test: unwind environment writes in tests/test_litellm with monkeypatch

[yuneng-berri]

TLDR

Problem this solves:

• 768 tests write straight into os.environ
• The value outlives the test, in that worker
• So a later test reads a key it never set
• TQ004 has sat at its seed since the rule shipped

How it solves it:

• 262 of those writes become monkeypatch.setenv, which unwinds at teardown
• Ratchet the TQ004 ceiling from 768 to 568

User Flow

No end-user behavior changes and no test behavior changes. The same 1,732 tests
pass, and one fewer environment variable survives the session.

Relevant issues

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• The handful of test files covering my change pass locally
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review

Screenshots / Proof of Fix

Shared setup: a leak counter, run as a pytest plugin, so the two sides are
comparable rather than asserted.

import os

_before = {}


def pytest_sessionstart(session):
    _before.update(os.environ)


def pytest_sessionfinish(session, exitstatus):
    added = sorted(set(os.environ) - set(_before))
    changed = sorted(k for k in _before if k in os.environ and os.environ[k] != _before[k])
    print("\nENV LEAKED AFTER SESSION: %d added, %d changed" % (len(added), len(changed)))
    for k in added:
        print("  + %s" % k)
    for k in changed:
        print("  ~ %s" % k)

Both sides run the same 38 files, in the same order, with the same command:

python -m pytest $(cat tq004_files.txt | tr '\n' ' ') -p no:randomly -q -p envleak

Before (ff02d5c)

The rule's count

1. python scripts/check_test_quality.py tests | grep -c ' TQ004 '

768

What the session leaves behind

1. The 38 files, run in order:

1469 passed, 15 warnings in 336.70s (0:05:36)

ENV LEAKED AFTER SESSION: 2 added, 1 changed
  + APISERPENT_API_KEY
  + LITELLM_LOCAL_MODEL_COST_MAP
  ~ LITELLM_LICENSE

2. LITELLM_LICENSE was already set on the machine, and a test overwrote it for
   every test that ran after it. APISERPENT_API_KEY was invented by a test and
   left behind

After (7219e7a)

The rule's count

1. python scripts/check_test_quality.py tests | grep -c ' TQ004 '

568

What the session leaves behind

1. The same 38 files, same order:

1469 passed, 15 warnings in 388.93s (0:06:28)

ENV LEAKED AFTER SESSION: 0 added, 0 changed

2. Nothing is left behind. An earlier revision of this PR read the remaining
   LITELLM_LOCAL_MODEL_COST_MAP as conftest-owned; it was not, it came from
   the two cost-calc files that now belong to test(cost-calc): stop 182 global writes leaking out of the cost-calc suites #37815

The ratchet

1. python scripts/test_quality_gate.py --update

Ratcheted TQ-rule limits down by 262 violations this branch fixed

2. python scripts/test_quality_gate.py --base origin/litellm_internal_staging

OK: every TQ rule is within its test-suite ceiling (base origin/litellm_internal_staging)

Type

✅ Test

Caveats (if any)

• Four shapes are skipped, each because the fixture would not bind
• A mock.patch family decorator wraps the test, so pytest may not inject
• A defaulted positional would shift, so the value lands wrong
• A nested def can shadow the name, so the inner one wins
• A test called directly by another test gets no fixture at all
• 506 writes remain, mostly inside those four shapes
• Five pop-then-restore tests now use delenv, so nothing is dropped

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[greptile-apps]

Greptile Summary

This PR replaces persistent test environment writes with pytest monkeypatch operations and lowers the corresponding test-quality budget.

• Converts direct environment assignments and deletions across the LiteLLM test suite to teardown-safe monkeypatch operations.
• Correctly fixes restoration of pre-existing RESEND_API_KEY, SENDGRID_API_KEY, and UI_PASSWORD values.
• Reduces the TQ004 ceiling from 768 to 568.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported environment-restoration issue is fixed at the current head.

Important Files Changed

Filename	Overview
test-quality-budget.json	Lowers the TQ004 ceiling to reflect the direct environment writes removed by this PR.
tests/test_litellm/enterprise/enterprise_callbacks/send_emails/test_resend_email.py	Uses monkeypatch.delenv so RESEND_API_KEY is absent during the test and its inherited state is restored afterward.
tests/test_litellm/enterprise/enterprise_callbacks/send_emails/test_sendgrid_email.py	Uses monkeypatch.delenv to test missing-key behavior without losing a pre-existing SENDGRID_API_KEY.
tests/test_litellm/proxy/auth/test_login_utils.py	Removes UI_PASSWORD through monkeypatch inside the patched environment, preserving fallback coverage and restoring the worker’s original state.

_{Reviews (3): Last reviewed commit: "chore(test): leave the two cost-calc fil..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[yuneng-berri]

Fixed: all five pop-then-restore sites now use monkeypatch.delenv, so the inherited value survives teardown. @greptileai

[yuneng-berri]

@greptileai

[yuneng-berri]

@greptileai please review the current head, no review has landed on it yet