[yurnov]

What type of PR is this?

/kind bug
/kind regression

What this PR does / why we need it:

Since v37.0.0, from kubernetes.client.apis import core_v1_api raises ImportError, which breaks code that still uses the deprecated alias, for example k8s_exec in ansible-collections/kubernetes.core#1273. The alias star-imports kubernetes.client.api, which now defines __all__ with class names only and loads its submodules lazily (#2631, #2652), so the submodule names are no longer re-exported. This adds a module-level __getattr__ (PEP 562) to the alias that resolves submodule names on demand, plus a regression test.

Which issue(s) this PR fixes:

Fixes #2726

Special notes for your reviewer:

• The alias is hand-written, so the fix survives regeneration: client/apis/__init__.py is not listed in [kubernetes/.openapi-generator/FILES][files], and gen only removes client/api, client/models and docs before generating ([CLEANUP_DIRS][cleanup], [removed one by one with rm -rf][rm]). The file hasn't changed since 2019 and was kept through the v7.24.0 and v7.25.0 regenerations. The new test sits next to the other hand-written tests in kubernetes/test/.
• Why __getattr__ rather than binding every submodule into the alias namespace with a pkgutil.iter_modules() loop: it is what the issue proposed, it keeps the change small, and type checkers can see it. With this change mypy resolves from kubernetes.client.apis import core_v1_api to ModuleType. On master it reports attr-defined, and dynamically bound names would still get that error.
• Unknown names raise AttributeError, so from ... import raises ImportError and hasattr() returns False. A ModuleNotFoundError for any other module, such as a missing dependency inside an API module, propagates unchanged. Private and non-identifier names are rejected without an import attempt, so probes like __wrapped__ or the __all__ lookup of a star import don't try to import anything.
• Remaining differences from 36.x: from kubernetes.client.apis import * does not export the submodule names, and dir(kubernetes.client.apis) does not list them. I left those as they are for a deprecated alias. Dotted imports (import kubernetes.client.apis.core_v1_api) didn't work in 36.x either.
• Importing the alias costs the same as before. Its star import already loads every API submodule through the lazy __getattr__ of kubernetes.client.api.
• The new test fails on master (test_from_import_submodule, all 66 subtests of test_every_api_submodule_is_reachable, and the two tests for the new guards) and passes with the fix. Other unit test results are the same as on master. Tested on Python 3.10, 3.12 and 3.14.
• Could this be cherry-picked to release-37.0 for a 37.0.1? The commit applies cleanly there, and the new test passes on it.

AI Disclosure
Parts of this PR (including complete unit tests) were generated using Anthropic Sonnet 5.5. All generated code has been manually reviewed, tested, and verified for accuracy.

Does this PR introduce a user-facing change?

Fixed importing API submodules such as `core_v1_api` from the deprecated `kubernetes.client.apis` package, which raised `ImportError` since v37.0.0.

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

N/A

[yliaog]

/lgtm
/approve

[kubernetes-prow]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: yliaog, yurnov

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [yliaog]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[yurnov]

Thanks @yliaog for the review. Can I request to cherry-pick it to the release-37 branch and plan a patch release, i.e., 37.0.1?

[yliaog]

Yes, that will done later today or early tomorrow.