Moving to Keystone v3¶
This blueprint is meant to capture the changes to the Neutron repos (neutron, neutron-lib and python-neutronclient) needed to fully integrate Neutron with the Keystone v3 API.
All the steps described in this document are applicable also to repos neutron-fwaas, neutron-lbaas, neutron-vpnaas and neutron-dynamic-routing. However, only the changes in neutron, neutron-lib and python-neutronclient are required to consider this blueprint completed.
The Keystone v3 API  was released on February 20, 2013. In the Icehouse release the Keystone v2 API was deprecated . The deprecation had to be reverted  because many OpenStack projects did not support the v3 API yet.
In the Mitaka release Keystone is again deprecating the v2 API . Projects will be able to use the v2 API for the next four releases, but after that it will not be supported. To be consistent with other OpenStack projects and to be able to continue using the Identity Service, we must update Neutron to exclusively use Keystone v3 when v2 is no longer supported.
With the Keystone v3 API, the concept of “tenant” is deprecated in favor of
“project”. In particular, the attribute
tenant_id is replaced with
project_id. Neutron must initially support
project_id as an alias for
tenant_id before finally replacing
tenant_id altogether. Deprecation
will be kept as long as Neutron CLI will be supported.
In requests to Neutron, the neutron server already treats
an alias for
tenant_id. Internally, Neutron should eventually replace
project_id in attributes and database fields.
In requests from Neutron to other OpenStack services, the neutron components should use keystoneauth, which abstracts the v2/v3 differences.
Change all existing uses in Neutron code from Keystone v2 to Keystone v3. Initially both versions of API will be supported. Migration will be handled in several steps.
Update Neutron server to accept Keystone v3 contexts in requests. [DONE]
Update the API bindings in python-neutronclient to properly handle both project_id and tenant_id.
Update Neutron server to use only keystoneauth for requests to other OpenStack services.
Update documentation about deprecation of tenant_id attribute:
OpenStack Networking API reference,
Neutron and neutronclient developer references.
project_idalias for all
tenant_idcolumns in DB.
Update codebase, changing all occurences of
project, where appropriate. Special care needs to be taken for:
When renaming arguments for callables, some callers may pass them using kwargs notation (tenant_id=…) instead of positional.
When changing code that is used by external projects, the change must not break the external project. Deprecation warnings should be set up.
tenant_id is used as a filter parameter in Neutron API. It is a
requirement that this use-case will continue to work.
Neutron API responses will be expanded to contain both
project_id. For now we are unable to phase out
tenant_id since that
would require an API version change. When microversioning  is adopted for the
Neutron API the task of removing deprecated attributes can be commenced.
Meanwhile we will expose
project_id through yet another API extension,
enabled by default.
For CLI commands, the Neutron client is being deprecated in favor of OpenStack
Client . Therefore the CLI portion of python-neutronclient will not be
updated. In the OpenStack Client the CLI commands have already adopted the
policy of using
project_ and do not support
The only part of python-neutronclient which will be modified is the Python API
bindings. All the places where the bindings accept
tenant_name need to be expanded to also accept
tenant_ parameters will be marked as deprecated and
to be removed in the future.
The neutronclient Python API documentation will be updated to show corresponding changes about bindings.
Neutron developer reference¶
tenant is used all across the documentation. All the places need to
be updated, with extra information about
tenant_id deprecation in favor of
Additional guidelines will be provided for all subprojects/affiliated projects, about modifications required to be consistent with Neutron API changes.
Renaming columns from
project_id and from
project_id will require downtime and will be done in a contract
For some time, database operations will see
attributes separately coming up in lists of properties and such, and care will
need to be taken in how they are filtered and passed around.
Once all existing columns have been renamed to
project_id, all new columns
added to any Neutron code shall use only
project_id. A hacking rule will
be added to prevent the reintroduction of
tenant_id in source code.
Server code changes will be submitted in reviewable chunks, without interruption to the external API or external project dependencies. Client code changes are additive until the deprecated keywords are removed.
Write tests to check responses for
project parameters, to be
sure that introduced changes are consistent.
project_idreturns the same response as
tenant_idis updated when