Add guidelines on adding a new service

Add docs to provide more info and guidelines on how to add a new service
to Kolla.

There are a lot of patterns and practices that have evolved in Kolla so
it's good to formalise them to make it easier for both developers and
reviewers to ensure nothing has being missed.

Change-Id: I9ab858fdd0c756e2505fc0e3487bd8e026da5db1
This commit is contained in:
Paul Bourke 2016-08-26 15:31:28 +01:00
parent ca100f1f32
commit 8229baff74

View File

@ -40,3 +40,120 @@ https://git.openstack.org/cgit/openstack-dev/sandbox, for learning, understandin
and testing the `Gerrit Workflow`_. and testing the `Gerrit Workflow`_.
.. _Gerrit Workflow: http://docs.openstack.org/infra/manual/developers.html#development-workflow .. _Gerrit Workflow: http://docs.openstack.org/infra/manual/developers.html#development-workflow
Adding a new service
====================
Kolla aims to both containerise and deploy all services within the OpenStack
"big tent". This is a constantly moving target as the ecosystem grows, so these
guidelines aim to help make adding a new service to Kolla a smooth experience.
The image
---------
Kolla follows Docker best practices
(https://docs.docker.com/engine/userguide/eng-image/dockerfile_best-practices/)
when designing and implementing services where at all possible.
We use ``jinja2`` templating syntax to help manage the volume and complexity
that comes with maintaining multiple Dockerfiles for multiple different base
operating systems.
Images should be created under the ``docker`` directory. OpenStack services
should inherit from the provided ``openstack-base`` image, while supporting and
infrastructure services (e.g. mongodb) should inherit from ``base``.
Services consisting of only one service should be placed in an image named the
same as that service, e.g. ``horizon``. Services that consist of multiple
processes generally use a base image and child images, e.g. ``glance-base``,
``glance-api``, and ``glance-registry``.
Jinja2 'blocks' are employed throughout the Dockerfile's to help operators
customise various stages of the build (refer to
http://docs.openstack.org/developer/kolla/image-building.html?highlight=override#dockerfile-customisation)
Some of these blocks are free form however, there are a subset that should be
common to every Dockerfile. The overall structure for a multi container service
is as follows::
FROM {{ namespace }}/{{ image_prefix }}openstack-base:{{ tag }}
MAINTAINER {{ maintainer }}
{% import "macros.j2" as macros with context %}
<< binary specific steps >>
<< source specific steps >>
<< common steps >>
{% block << service >>_footer %}{% endblock %}
{% block footer %}{% endblock %}
{{ include_footer }}
.. NOTE::
The generic footer block ``{% block footer %}{% endblock %}`` should not be
included in base images (e.g. glance-base).
{{ include_footer }} is legacy and should not be included in new services, it
is superseded by {% block footer %}{% endblock %}
Orchestration
-------------
As of the Newton release there are two main orchestration methods in existence
for Kolla, Ansible and Kubernetes. Ansible is the most mature and generally
regarded as the reference implementation.
When adding a role for a new service in Ansible, there are couple of patterns
that Kolla uses throughout that should be followed.
* The sample inventories
Entries should be added for the service in each of
``ansible/inventory/multinode`` and ``ansible/inventory/all-in-one``.
* The playbook
The main playbook that ties all roles together is in ``ansible/site.yml``,
this should be updated with appropriate roles, tags, and conditions. Ensure
also that supporting hosts such as haproxy are updated when necessary.
* The common role
A ``common`` role exists which sets up logging, ``kolla-toolbox`` and other
supporting components. This should be included in all services within
``meta/main.yml`` of your role.
* Common tasks
All services should include the following tasks:
- ``do_reconfigure.yml`` : Used to push new configuration files to the host and
restart the service.
- ``pull.yml`` : Used to pre fetch the image into the Docker image cache on hosts,
to speed up intial deploys.
- ``upgrade.yml`` : Used for upgrading the service in a rolling fashion. May
include service specific setup and steps as not all services can be upgraded
in the same way.
Other than the above, most roles follow the following pattern::
# Register
Involves registering the service with Keystone, creating endpoints, roles, users,
etc.
# Config
Distributes the config files to the nodes to be pulled into the container on
startup.
# Bootstrap
Creating the database (but not tables), database user for the service,
permissions, etc.
# Bootstrap Service
Starts a one shot container on the host to create the database tables, and other
initial run time config.
# Start
Start the service(s).