fix(vertex): resolve blocking I/O in async contexts

[smart-cau]

Description

Fixes blocking I/O errors when using ChatVertexAI in async contexts (e.g., LangGraph applications). The issue occurred because importlib.metadata.version() performs blocking file I/O during async client initialization, which blocks the event loop and triggers BlockingError in async-aware environments.

Solution: Cache package version at module import time instead of reading it on every get_user_agent() call. This eliminates blocking I/O in async contexts while improving performance.

Relevant issues

Related to #1231
Related to #873

Type

🐛 Bug Fix

Changes

Core Changes

• libs/vertexai/langchain_google_vertexai/_utils.py
  • Added module-level _LANGCHAIN_VERTEXAI_VERSION cache
  • Modified get_user_agent() to use cached version

Test Changes

• libs/vertexai/tests/unit_tests/test_utils.py
  • Added test_version_is_cached_at_module_level()
  • Added test_get_user_agent_no_blocking_in_async_context()
  • Added test_async_context_execution()
  • Updated existing tests to use cached version

Testing

Unit Tests

• ✅ All existing unit tests pass in libs/vertexai
• ✅ New tests verify version caching behavior
• ✅ New tests ensure no blocking I/O in async contexts

Test Commands Run

cd libs/vertexai && make test

Note

This PR is part of the fix for issue #1231 and #873. A companion PR (#1264) addresses the same issue in langchain-google-genai.

The solution:

• ✅ No breaking changes to public APIs
• ✅ Backward compatible
• ✅ Performance improvement (version read only once)
• ✅ Zero overhead for subsequent calls
• ✅ Safe in all contexts (sync and async)