Litellm stability fix v2

[ishaan-jaff]

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• 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)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Error		Feb 28, 2026 11:29pm

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR improves CI test stability across multiple test suites by making spend-data assertions resilient to slow DB writes and fixing test mocking for auth-related changes.

• Anthropic passthrough tests: Replaces hard assertions on spend data with defensive null/type checks and graceful skip when data is unavailable in CI
• Gemini spend tests (JS): Adds polling with up to 6 retries (10s intervals), jest.retryTimes(3), increases timeout from 25s to 90s, and gracefully skips spend assertions if data never arrives
• MCP server test: Fixes test_sse_mcp_handler_mock by mocking extract_mcp_auth_context and set_auth_context so the handler reaches sse_session_manager.handle_request
• Files endpoint test: Adds proper auth mocking (master_key, prisma_client, user_api_key_auth dependency override) with try/finally cleanup
• Parallel request limiter test: Updates @pytest.mark.flaky from reruns=3 to retries=3, delay=2 for pytest-rerunfailures v14 compatibility (though 2 other tests in the same file still use the old parameter)
• Docs: Adds CURSOR_API_BASE environment variable documentation

Confidence Score: 4/5

• This PR is safe to merge — it only modifies test files and documentation, with no production code changes.
• All changes are test-only fixes and a minor docs addition. The fixes are well-structured: proper auth mocking, defensive spend data handling, and correct pytest plugin parameter usage. One minor inconsistency with flaky markers in the same file was noted.
• tests/test_litellm/proxy/hooks/test_parallel_request_limiter_v3.py has inconsistent @pytest.mark.flaky parameters — only one of three tests was updated to the new syntax.

Important Files Changed

Filename	Overview
docs/my-website/docs/proxy/config_settings.md	Adds documentation for the CURSOR_API_BASE environment variable. Straightforward doc addition, no issues.
tests/mcp_tests/test_mcp_server.py	Adds mocks for extract_mcp_auth_context and set_auth_context to fix the SSE MCP handler test. Mock return value correctly matches the 6-tuple signature. Properly uses AsyncMock for async functions.
tests/pass_through_tests/test_anthropic_passthrough.py	Replaces hard assertions on spend data with graceful skip logic. Defensive checks are thorough (None, type, length, key existence). Tests can now silently pass when spend logging is slow in CI, which trades test coverage for stability.
tests/pass_through_tests/test_gemini_with_spend.test.js	Adds polling with retries for spend data and jest.retryTimes(3) for flaky test resilience. Timeout increased from 25s to 90s. Graceful skip when spend data unavailable. Duplicated polling logic could be extracted into a helper.
tests/test_litellm/proxy/hooks/test_parallel_request_limiter_v3.py	Updates @pytest.mark.flaky(reruns=3) to @pytest.mark.flaky(retries=3, delay=2) for one test, aligning with pytest-rerunfailures v14 parameter names. However, two other tests in the same file still use the old reruns=3 parameter.
tests/test_litellm/proxy/openai_files_endpoint/test_files_endpoint.py	Properly fixes the files endpoint test by mocking master_key/prisma_client and overriding user_api_key_auth FastAPI dependency. Wraps test body in try/finally for cleanup. Follows FastAPI testing best practices.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[CI Test Run] --> B{Spend Data Available?}
    B -->|Yes| C[Run Spend Assertions]
    B -->|No| D[Poll with Retries]
    D --> E{Data arrived?}
    E -->|Yes| C
    E -->|No after retries| F[Graceful Skip - return early]
    C --> G[Test Passes]
    F --> G

    H[MCP SSE Test] --> I[Mock extract_mcp_auth_context]
    I --> J[Mock set_auth_context]
    J --> K[Verify sse_session_manager.handle_request called]

    L[Files Endpoint Test] --> M[Mock master_key & prisma_client]
    M --> N[Override user_api_key_auth dependency]
    N --> O[Run file upload assertions]
    O --> P[Cleanup dependency overrides]

Loading

_{Last reviewed commit: 863d070}

[greptile-apps]

_{6 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Inconsistent flaky marker parameters in same file

This test's marker was updated from reruns=3 to retries=3, delay=2 to align with pytest-rerunfailures v14, but two other tests in this same file still use the old reruns=3 parameter (line 44 for test_sliding_window_rate_limit_v3 and line 248 for test_normal_router_call_tpm_v3). For consistency and correctness, those should also be updated to retries=3, delay=2.