Ghanshyam Mann 70f7b89e58 Resolution to drop the lower constraints maintenance

In zed PTG, TC agreed to drop the lower constraints
and their testing
- https://etherpad.opendev.org/p/tc-zed-ptg#L326

Change-Id: Id14083a00014ac6632f16f6700e24d0d54a0d903

2022-04-21 13:56:33 -05:00

4.1 KiB

Raw Blame History

2022-04-14 Drop Lower Constraints Maintenance

History

OpenStack has tried to maintain the lower bounds of dependency in lower-constraints.txt file and also testing them in unit test jobs on every project's repository. This was the best effort to tell that these are the minimum versions of the dependencies you should have in your environment for a particular OpenStack release.

Lower bounds of dependencies are present in lower-constraints.txt as well as in requirements.txt file. The former one is used in lower constraints unit testing job but latter one is not tested at all.

Challenges

Lower constraints are not easy to test and they need more maintenance to keep them correct. In every distro upgrade in our testing, we spend most of the time fixing the lower constraints testing. In the Yoga cycle, it was broken by pip's latest resolver feature¹. From there discussion started on whether it is worthwhile to maintain and test these lower constraints² ³ ⁴.
We do not test them completely. The lower constraints are tested only in unit tests and we are not even sure that lower constraints maintained by one project are compatible with lower constraints maintained by the other dependent/related project. We can only manage the OpenStack direct dependencies lower bound since indirect dependencies, e.g. from packages outside of OpenStack, do not maintain compatible lower bounds.
We do not test them correctly. Before pip's coherent dependency resolver, we were testing with versions of dependencies that were different from what we thought they were⁵. Adapting an individual project's lower-constraints.txt files to pass the coherent resolver was a lot of work, and revealed that there were problems maintaining transitive dependencies in the lower-constraints.txt (we were not systematically tracking when dependencies added new dependencies).
In the current situation, when we migrated the testing of lower constraints to min python version supported (python 3.8 on ubuntu focal) then it was failing on stable branches. This kind of python version migration requires a separate job to maintain for stable branches and master lower constraints testing.
To check lower constraints testing worthiness, TC did check who uses the lower constraints and it seems only Debian package maintainers use them as a reference and the rest of the distro/packagers use upper constraints⁶. Based on that, TC agreed to explicitly clarify that the testing of lower constraints is optional and up to projects to keep or drop it⁷. And many projects like Oslo have dropped it entirely, Neutron dropped it from all stable branches.

In summary, lower constraints and their testing is half baked and maintaining them without correct and complete testing does not make sense. Distros can always test the versions of their packages with OpenStack master code.

Proposal

In Zed cycle PTG⁸, after discussing all the challenges mentioned above, TC agreed on the below points:

Keep the lower bounds in the requirements.txt file with some statements at the top of this file saying that these are not confirmed or tested lower bounds and we try our best to keep them updated.
Drop the lower-constraints.txt file, its env, and testing from master as well as from stable branches.

4.1 KiB Raw Blame History

2022-04-14 Drop Lower Constraints Maintenance

History

Challenges

Proposal

4.1 KiB

Raw Blame History