_resource') do |res|
# ...
end
# Stubs for any resource.
stubs_for_resource do |res|
# ...
end
# Stubs for the provider for just the one resource.
stubs_for_provider('my_custom_resource[something]') do |res|
# Can use any RSpec Mocks code here, see below.
allow(res).to receive(:something)
end
# And similar to the above for any provider of a type or any overall.
end
By default, stubs for the resource will also be set up on the current_resource
and after_resource
objects that are
created via load_current_value
. This can be disabled by using stubs_for_resource('my_custom_resource[something]', current_value: false)
. You can also manually set stubs for only the current_resource
and after_resource
objects using
stubs_for_current_value
.
For more complex Ruby code, in recipes, libraries, or custom resources, you have the full power of RSpec and RSpec Mocks available to you.
One issue that comes up often is stubbing filesystem checks such as File.exist?
. Since those are global class methods
by stubbing them they will be stubbed throughout the entire chef-client codebase that chefspec relies upon. There are
many calls to File.exist?
in any chefspec test that are not immediately visible to the user. In order to
make the client behave correctly the pattern that should be followed is to allow all the chef-client calls to operate
normally using and_call_original
and then to stub the exact path the test needs:
before do
allow(File).to receive(:exist?).and_call_original
allow(File).to receive(:exist?).with('/test/path').and_return(true)
end
All the ruby methods off of the File, Dir and FileUtils classes along with any other global class methods that the client might use, should follow a similar pattern for stubbing.
Check out the RSpec Mocks documentation for more information about setting up Ruby method stubs.
- Fork the repository from GitHub.
- Clone your fork to your local machine:
$ git clone [email protected]:USER/chefspec.git
- Create a git branch
$ git checkout -b my_bug_fix
-
Write tests
-
Make your changes/patches/fixes, committing appropriately
-
Run the tests:
bundle exec rake
-
Push your changes to GitHub
-
Open a Pull Request
ChefSpec is on [Travis CI][travis] which tests against multiple Chef and Ruby versions.
If you are contributing, please see the Contributing Guidelines for more information.
MIT - see the accompanying LICENSE file for details.