Python: [RFC] Syncing Kubernetes Client versions with upstream Kubernetes versions

[palnabarun]

Current Scenario

The Kubernetes Python client follows a versioning schema x.y.z{a|b}N where x is Kubernetes Minor Version - 4. For example, the Kubernetes Python client 12.0.0 is based on Kubernetes 1.16.

Reasons

To make the numbering coherent and reduce confusion.

What other clients do

• Go: client-go used to use a versioning scheme like us, but they eventually moved to semver compatible versions - v0.y.z where y and z are from Kubernetes versions 1.y.z. Ref1 Ref2
• Java: A similar schema like but the Java client major version number is Kubernetes minor version number - 8. For example, Java client 9.0.0 is based on Kubernetes 1.17.

Proposed Versioning Schemes

We can have a versioning scheme similar to client-go. Kubernetes 1.x.y would correspond to Python client 0.x.y.
Or, the versions can be 1.x.y exactly equal to the Kubernetes versions. This results in us being not able to do our own patch releases since there are changes done on the client code too. Also, the client-go adopted the conventions they have currently because of certain limitations with Go Modules, which we don't have. Moving back version numbers is also detrimental since pip install kubernetes without a version number can

Another option we have is to have client releases versioned as x.y.p where x is the Kubernetes Minor release number, y is the Kubernetes patch release number, and p is the Python client patch specific number. In order to achieve this option, client releases henceforth will jump a few version numbers to achieve the coherency, and the same needs to be documented in the README and CHANGELOG along with proper communication to k-dev when the release happens.

Resolution

Based on the discussion in the bi-weekly meeting, it looks like the latter option is preferable.

Action Items

• [x] Create release-17.0 branch for client based on Kubernetes 1.17.p @roycaihw / @yliaog
• [x] Create release-18.0 branch for client based on Kubernetes 1.18.p @roycaihw / @yliaog
• [x] Create release-19.0 branch for client based on Kubernetes 1.19.p @roycaihw / @yliaog
• [x] Write in the Python Client 17.0.0 release notes about this change in versioning scheme @palnabarun
• [ ] Write in the Python Client 18.0.0 release notes about this change in versioning scheme @palnabarun
• [ ] Write in the Python Client 19.0.0 release notes about this change in versioning scheme @palnabarun
• [ ] Close this issue after all of the above has been done. @palnabarun

Edits

• Updated the issue after the discussion on the same on 14th September.
• Updated the action items to include the creation of branches

/assign

[palnabarun]

cc: @roycaihw @yliaog

[palnabarun]

/kind design
/remove-kind documentation

[roycaihw]

Using 0.x.y corresponding to Kubernetes 1.x.y has the following downsides:

• The python client version goes backwards (from 11.0.0 to 0.x.y). People running pip install --upgrade may not be able to pick up the latest versions.
• Sometimes we do patch releases in this client for bug fixes (e.g. 8.0.1, 8.0.2). We don't have digits left to number our own patches.

[roycaihw]

(For the record:) The proposal was accepted in the bi-weekly client-python meeting, and has been implemented for client v17: https://github.com/kubernetes-client/python#homogenizing-the-kubernetes-python-client-versions. Please feel free to close this issue once the action items are finished.