fix(tests): drop module-level test calls that break local_testing collection

[mateo-berri]

Relevant issues

CircleCI failures on pipeline 80150 in langfuse_logging_unit_tests (job 1757575) and litellm_assistants_api_testing (job 1757577)

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

CI (LiteLLM team)

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🐛 Bug Fix

Changes

Both failing jobs glob tests/local_testing/**/test_*.py and then narrow with -k "langfuse" or -k "assistants". The -k filter only applies after pytest has collected every file in that glob, so anything that breaks collection of any file breaks the whole job regardless of the keyword.

tests/local_testing/test_register_model.py ended with a bare test_update_model_cost_via_completion() at module scope. That call runs while the module is being imported during collection, and the function fires a real litellm.completion(model="gpt-3.5-turbo", ...). On this run OpenAI returned a 429 (quota exceeded), the function's except block called pytest.fail(...), and pytest reported it as ERROR collecting tests/local_testing/test_register_model.py. With -x the session stopped there, so both jobs failed at collection time even though neither one actually exercises register_model.

The CircleCI reruns you linked are the "rerun from failed" attempt; their logs only show the rerun plugin choking on the recorded failure (file or directory not found: pytest.py), which is why the real cause is not visible there. The original attempt (jobs 1756105 and 1756113) shows the 429 collection error at test_register_model.py:65.

Three sibling files had the same leftover pattern (test_wandb.py, test_lunary.py, test_multiple_deployments.py). Two of them swallow exceptions so they did not error collection, but test_multiple_deployments.py also calls pytest.fail(...) in its except block, so it was one bad network call away from taking down every local_testing job the same way. pytest discovers and runs all four test_* functions on its own, so the top-level calls were dead and only harmful; this removes all four.

To keep the class of bug from coming back, test_no_top_level_test_invocations.py parses every file under tests/local_testing/ and fails if any test function is invoked at module scope. It flags all four call sites on the pre-fix tree and passes once they are gone. It is pure AST parsing with no imports or network, so it is fast and deterministic.

Screenshots / Proof of Fix

This is a pytest collection-time bug in CI, so there is no proxy endpoint to curl; the faithful reproduction is the exact collection the jobs do.

Before (the original job logs, gh tests/local_testing collection):

==================================== ERRORS ====================================
_________ ERROR collecting tests/local_testing/test_register_model.py __________
...
tests/local_testing/test_register_model.py:65: in <module>
    test_update_model_cost_via_completion()
...
E   litellm.exceptions.RateLimitError: OpenAIException - You exceeded your current quota
ERROR tests/local_testing/test_register_model.py - Failed: An error occurred: litellm.RateLimitError
!!!!!!!!!!!!!!!!!!!!!!!!!! stopping after 1 failures !!!!!!!!!!!!!!!!!!!!!!!!!!!
============================== 1 error in 19.09s ===============================

After (collection no longer imports-and-runs anything, no network call):

$ uv run --no-sync python -m pytest \
    tests/local_testing/test_register_model.py \
    tests/local_testing/test_wandb.py \
    tests/local_testing/test_lunary.py \
    tests/local_testing/test_multiple_deployments.py --collect-only -q
...
tests/local_testing/test_register_model.py::test_update_model_cost_via_completion
tests/local_testing/test_wandb.py::test_wandb_logging
tests/local_testing/test_wandb.py::test_wandb_logging_async
11 tests collected in 0.08s

Regression test green, and confirmed it flags the pre-fix tree:

$ uv run --no-sync python -m pytest tests/local_testing/test_no_top_level_test_invocations.py -q
1 passed

# scanning the HEAD (pre-fix) versions reports 4 violations:
tests/local_testing/test_register_model.py:65 calls test_update_model_cost_via_completion()
tests/local_testing/test_wandb.py:54 calls test_wandb_logging_async()
tests/local_testing/test_lunary.py:29 calls test_lunary_logging()
tests/local_testing/test_multiple_deployments.py:54 calls test_multiple_deployments()

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[mateo-berri]

Superseded by #29520; same fix, branch renamed to the litellm_ prefix to follow the internal-contributor convention

---

Generated by Claude Code