governance/resolutions/20140211-tc_defcore_response.rst
Joe Gordon 0acaeb5078 Fix tiny typo: compatability => compatibility
Change-Id: I355ca5ccdefa5945d6ac00d34eb7c7f7bc94b281
2015-04-10 15:14:09 -04:00

6.6 KiB

2014-02-11 DefCore Response

This resolution is the draft of an email from the TC to the Board DefCore committee on the issue of identifying "required sections" of code.

---

The Technical Committee thanks the Board for requesting our assistance in defining the requirements for usage of the OpenStack trademark [1] in defining "core" components of OpenStack [2]. This is an important and complicated issue, so the Technical Committee feels that it needs to be solved in stages. In order to improve the engagement of the Technical Committee on this issue, we have selected Michael Still as our representative to the "DefCore" subcommittee, with Anne Gentle as his backup.

The Technical Committee understands that the end goal of the Board is to define criteria that can be used to define who shall be granted a license to use the OpenStack trademark. We understand that these criteria will be defined per OpenStack release, and that there is therefore scope to incrementally refine the requirements over time.

Designating areas of codes where a private implementation can be substituted for the mainline code [3] is time consuming, but otherwise a technical and objective answer. However, deciding which, amongst those areas, constitute a required use is more subjective and depends on the stakeholders we want to best serve.

The Technical Committee warns that we are unlikely to reach a consensus position in a time that is reasonable for the Icehouse release. Additionally, we believe we would have the following further policy questions for the board:

  • how granular and specific does the board want this to be? How do we handle drift in the code over time from operations such as refactoring? How do we identify designated sections -- both line numbers and cryptographic hashes of the lines of code have been proposed, but both methods are flawed.
  • how limiting should it be? For example, are backports from master of fixes or features to the designated areas allowed?
  • many parts of OpenStack are pluggable in ways that do not replace designated sections and are not alternate implementations of a plugin interface - for example you can add new WSGI middlewares or scheduler filters - how much of this extensibility does the board wish to allow?
  • altering libraries (3rd party, or part of OpenStack itself) can have as much impact on behaviour as altering OpenStack services itself. Is there a desire to address this?
  • at some point, these sorts of requirements, and the desire to force people to engage upstream, seem confusingly at conflict with our "business friendly" choice of a permissive license (i.e. the Apache License vs, say, the AGPL). How does the board reconcile the use of an extremely permissive software license with an extremely restrictive trademark license?

There are a variety of stake holders in this discussion, and it is hard to balance the requirements of those stake holders in the requirements definition process.

The Technical Committee sees the stake holders as:

  • end users -- end users care about interoperability of deployed OpenStack clouds. This interoperability is best measured by API testing, with tempest being a likely vehicle to drive such testing.
  • deployers -- organizations deploying OpenStack in their datacenters want confidence that their OpenStack distribution is providing code that will be supported long term by the OpenStack community. Privately implemented features increase the risk to these deployments as they are reliant on a set of developers that is smaller than the entire OpenStack community.
  • distributors -- distributors need the ability to differentiate their products, but they also need to be able to entice deployers away from other distributors. This compatibility of different distributions is one of the competitive advantages of OpenStack, and should be protected by our trademark policy. Therefore, we do not believe that private features are the right way for distributors to differentiate their products.
  • developers -- the OpenStack development community is also an important part of the OpenStack ecosystem, and should be considered in this process. We feel that encouraging distributors and other employers of developers to engage with the upstream community is of value, and that the preference is for all code to end up going through our development process instead of being held in private.

We understand that the DefCore committee is intending the "defined sections of code" requirement as a way of protecting deployers and developers from private patches being held by distributors on critical sections of the OpenStack code.

However, addressing the needs of all of these stake holders is not possible in the time available in the Icehouse release. Therefore, we would like to suggest a focus on ensuring that end users have a good experience with Icehouse being the priority at this time. The Technical Committee encourages the DefCore committee to therefore focus on defining the set of API tests that a compliant OpenStack should pass for this release. We can then iterate in future releases to extend the requirements for trademark licensing to cover the needs of additional stake holder groups. The Technical Committee suggests that tempest is the right vehicle to implement these interoperability tests.

It is felt by the Technical Committee that whilst API testing is a good first step, it should not be sufficient to remove the requirement that:

"[distributors] include the entirety of the OpenStack [..] code from either of the latest two releases and associated milestones" [4]

This requirement should remain until the further issues outlined in this response can be addressed.

In summary, the Technical Committee proposes a focus on black box API interoperability testing for the Icehouse release, with a plan to iterate the requirements for future releases.

---

1: http://eavesdrop.openstack.org/meetings/tc/2014/tc.2014-02-04-20.02.log.html#l-332

<joshuamckenty> we need the PTLs to decide what code sections are designated sections <joshuamckenty> nova might make the "conductor" a designated section <joshuamckenty> capabilities don't have to be implemented with the same code unless that code is a designated section <joshuamckenty> e.g., neutron plugins

https://wiki.openstack.org/wiki/Governance/CoreDefinition

2: https://wiki.openstack.org/wiki/Governance/CoreDefinition (section 4)

3: http://lists.openstack.org/pipermail/openstack-dev/2014-February/026413.html

4: http://lists.openstack.org/pipermail/openstack-dev/2014-February/026559.html