2017-06-13 Document current level of community database support

Over the years, OpenStack’s ecosystem and developer community grew around MySQL (as defined in [2]). Emerging from the OpenStack User Survey 2017 [1] there is a trend visible that most production deployments today settle on MySQL as the configured database backend.

Although the OpenStack User Survey is non-authoritative, its reported bias (> 10 to 1 for MySQL [2] vs. PostgreSQL [3]) is also mirrored in openstack-operator mailing list posts, ask.openstack.org posts, and content during operator meetups. In contrast, production deployments using PostgreSQL appear to be less popular ([4]).

The OpenStack community is generally offering wide ranges of options to the users. For DBMS we advocate the use of abstraction layers like for example oslo.db and SQLAlchemy to avoid a lock-in on a specific database implementation.

Still the development community could design and develop individual features with specific backends in mind. Even if portability issues are caught during code review and various levels of automated QA, there is still impact on the community like documentation and other cost of handling peculiarities of each DBMS like best practices for each implementation choice.

Therefore, we should clearly state that currently in the OpenStack community there is a bias towards MySQL [2] over other databases like for example PostgreSQL.

As a set of concrete steps we should do the following:

• The OpenStack documentation should be updated to document the community preference:

  .. warning::
  
     While OpenStack can be used with many SQL DBMS, please
     be aware that the upstream OpenStack developer community
     is so far focusing on the MySQL family of databases
     as a backend and tests OpenStack against the version of
     MySQL provided with Operating System distributions.
  
     If you exclusively rely on the OpenStack upstream
     community and not on an OpenStack vendor for support of your
     OpenStack installation you are proceeding at your own
     risk by choosing a different DBMS backend.
  

• We should state to the OpenStack Foundation that there needs to be a way for distributors to report back aggregate data about their users, who are not self reporting to the survey. The upstream community does not have infinite resources, and is legitimately trying to make hard and informed decisions about things not to do. Because there is no mandatory registration of clouds, the best information we can get is via the User Survey and what vendors decide to share.

In the future we may want to make stronger statements, but the important thing is being clear to users that PostgreSQL is currently not first class supported upstream and there isn’t any coordinated effort to change that.

This should not exclude vendors claiming support for other DBMS choices for their customers. However, for users bypassing vendors and going straight to upstream, the current reality of OpenStack enablement needs to be clear.

Footnotes

[1]

April 2017 User Survey https://www.openstack.org/assets/survey/April2017SurveyReport.pdf

[2] (1,2,3)

The term MySQL is used as a reference to the MySQL database and the close variants MariaDB and Galera.

[3]

The term PostgreSQL is used as a reference to the PostgreSQL database and close derivatives.

[4]

As of the Boston Forum (May 2017) known users of PostgreSQL are:

• SUSE: investigating migrating to Galera

• Huawei: not actually PostgreSQL, but an internal db based on pg.

• Windriver

• SAP: consuming OpenStack directly, not via vendor. They did not respond to the user survey, so their usage is not recorded there.