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

Project	Deployment	Actions	Updated (UTC)
litellm	Error		Jan 31, 2026 10:59pm

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 Overview

Greptile Summary

Changes Overview

This PR fixes a test reliability issue where tests using @respx.mock were making real API calls instead of using mocked responses. The root cause was that litellm uses aiohttp transport by default, which bypasses respx's httpx-only interception.

What Changed

1. Added disable_aiohttp_transport fixture: Five test files now include an autouse fixture that:

   • Sets litellm.disable_aiohttp_transport = True to force httpx transport
   • Clears the client cache via in_memory_llm_clients_cache.flush_cache() to prevent reuse of old aiohttp clients
   • Properly restores the original state after tests complete

2. Fixed isinstance checks in vector store tests: Replaced direct isinstance checks with type name comparisons (type(obj).__name__ == "ClassName") to avoid module identity issues caused by sys.path.insert in test setup.

Technical Analysis

The fix is well-designed and addresses the regression from PR #19829. By forcing httpx transport, respx can now properly intercept HTTP calls in these test files:

• test_resend_email.py - Email sending tests
• test_sendgrid_email.py - Email sending tests
• test_main.py - Image edit header merging test
• test_vector_store_registry.py - Vector store registry credential tests
• test_vector_store_create_provider_logic.py - Vector store provider logic tests

The fixture implementation follows best practices by:

• Using getattr with None default for safe attribute access
• Properly restoring original values or deleting the attribute if it didn't exist
• Clearing the client cache to ensure fresh clients are created with the new transport setting

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are isolated to test files only, properly implement cleanup/restoration logic, and directly address the identified regression. The fixture pattern is consistent across all modified files and follows pytest best practices.
• No files require special attention

Sequence Diagram

sequenceDiagram
    participant Test as Test Execution
    participant Fixture as disable_aiohttp_transport
    participant LiteLLM as litellm module
    participant Cache as in_memory_llm_clients_cache
    participant Respx as @respx.mock
    participant HTTPX as httpx transport
    
    Note over Test,HTTPX: Before PR (Broken)
    Test->>Respx: Start test with @respx.mock
    Respx->>LiteLLM: Intercept HTTP calls
    LiteLLM->>LiteLLM: Use aiohttp transport (default)
    Note over Respx,LiteLLM: respx cannot intercept aiohttp!
    LiteLLM-->>Test: Real API call made ❌
    
    Note over Test,HTTPX: After PR (Fixed)
    Test->>Fixture: autouse=True fixture runs
    Fixture->>LiteLLM: Set disable_aiohttp_transport = True
    Fixture->>Cache: flush_cache() to clear old clients
    Cache-->>Fixture: Cache cleared
    Fixture->>Test: Setup complete
    
    Test->>Respx: Start test with @respx.mock
    Respx->>LiteLLM: Intercept HTTP calls
    LiteLLM->>HTTPX: Use httpx transport (forced)
    Respx->>HTTPX: Mock HTTP request
    HTTPX-->>Respx: Return mocked response
    Respx-->>Test: Test passes with mock ✓
    
    Test->>Fixture: Test complete, teardown
    Fixture->>LiteLLM: Restore original disable_aiohttp_transport
    Fixture-->>Test: Cleanup complete