Skip to content

Latest commit

 

History

History
835 lines (631 loc) · 27.7 KB

README_DEVELOPER.md

File metadata and controls

835 lines (631 loc) · 27.7 KB
Raw

Developer README

This file is intended to provide a place for developers and contributors to document what other developers need to know about changes made to Puppet.

Internal Structures

Two Types of Catalog

When working on subsystems of Puppet that deal with the catalog it is important to be aware of the two different types of Catalog. Developers will often find this difference while working on the static compiler and types and providers.

The two different types of catalog becomes relevant when writing spec tests because we frequently need to wire up a fake catalog so that we can exercise types, providers, or terminii that filter the catalog.

The two different types of catalogs are so-called "resource" catalogs and "RAL" (resource abstraction layer) catalogs. At a high level, the resource catalog is the in-memory object we serialize and transfer around the network. The compiler terminus is expected to produce a resource catalog. The agent takes a resource catalog and converts it into a RAL catalog. The RAL catalog is what is used to apply the configuration model to the system.

Resource dependency information is most easily obtained from a RAL catalog by walking the graph instance produced by the relationship_graph method.

Resource Catalog

If you're writing spec tests for something that deals with a catalog "server side," a new catalog terminus for example, then you'll be dealing with a resource catalog. You can produce a resource catalog suitable for spec tests using something like this:

let(:catalog) do
  catalog = Puppet::Resource::Catalog.new("node-name-val") # NOT certname!
  rsrc = Puppet::Resource.new("file", "sshd_config",
    :parameters => {
      :ensure => 'file',
      :source => 'puppet:///modules/filetest/sshd_config',
    }
  )
  rsrc.file = 'site.pp'
  rsrc.line = 21
  catalog.add_resource(rsrc)
end

The resources in this catalog may be accessed using catalog.resources. Resource dependencies are not easily walked using a resource catalog however. To walk the dependency tree convert the catalog to a RAL catalog as described in

RAL Catalog

The resource catalog may be converted to a RAL catalog using catalog.to_ral. The RAL catalog contains Puppet::Type instances instead of Puppet::Resource instances as is the case with the resource catalog.

One very useful feature of the RAL catalog are the methods to work with resource relationships. For example:

irb> catalog = catalog.to_ral
irb> graph = catalog.relationship_graph
irb> pp graph.edges
[{ Notify[alpha] => File[/tmp/file_20.txt] },
 { Notify[alpha] => File[/tmp/file_21.txt] },
 { Notify[alpha] => File[/tmp/file_22.txt] },
 { Notify[alpha] => File[/tmp/file_23.txt] },
 { Notify[alpha] => File[/tmp/file_24.txt] },
 { Notify[alpha] => File[/tmp/file_25.txt] },
 { Notify[alpha] => File[/tmp/file_26.txt] },
 { Notify[alpha] => File[/tmp/file_27.txt] },
 { Notify[alpha] => File[/tmp/file_28.txt] },
 { Notify[alpha] => File[/tmp/file_29.txt] },
 { File[/tmp/file_20.txt] => Notify[omega] },
 { File[/tmp/file_21.txt] => Notify[omega] },
 { File[/tmp/file_22.txt] => Notify[omega] },
 { File[/tmp/file_23.txt] => Notify[omega] },
 { File[/tmp/file_24.txt] => Notify[omega] },
 { File[/tmp/file_25.txt] => Notify[omega] },
 { File[/tmp/file_26.txt] => Notify[omega] },
 { File[/tmp/file_27.txt] => Notify[omega] },
 { File[/tmp/file_28.txt] => Notify[omega] },
 { File[/tmp/file_29.txt] => Notify[omega] }]

If the relationship_graph method is throwing exceptions at you, there's a good chance the catalog is not a RAL catalog.

Settings Catalog

Be aware that Puppet creates a mini catalog and applies this catalog locally to manage file resource from the settings. This behavior made it difficult and time consuming to track down a race condition in 2888.

Even more surprising, the File[puppetdlockfile] resource is only added to the settings catalog if the file exists on disk. This caused the race condition as it will exist when a separate process holds the lock while applying the catalog.

It may be sufficient to simply be aware of the settings catalog and the potential for race conditions it presents. An effective way to be reasonably sure and track down the problem is to wrap the File.open method like so:

# We're wrapping ourselves around the File.open method.
# As described at: http://goo.gl/lDsv6
class File
  WHITELIST = [ /pidlock.rb:39/ ]

  class << self
    alias xxx_orig_open open
  end

  def self.open(name, *rest, &block)
    # Check the whitelist for any "good" File.open calls against the #
    puppetdlock file
    white_listed = caller(0).find do |line|
      JJM_WHITELIST.find { |re| re.match(line) }
    end

    # If you drop into IRB here, take a look at your caller, it might be
    # the ghost in the machine you're looking for.
    binding.pry if name =~ /puppetdlock/ and not white_listed
    xxx_orig_open(name, *rest, &block)
  end
end

The settings catalog is populated by the Puppet::Util::Settings#to\_catalog method.

Ruby Dependencies

Puppet is considered an Application as it relates to the recommendation of adding a Gemfile.lock file to the repository and the information published at Clarifying the Roles of the .gemspec and Gemfile

To install the dependencies run: bundle install to install the dependencies.

A checkout of the source repository should be used in a way that provides puppet as a gem rather than a simple Ruby library. The parent directory should be set along the GEM_PATH, preferably before other tools such as RVM that manage gemsets using GEM_PATH.

For example, Puppet checked out into /workspace/src/puppet using git checkout https://github.com/puppetlabs/puppet in /workspace/src can be used with the following actions. The trick is to symlink gems to src.

$ cd /workspace
$ ln -s src gems
$ mkdir specifications
$ pushd specifications; ln -s ../gems/puppet/puppet.gemspec; ln -s ../gems/puppet/lib; popd
$ export GEM_PATH="/workspace:${GEM_PATH}"
$ gem list puppet

This should list out

puppet (2.7.19)

The final directory structure should look like this:

/workspace/src --- git working directory
          /gems -> src
          /specifications/puppet.gemspec -> ../gems/puppet/puppet.gemspec
                         /lib -> ../gems/puppet/lib

Bundler

With a source checkout of Puppet properly setup as a gem, dependencies can be installed using Bundler

$ bundle install
Fetching gem metadata from http://rubygems.org/........
Using diff-lcs (1.1.3)
Installing facter (1.6.11)
Using metaclass (0.0.1)
Using mocha (0.10.5)
Using puppet (2.7.19) from source at /workspace/puppet-2.7.x/src/puppet
Using rack (1.4.1)
Using rspec-core (2.10.1)
Using rspec-expectations (2.10.0)
Using rspec-mocks (2.10.1)
Using rspec (2.10.0)
Using bundler (1.1.5)
Your bundle is complete! Use `bundle show [gemname]` to see where a bundled gem is installed.

Running Tests

Puppet Labs projects use a common convention of using Rake to run unit tests. The tests can be run with the following rake task:

rake spec
# Or if using Bundler
bundle exec rake spec

This allows the Rakefile to set up the environment beforehand if needed. This method is how the unit tests are run in Jenkins.

Under the hood Puppet's tests use rspec. To run all of them, you can directly use 'rspec':

rspec
# Or if using Bundler
bundle exec rspec

To run a single file's worth of tests (much faster!), give the filename, and use the nested format to see the descriptions:

rspec spec/unit/ssl/host_spec.rb --format nested

A brief introduction to testing in Puppet

Puppet relies heavily on automated testing to ensure that Puppet behaves as expected and that new features don't interfere with existing behavior. There are three primary sets of tests that Puppet uses: unit tests, integration tests, and acceptance tests.

Unit tests are used to test the individual components of Puppet to ensure that they function as expected in isolation. Unit tests are designed to hide the actual system implementations and provide canned information so that only the intended behavior is tested, rather than the targeted code and everything else connected to it. Unit tests should never affect the state of the system that's running the test.

Integration tests serve to test different units of code together to ensure that they interact correctly. While individual methods might perform correctly, when used with the rest of the system they might fail, so integration tests are a higher level version of unit tests that serve to check the behavior of individual subsystems.

All of the unit and integration tests for Puppet are kept in the spec/ directory.

Acceptance tests are used to test high level behaviors of Puppet that deal with a number of concerns and aren't easily tested with normal unit tests. Acceptance tests function by changing system state and checking the system after the fact to make sure that the intended behavior occurred. Because of this acceptance tests can be destructive, so the systems being tested should be throwaway systems.

All of the acceptance tests for Puppet are kept in the acceptance/tests/ directory.

Puppet Continuous integration

RSpec

Puppet uses RSpec to perform unit and integration tests. RSpec handles a number of concerns to make testing easier:

  • Executing examples and ensuring the actual behavior matches the expected behavior (examples)
  • Grouping tests (describe and contexts)
  • Setting up test environments and cleaning up afterwards (before and after blocks)
  • Isolating tests (mocks and stubs)

Examples and expectations

At the most basic level, RSpec provides a framework for executing tests (which are called examples) and ensuring that the actual behavior matches the expected behavior (which are done with expectations)

# This is an example; it sets the test name and defines the test to run
specify "one equals one" do
  # 'should' is an expectation; it adds a check to make sure that the left argument
  # matches the right argument
  1.should == 1
end

# Examples can be declared with either 'it' or 'specify'
it "one doesn't equal two" do
  1.should_not == 2
end

Good examples generally do as little setup as possible and only test one or two things; it makes tests easier to understand and easier to debug.

More complete documentation on expectations is available at https://www.relishapp.com/rspec/rspec-expectations/docs

Example groups

Example groups are fairly self explanatory; they group similar examples into a set.

describe "the number one" do

  it "is larger than zero" do
    1.should be > 0
  end

  it "is an odd number" do
    1.odd?.should be true
  end

  it "is not nil" do
    1.should_not be_nil
  end
end

Example groups have a number of uses that we'll get into later, but one of the simplest demonstrations of what they do is how they help to format documentation:

rspec ex.rb --format documentation

the number one
  is larger than zero
  is an odd number
  is not nil

Finished in 0.00516 seconds
3 examples, 0 failures

Setting up and tearing down tests

Examples may require some setup before they can run, and might need to clean up afterwards. before and after blocks can be used before this, and can be used inside of example groups to limit how many examples they affect.

describe "something that could warn" do
  before :each do
    # Disable warnings for this test
    $VERBOSE = nil
  end

  after do
    # Enable warnings afterwards
    $VERBOSE = true
  end

  it "doesn't generate a warning" do
    MY_CONSTANT = 1
    # reassigning a normally prints out 'warning: already initialized constant FOO'
    MY_CONSTANT = 2
  end
end

Setting up helper data

Some examples may require setting up data before hand and making it available to tests. RSpec provides helper methods with the let method call that can be used inside of tests.

describe "a helper object" do
  # This creates an array with three elements that we can retrieve in tests. A
  # new copy will be made for each test.
  let(:my_helper) do
    ['foo', 'bar', 'baz']
  end

  it "should be an array" do
    my_helper.should be_a_kind_of Array
  end

  it "should have three elements" do
    my_helper.should have(3).items
  end
end

Like before blocks, helper objects like this are used to avoid doing a lot of setup in individual examples and share setup between similar tests.

Isolating tests with stubs

RSpec allows you to provide fake data during testing to make sure that individual tests are only running the code being tested. You can stub out entire objects, or just stub out individual methods on an object. When a method is stubbed the method itself will never be called.

While RSpec comes with its own stubbing framework, Puppet uses the Mocha framework.

A brief usage guide for Mocha is available at http://gofreerange.com/mocha/docs/#Usage, and an overview of Mocha expectations is available at http://gofreerange.com/mocha/docs/Mocha/Expectation.html

describe "stubbing a method on an object" do
  let(:my_helper) do
    ['foo', 'bar', 'baz']
  end

  it 'should have three items before being stubbed' do
    my_helper.size.should == 3
  end

  describe 'when stubbing the size' do
    before do
      my_helper.stubs(:size).returns 10
    end

    it 'should have the stubbed value for size' do
      my_helper.size.should == 10
    end
  end
end

Entire objects can be stubbed as well.

describe "stubbing an object" do
  let(:my_helper) do
    stub(:not_an_array, :size => 10)
  end

  it 'should have the stubbed size'
    my_helper.size.should == 10
  end
end

Adding expectations with mocks

It's possible to combine the concepts of stubbing and expectations so that a method has to be called for the test to pass (like an expectation), and can return a fixed value (like a stub).

describe "mocking a method on an object" do
  let(:my_helper) do
    ['foo', 'bar', 'baz']
  end

  describe "when mocking the size" do
    before do
      my_helper.expects(:size).returns 10
    end

    it "adds an expectation that a method was called" do
      my_helper.size
    end
  end
end

Like stubs, entire objects can be mocked.

describe "mocking an object" do
  let(:my_helper) do
    mock(:not_an_array)
  end

  before do
    not_an_array.expects(:size).returns 10
  end

  it "adds an expectation that the method was called" do
    not_an_array.size
  end
end

Writing tests without side effects

When properly written each test should be able to run in isolation, and tests should be able to be run in any order. This makes tests more reliable and allows a single test to be run if only that test is failing, instead of running all 17000+ tests each time something is changed. However, there are a number of ways that can make tests fail when run in isolation or out of order.

Using instance variables

Puppet has a number of older tests that use before blocks and instance variables to set up fixture data, instead of let blocks. These can retain state between tests, which can lead to test failures when tests are run out of order.

# test.rb
RSpec.configure do |c|
  c.mock_framework = :mocha
end

describe "fixture data" do
  describe "using instance variables" do

    # BAD
    before :all do
      # This fixture will be created only once and will retain the `foo` stub
      # between tests.
      @fixture = stub 'test data'
    end

    it "can be stubbed" do
      @fixture.stubs(:foo).returns :bar
      @fixture.foo.should == :bar
    end

    it "should not keep state between tests" do
      # The foo stub was added in the previous test and shouldn't be present
      # in this test.
      expect { @fixture.foo }.to raise_error
    end
  end

  describe "using `let` blocks" do

    # GOOD
    # This will be recreated between tests so that state isn't retained.
    let(:fixture) { stub 'test data' }

    it "can be stubbed" do
      fixture.stubs(:foo).returns :bar
      fixture.foo.should == :bar
    end

    it "should not keep state between tests" do
      # since let blocks are regenerated between tests, the foo stub added in
      # the previous test will not be present here.
      expect { fixture.foo }.to raise_error
    end
  end
end
bundle exec rspec test.rb -fd

fixture data
  using instance variables
    can be stubbed
    should not keep state between tests (FAILED - 1)
  using `let` blocks
    can be stubbed
    should not keep state between tests

Failures:

  1) fixture data using instance variables should not keep state between tests
     Failure/Error: expect { @fixture.foo }.to raise_error
       expected Exception but nothing was raised
     # ./test.rb:17:in `block (3 levels) in <top (required)>'

Finished in 0.00248 seconds
4 examples, 1 failure

Failed examples:

rspec ./test.rb:16 # fixture data using instance variables should not keep state between tests

RSpec references

Puppet-acceptance

Puppet has a custom acceptance testing framework called puppet-acceptance for running acceptance tests. Puppet-acceptance runs the tests by configuring one or more VMs, copying the test cases onto the VMs, performing the tests and collecting the results, and ensuring that the results match the intended behavior. It uses test::unit to perform the actual assertions.

UTF-8 Handling

As Ruby 1.9 becomes more commonly used with Puppet, developers should be aware of major changes to the way Strings and Regexp objects are handled. Specifically, every instance of these two classes will have an encoding attribute determined in a number of ways.

  • If the source file has an encoding specified in the magic comment at the top, the instance will take on that encoding.
  • Otherwise, the encoding will be determined by the LC_LANG or LANG environment variables.
  • Otherwise, the encoding will default to ASCII-8BIT

References

Excellent information about the differences between encodings in Ruby 1.8 and Ruby 1.9 is published in this blog series: Understanding M17n

Encodings of Regexp and String instances

In general, please be aware that Ruby 1.9 regular expressions need to be compatible with the encoding of a string being used to match them. If they are not compatible you can expect to receive and error such as:

Encoding::CompatibilityError: incompatible encoding regexp match (ASCII-8BIT
regexp with UTF-8 string)

In addition, some escape sequences were valid in Ruby 1.8 are no longer valid in 1.9 if the regular expression is not marked as an ASCII-8BIT object. You may expect errors like this in this situation:

SyntaxError: (irb):7: invalid multibyte escape: /\xFF/

This error is particularly common when serializing a string to other representations like JSON or YAML. To resolve the problem you can explicitly mark the regular expression as ASCII-8BIT using the /n flag:

"a" =~ /\342\230\203/n

Finally, any time you're thinking of a string as an array of bytes rather than an array of characters, common when escaping a string, you should work with everything in ASCII-8BIT. Changing the encoding will not change the data itself and allow the Regexp and the String to deal with bytes rather than characters.

Puppet provides a monkey patch to String which returns an encoding suitable for byte manipulations:

# Example of how to escape non ASCII printable characters for YAML.
>> snowman = "☃"
>> snowman.to_ascii8bit.gsub(/([\x80-\xFF])/n) { |x| "\\x#{x.unpack("C")[0].to_s(16)} }
=> "\\xe2\\x98\\x83"

If the Regexp is not marked as ASCII-8BIT using /n, then you can expect the SyntaxError, invalid multibyte escape as mentioned above.

Windows

If you'd like to run Puppet from source on Windows platforms, the include ext/envpuppet.bat will help.

To quickly run Puppet from source, assuming you already have Ruby installed from rubyinstaller.org.

C:\> cd C:\work\puppet
C:\work\puppet> set PATH=%PATH%;C:\work\puppet\ext
C:\work\puppet> envpuppet bundle install
C:\work\puppet> envpuppet puppet --version
2.7.9

When writing a test that cannot possibly run on Windows, e.g. there is no mount type on windows, do the following:

describe Puppet::MyClass, :unless => Puppet.features.microsoft_windows? do
  ..
end

If the test doesn't currently pass on Windows, e.g. due to on going porting, then use an rspec conditional pending block:

pending("porting to Windows", :if => Puppet.features.microsoft_windows?) do
  <example1>
end

pending("porting to Windows", :if => Puppet.features.microsoft_windows?) do
  <example2>
end

Then run the test as:

C:\work\puppet> envpuppet bundle exec rspec spec

Common Issues

  • Don't assume file paths start with '/', as that is not a valid path on Windows. Use Puppet::Util.absolute_path? to validate that a path is fully qualified.

  • Use File.expand_path('/tmp') in tests to generate a fully qualified path that is valid on POSIX and Windows. In the latter case, the current working directory will be used to expand the path.

  • Always use binary mode when performing file I/O, unless you explicitly want Ruby to translate between unix and dos line endings. For example, opening an executable file in text mode will almost certainly corrupt the resulting stream, as will occur when using:

    IO.open(path, 'r') { |f| ... } IO.read(path)

    If in doubt, specify binary mode explicitly:

    IO.open(path, 'rb')

  • Don't assume file paths are separated by ':'. Use File::PATH_SEPARATOR instead, which is ':' on POSIX and ';' on Windows.

  • On Windows, File::SEPARATOR is '/', and File::ALT_SEPARATOR is ''. On POSIX systems, File::ALT_SEPARATOR is nil. In general, use '/' as the separator as most Windows APIs, e.g. CreateFile, accept both types of separators.

  • Don't use waitpid/waitpid2 if you need the child process' exit code, as the child process may exit before it has a chance to open the child's HANDLE and retrieve its exit code. Use Puppet::Util.execute.

  • Don't assume 'C' drive. Use environment variables to look these up:

    "#{ENV['windir']}/system32/netsh.exe"

Configuration Directory

In Puppet 3.x we've simplified the behavior of selecting a configuration file to load. The intended behavior of reading puppet.conf is:

  1. Use the explicit configuration provided by --confdir or --config if present
  2. If running as root (Puppet.features.root?) then use the system puppet.conf
  3. Otherwise, use ~/.puppet/puppet.conf.

When Puppet master is started from Rack, Puppet 3.x will read from ~/.puppet/puppet.conf by default. This is intended behavior. Rack configurations should start Puppet master with an explicit configuration directory using ARGV << "--confdir" << "/etc/puppet". Please see the ext/rack/files/config.ru file for an up-to-date example.

Determining the Puppet Version

If you need to programmatically work with the Puppet version, please use the following:

require 'puppet/version'
# Get the version baked into the sourcecode:
version = Puppet.version
# Set the version (e.g. in a Rakefile based on `git describe`)
Puppet.version = '2.3.4'

Please do not monkey patch the constant Puppet::PUPPETVERSION or obtain the version using the constant. The only supported way to set and get the Puppet version is through the accessor methods.

Static Compiler

The static compiler was added to Puppet in the 2.7.0 release. 1

The static compiler is intended to provide a configuration catalog that requires a minimal amount of network communication in order to apply the catalog to the system. As implemented in Puppet 2.7.x and Puppet 3.0.x this intention takes the form of replacing all of the source parameters of File resources with a content parameter containing an address in the form of a checksum. The expected behavior is that the process applying the catalog to the node will retrieve the file content from the FileBucket instead of the FileServer.

The high level approach can be described as follows. The StaticCompiler is a terminus that inserts itself between the "normal" compiler terminus and the request. The static compiler takes the resource catalog produced by the compiler and filters all File resources. Any file resource that contains a source parameter with a value starting with 'puppet://' is filtered in the following way in a "standard" single master / networked agents deployment scenario:

  1. The content, owner, group, and mode values are retrieved from th FileServer by the master.
  2. The file content is stored in the file bucket on the master.
  3. The source parameter value is stripped from the File resource.
  4. The content parameter value is set in the File resource using the form '{XXX}1234567890' which can be thought of as a content address indexed by checksum.
  5. The owner, group and mode values are set in the File resource if they are not already set.
  6. The filtered catalog is returned in the response.

In addition to the catalog terminus, the process requesting the catalog needs to obtain the file content. The default behavior of puppet agent is to obtain file contents from the local client bucket. The method we expect users to employ to reconfigure the agent to use the server bucket is to declare the Filebucket[puppet] resource with the address of the master. For example:

node default {
  filebucket { puppet:
    server => $server,
    path   => false,
  }
  class { filetest: }
}

This special filebucket resource named "puppet" will cause the agent to fetch file contents specified by checksum from the remote filebucket instead of the default clientbucket.

Trying out the Static Compiler

Create a module that recursively downloads something. The jeffmccune-filetest module will recursively copy the rubygems source tree.

$ puppet module install jeffmccune-filetest

Start the master with the StaticCompiler turned on:

$ puppet master \
    --catalog_terminus=static_compiler \
    --verbose \
    --no-daemonize

Add the special Filebucket[puppet] resource:

# site.pp
node default {
  filebucket { puppet: server => $server, path => false }
  class { filetest: }
}

Get the static catalog:

$ puppet agent --test

You should expect all file metadata to be contained in the catalog, including a checksum representing the content. When managing an out of sync file resource, the real contents should be fetched from the server instead of the clientbucket.

Package Maintainers

Software Version API

Please see the public API regarding the software version as described in lib/puppet/version.rb. Puppet provides the means to easily specify the exact version of the software packaged using the VERSION file, for example:

$ git describe --match "3.0.*" > lib/puppet/VERSION
$ ruby -r puppet/version -e 'puts Puppet.version'
3.0.1-260-g9ca4e54

EOF