Coding Style


this document is a work in progress and the content will evolve. Any contribution is welcome though ;-)

  • Read this page

  • Make sure that what you’re going to code is not already a work in progress

  • Make sure you’re familiar with Puppet Syntax, Lint, Rspec and Litmus

  • If you want to create a new module, read New Module.

Best practices


Any patch must be backward compatible.

It meansː

  • do not break the interface (deprecate parameters for at least one cycle, and add a warning for our users)

  • do not change default parameters (except if you have a good reason but your commit message must explain it)

Commit message

Please read GitCommitMessages


We have a lot of modules to maintain, please keep our code consistent. Before adding new parameters or new classes, see if other modules already implement them and keep consistent. See also our common libraries OpenStackLib

Class structure and dependencies

The structure of classes should be consistent between modules for common resources. There should be as few include or require statements as possible unless absolutely neccesary, such as in the case when backward compatibility is required.

This helps to prevent redeclaration problems and issues where consumers of the modules are forced to place resources in the proper order in their manifests to not cause those issues. We should strive towards not enforcing ordering of resources in code, if we have to do that we should inform our users properly and fail early unless the condition is met.

Note that this is not about the runtime dependency chain of resources but about placement of code that can cause redeclaration issues because we include a class in another and the consumer specifying that class at the same time.

Classes should be used to logically split up the functionality of modules to not place everything in the same place. For example in the puppet-cinder instead of having cinder::nova_username which configures [nova]/username you should instead supply the cinder::nova class that takes care of the whole nova config section.

Same structure should be used for database, logging etc to not group up all the configuration options in the init class of the modules. Examples of this is the <modules>::db```and ``<module>::logging classes we have today.

Config file defaults and parameters

We want users that don’t override a value to get the default for the service being configured. Many of the parameters passed in to classes and defined types translate directly into a config file setting. When a parameter translates directly to a config file value, and the value is optional, it should be set to $::os_service_default. This is a special value that will ensure that the service default is used by removing any existing value from the config file.

Empty parameters

When you need to specify an empty (nil) parameter, using undef is the best choice. Do not useː “ (not Puppetish) or false (undef is false if tested as a boolean).

Adding new depdnency

When you add a new dependency, update metadata.json in that repository. If the new dependency is not yet used by the other modules, then that should be added to Puppetfile so that the new dependency is installed during tests. If the new depdnency is already used in the other modules, then set the consistent version constraints unless you have a specific requirement,


Everything about testing you can find here.


  • Validate all parameters are documented. They are required and lint will check it.

  • If possible, keep examples/*.pp updated, they are very useful for our users.

  • Comment your code when needed (temporary workarounds, TODO, etc).

  • Any change in interface (like new parameters, deprecations, and etc) or fundamental behavior should be documented in a release note. In Puppet OpenStack, we use reno to maintain release notes.