GKE code executor unit tests fail with kubernetes 37.0.0, turning main CI red #7443

🔴 Required Information

Describe the Bug:

Since kubernetes 37.0.0 was published to PyPI (2026-10-07, 00:51 UTC), five tests in tests/unittests/code_executors/test_gke_code_executor.py fail. The Unit Tests jobs of the Continuous Integration workflow have failed on main for Python 3.11–3.14 on every push since 51a20e2, because CI installs the newest version kubernetes>=29 allows.

kubernetes 37 validates model fields with pydantic. In these tests create_namespaced_job returns the default MagicMock, and GkeCodeExecutor._add_owner_reference copies that Job's api_version, kind, metadata.name and metadata.uid into a V1OwnerReference, which now raises a ValidationError. execute_code catches it and returns "An unexpected executor error occurred", so the assertions fail.

Only the tests are affected. Against a real cluster, create_namespaced_job returns a V1Job with string fields. The other 13 tests in the file pass on 37.0.0, including the one that builds a real Job manifest.

Steps to Reproduce:

1. pip install -e . pytest pytest-asyncio pytest-mock "kubernetes==37.0.0"
2. pytest tests/unittests/code_executors/test_gke_code_executor.py

Expected Behavior:

All 18 tests pass, as they do with kubernetes 36.0.3 (the version pinned in constraints-3.*.txt).

Observed Behavior:

FAILED ...::test_execute_code_success - AssertionError: assert '' == 'hello world'
FAILED ...::test_execute_code_job_failed - assert 'Job failed. Logs:' in "An unexpected executor error occurred: 4 validation errors for V1OwnerReference ..."
FAILED ...::test_execute_code_job_failed_without_terminated_container - assert None == 1
FAILED ...::test_execute_code_timeout - assert 'Executor timed out' in "An unexpected executor error occurred: 4 validation errors for V1OwnerReference ..."
FAILED ...::test_execute_code_forks_to_job - AssertionError: assert '' == 'job stdout'
5 failed, 13 passed

The underlying error:

4 validation errors for V1OwnerReference
api_version
  Input should be a valid string [type=string_type, input_value=<MagicMock name='mock.api...ersion' id='...'>, input_type=MagicMock]
kind
  Input should be a valid string [type=string_type, input_value=<MagicMock name='mock.kind' id='...'>, input_type=MagicMock]
name
  Input should be a valid string [type=string_type, input_value=<MagicMock name='mock.met...a.name' id='...'>, input_type=MagicMock]
uid
  Input should be a valid string [type=string_type, input_value=<MagicMock name='mock.met...ta.uid' id='...'>, input_type=MagicMock]

Environment Details:

• ADK Library Version: main (2.11.0), since 51a20e2
• Desktop OS: Linux (GitHub Actions) and macOS
• Python Version: 3.11, 3.12, 3.13 and 3.14

Model Information:

• Are you using LiteLLM: N/A
• Which model is being used: N/A, no model is involved

🟡 Optional Information

Regression:

Yes, but the regression comes from the dependency release, not from an ADK commit. The same tests pass with kubernetes 36.0.3.

Logs:

CI run on main: https://github.com/google/adk-python/actions/runs/37674194847

Additional Context:

constraints-3.*.txt already pin kubernetes==36.0.3, but the unit test job installs without them. A fix that keeps the tests passing on both versions is to have the mocked create_namespaced_job return a real client.V1Job with string fields. I'll open a PR.

How often has this issue occurred?:

• Always (100%)