Lance Bragstad 429e115686 Add contributor/ directory for docs

The specification detailing the great documentation migration
requires a doc/source/contributor/ directory for all contributor
documentation. Keystone had a directory specifically for this type
of content but it was in doc/source/devref/.

This commit renames devref/ to contributor/ because the term
`contributor` is more inclusive of other community members that
aren't necessarily developers. The specification also requires this
convention for consistency with other projects.

Change-Id: Id6363213dc6ba6946bd272ca5ee67b4604afc5ac
Partial-Bug: 1694460

2017-08-01 21:14:46 +00:00

4.1 KiB

Raw Blame History

Developing Keystone Drivers

A driver, also known as a backend, is an important architectural component of Keystone. It is an abstraction around the data access needed by a particular subsystem. This pluggable implementation is not only how Keystone implements its own data access, but how you can implement your own!

Each major subsystem (that has data access needs) implements the data access by using drivers. Some examples of Keystone's drivers:

keystone.identity.backends.ldap.Identity
keystone.token.providers.fernet.core.Provider
keystone.contrib.federation.backends.sql.Federation

In/Out of Tree

It's best to start developing your custom driver outside of the Keystone development process. This means developing it in your own public or private git repository and not worrying about getting it upstream (for now).

This is better for you because it gives you more freedom and you are not bound to the strict OpenStack development rules or schedule. You can iterate faster and take whatever shortcuts you need to get your product out of the door.

This is also good for Keystone because it will limit the amount of drivers that must be maintained by the team. If the team had to maintain a driver for each NoSQL DB that deployers want to use in production there would be less time to make Keystone itself better. Not to mention that the team would have to start gaining expertise in potentially dozens of new technologies.

As you'll see below there is no penalty for open sourcing your driver, on GitHub for example, or even keeping your implementation private. We use Setuptools entry points to load your driver from anywhere in the Python path.

How To Make a Driver

The TLDR; steps (and too long didn't write yet):

Determine which subsystem you would like write a driver for
Subclass the most current version of the driver interface
Implement each of the abstract methods for that driver
1. We are currently not documenting the exact input/outputs of the driver methods. The best approach right now is to use an existing driver as an example of what data your driver will receive and what data your driver will be required to return.
2. There is a plan in place to document these APIs in more detail.
Register your new driver as an entry point
Configure your new driver in keystone.conf
Sit back and enjoy!

Driver Interface Changes

We no longer support driver versioning. Thus, if a driver interface changes, you will need to upgrade your custom driver to meet the new driver contract.

Removing Methods

Newer driver interfaces may remove methods that are currently required. Methods are removed when they are no longer required or invoked by Keystone. There is no reason why methods removed from the Keystone interface need to be removed from custom drivers.

Adding Methods

The most common API changes will be adding methods to support new features. The new method must be implemented by custom driver implementations.

Updating Methods

We will do our best not to update existing methods in ways that will break custom driver implementations. However, if that is not possible, again you will need to upgrade your custom driver implementation to meet the new driver contract.

4.1 KiB Raw Blame History