fix(test): prevent flaky failure in test_log_langfuse_v2_handles_null_usage_values

[jquinter]

Summary

This test has repeatedly failed in CI (previously fixed in #17599, #20475, #21093) with:

AssertionError: Expected '_add_prompt_to_generation_params' to have been called once. Called 0 times.

• _add_prompt_to_generation_params is only reached when _supports_prompt() returns True
• Under cross-test state contamination in CI (parallel workers running in the same process), langfuse_sdk_version can end up in an unexpected state
• _supports_prompt() then returns False, silently skipping the call (the outer try/except Exception in _log_langfuse_v2 swallows any exception)
• The previous fix used reset_mock() without side_effect=True, so setUp's trace.side_effect persisted and overrode the return_value set afterward

Fixes:

• Use reset_mock(side_effect=True) to clear setUp's side effects so the explicit return_value assignments actually take effect
• Add patch.object(self.logger, "_supports_prompt", return_value=True) to make the _add_prompt_to_generation_params assertion fully independent of SDK version state contamination

Test plan

• poetry run pytest tests/test_litellm/integrations/test_langfuse.py::TestLangfuseUsageDetails -v — all 7 tests pass

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 17, 2026 10:19pm

[greptile-apps]

Greptile Summary

This PR fixes a repeatedly flaky test (test_log_langfuse_v2_handles_null_usage_values) that has failed intermittently in CI due to cross-test state contamination. Two root causes are addressed:

• reset_mock(side_effect=True): The previous reset_mock() calls (without side_effect=True) did not clear the side_effect set in setUp(), so the subsequent return_value assignments on lines 281-284 were silently ignored. This caused trace() to not return the expected mock, leading to downstream assertion failures.
• patch.object(self.logger, "_supports_prompt", return_value=True): The _add_prompt_to_generation_params assertion depends on _supports_prompt() returning True, which in turn depends on langfuse_sdk_version. Under parallel CI workers, this global state could be contaminated. Patching _supports_prompt directly makes the assertion fully independent of SDK version state.

Both fixes are minimal, well-targeted, and don't affect any production code. The test file is mock-only, consistent with the project rule requiring no real network calls in this test directory.

Confidence Score: 5/5

• This PR is safe to merge — it only modifies a single test method with well-understood mock behavior fixes.
• The changes are limited to a single test method, address clearly documented flaky behavior (referenced in 3 prior PRs), and both fixes are correct: reset_mock(side_effect=True) properly clears side_effect to allow return_value, and patching _supports_prompt isolates the test from global state. No production code is touched.
• No files require special attention.

Important Files Changed

Filename	Overview
tests/test_litellm/integrations/test_langfuse.py	Two targeted fixes to prevent flaky test: (1) reset_mock(side_effect=True) clears setUp's side_effect so return_value assignments take effect, (2) patch.object(self.logger, "_supports_prompt", return_value=True) isolates the test from SDK version state contamination. Both changes are correct and well-scoped.

Flowchart

flowchart TD
    A[test_log_langfuse_v2_handles_null_usage_values] --> B[reset_mock - side_effect=True]
    B --> C[Re-setup mock chain with return_value]
    C --> D[patch _supports_prompt → True]
    D --> E[_log_langfuse_v2 called]
    E --> F{_supports_prompt?}
    F -->|True - patched| G[_add_prompt_to_generation_params called]
    G --> H[trace.generation called]
    H --> I[Assert usage values converted None → 0]
    I --> J[Assert _add_prompt_to_generation_params called once]

    style B fill:#90EE90,stroke:#333
    style D fill:#90EE90,stroke:#333
    style F fill:#FFD700,stroke:#333

Loading

_{Last reviewed commit: 703f02b}

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Greptile Summary

This PR fixes a repeatedly flaky test (test_log_langfuse_v2_handles_null_usage_values) that has failed in CI multiple times due to cross-test state contamination when running with parallel workers.

• reset_mock(side_effect=True): The setUp method sets mock_langfuse_client.trace.side_effect to _trace_side_effect. Previously, reset_mock() was called without side_effect=True, which left the side effect in place, overriding the return_value assignments that followed. This is fixed by passing side_effect=True to all three reset_mock() calls.
• patch.object(self.logger, "_supports_prompt", return_value=True): The _add_prompt_to_generation_params function is only called when _supports_prompt() returns True, which depends on langfuse_sdk_version. Under parallel CI, this attribute can be contaminated by other tests. Patching _supports_prompt directly makes the assertion independent of global state.
• Minor whitespace cleanup (trailing whitespace removed on one line).

The changes are minimal, well-targeted, and correctly address the two root causes of the flaky test.

Confidence Score: 5/5

• This PR is safe to merge — it only modifies test code with two well-reasoned fixes for a known flaky test.
• The changes are limited to a single test file, are minimal in scope, and correctly address two documented root causes of CI flakiness. No production code is modified. The fixes are consistent with Python unittest.mock semantics.
• No files require special attention.

Important Files Changed

Filename	Overview
tests/test_litellm/integrations/test_langfuse.py	Two targeted fixes for flaky test: (1) reset_mock(side_effect=True) to properly clear setUp's _trace_side_effect so return_value assignments take effect, and (2) patch.object on _supports_prompt to decouple the _add_prompt_to_generation_params assertion from global SDK version state. Both changes are correct and well-scoped.

Flowchart

flowchart TD
    A["test_log_langfuse_v2_handles_null_usage_values"] --> B["reset_mock(side_effect=True)"]
    B --> C["Clear setUp's _trace_side_effect"]
    C --> D["Set return_value on trace mock"]
    D --> E["patch _supports_prompt → True"]
    E --> F["Call _log_langfuse_v2"]
    F --> G{"_supports_prompt()"}
    G -->|"True (patched)"| H["_add_prompt_to_generation_params called"]
    H --> I["Assert generation called with correct usage values"]
    I --> J["Assert _add_prompt_to_generation_params called once"]
    G -->|"False (unpatched, flaky)"| K["_add_prompt_to_generation_params skipped"]
    K --> L["❌ AssertionError: Called 0 times"]

Loading

_{Last reviewed commit: 703f02b}

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}