9.3 KiB
Proposals for new libraries
Quite often Oslo (as a group and project) is requested to create (or adopt) new libraries for various purposes (some relevant, some not) and it would be good to (as a group) have a stance on new libraries and the process they need to go through to either become a new library in Oslo (that the Oslo group will own and maintain going forward) or to become a new library that is independently maintained on pypi (or some hybrid of both of these approaches).
Problem Description
Assume developer Bob wants to make a new library for project(s) in OpenStack to consume.
Bob at that point has a few options when building out this library:
Keep the library independent from the Oslo project. Create library in some repository site and produce a functional and useful artifact from that library (typically a pypi release) and integrate it into the various OpenStack projects once that artifact is created.
Some advantages of this are:
- Bob can work, release, iterate at his own pace.
- Wider python community adoption can start immediately.
Some disadvantages of this are:
- Hard to define/find purpose for a library when it is not initially integrated into any project (typically to create a good library it helps to have an active user of that library for feedback).
- Gamble in that adoption (in the larger python community or in general) may not occur.
- No direct ability to take advantage of expertise (reviews, experience...) developed in the wider Oslo group (aka the creator is on his/her own).
Create library in some repository site and produce a functional and useful artifact from that library (typically a pypi release) and integrate it into the various OpenStack projects (either by oneself or with others to help) and then propose that the library be moved into the Oslo group's ownership for some reasons like it's not maintained by owner but it is used across OpenStack projects.
Some advantages of this are (including prior option's advantages):
- Continued maintenance (and reviews) of created library are now shared among the oslo reviewers (and group).
Some disadvantages of this are (including prior options disadvantages):
- Continued maintenance (and reviews) of created library are now shared among the oslo reviewers (and group). The creator may lose of control. must follow Oslo's policies.
There are also a few misconceptions about this approach that should also be cleared up (these conceptions may have been previously warranted but due to changes in the community they no longer should be deemed so):
- Adoption in oslo does not guarantee success of a library (it will still be a large amount of convincing and hard and dirty work to be successful).
- The big tent makes it easier for a library to integrate into the OpenStack ecosystems processes without having to join Oslo to also benefit from that same ecosystems processes.
Example oslo libraries that followed this workflow:
- automaton
- futurist
Create library immediately having the Oslo group or subgroup (and by default use the OpenStack ecosystem) help in its creation and produce a functional and useful artifact from that library (typically a pypi release) and integrate it into the various openstack projects.
Some advantages of this are (including prior options advantages):
- The expertise (years of experience, smart people) the Oslo group has can help guide the libraries success.
- Continued maintenance (and reviews) of created library are now shared among the Oslo reviewers (and group). The creator may lose of control, since the project is now under Oslo's purview they can approve specs and changes that creator might not agree with.
Some disadvantages of this are (including prior options disadvantages):
- Continued maintenance (and reviews) of created library are now shared among the Oslo reviewers (and group).
- Same misconceptions as stated in previous item.
Example oslo libraries that followed this workflow:
- tooz
- debtcollector
- oslo.privsep
Create library inside of a project, prove it is useful to its host project (without becoming so specific to that host project that it is not useful to anyone else) then extract that library to some repository site and produce a useful artifact. Following this further maintenance at that point will be via that repository site.
Some advantages of this are:
- Usefulness/purpose of the library will be more well-known due to integration with a host project.
Some disadvantages of this are:
- Usefulness/purpose can be too specific to host project (and further extraction/refactoring work will be needed to generalize).
Example oslo libraries that followed this workflow:
- taskflow
- oslo.versionedobjects
Create source code in a host project, at a point where another host project would benefit from that same code then extract source code into a common incubator. At that point iterate over versions in that incubator and periodically sync incubator version into host consuming projects. At some point when deemed stable extract the incubators version into a library and then produce a functional and useful artifact from that library (typically a pypi release). Following this further maintenance at that point will be via the new library repository site (and typically versioning will be followed).
Some advantages of this are:
- Usefulness/purpose of the library will be more well-known due to integration with a set of host projects.
Some disadvantages of this are:
- Typically a larger number of iterations required to extract a isolated artifact.
- Multiple rounds of non-versioned syncing and potential API collisions/conflicts (due to incubator copy/paste routine) and change iteration.
- No ability for wider python community adoption from the get go.
- Harder to cleanup and track.
- Implementations will likely diverge and the amount of person time required to keep in sync.
Example oslo libraries that followed this workflow:
- oslo.config
- oslo.cache
- oslo.concurrency
- oslo.db
- oslo.log
- oslo.messaging
- oslo.policy
- oslo.serialization
- ...
Bob will also have to pick which repository site he will use. For sake of this document we will assume the majority will choose to use the OpenStack ecosystems gerrit review system and git based hosting system (but Bob if he desires can use something like github and pull requests if he so chooses, as long as Bob takes into consideration that doing this will be make it harder to get contributions from folks in the OpenStack community).
Proposed Policy
In order to help Bob (or other author) make a smart selection from the options listed above in the problem statement we as a group (who has made these decisions many times previously) would like to help new libraries (and their authors) become successful by having new library proposals go through a sort of litmus test that we as a group believe will help library creators figure out which of the above listed options will be better suited for them (and be better suited for their own library's future success).
To aid in this process we as a group believe that when Bob (or other author) starts to figure out which of the options he (or she) will take it would be best for that developer to fill out the template new-library-template.rst and post it for review on the openstack-discuss mailing list with the [oslo] tag in the subject. And then let the mailing list figure out which of the above options will best work for the authors and the community as a whole). This same information should also be proposed to the oslo-specs repository itself (if/when the mailing list agrees that it should be a new oslo library).
In order to make the new oslo library healthy and continuous development, new core contributors for that adopted library are needed, it needs at least two individuals from the community committed to triaging and fixing bugs, and responding to test failures in a timely manner.
Alternatives & History
The other options are more along the line of what the Oslo group has already done which is to have a sort of impromptu and tribal knowledge around the area of new libraries and the options available to developers wanting (and/or willing) to make new libraries. This policy will aim to solidify that knowledge into a document that can be easily referenced and agreed upon.
Implementation
Author(s)
Primary author: harlowja
Other contributors: gcb
Milestones
Pike
Work Items
- Draft policy
- Get feedback on policy
- Repeat
- Approve policy
References
Revision History
| Release Name | Description |
|---|---|
| Pike | Introduced |
Note
This work is licensed under a Creative Commons Attribution 3.0 Unported License. http://creativecommons.org/licenses/by/3.0/legalcode