BlockingError when using ChatGoogleGenerativeAI with async in LangGraph dev

[Trinkes]

Bug: BlockingError when using ChatGoogleGenerativeAI with async in LangGraph dev

Description

When using ChatGoogleGenerativeAI with async/await in a LangGraph application, langgraph dev throws a blockbuster.blockbuster.BlockingError due to blocking I/O calls during async client initialization. The blocking call occurs when langchain_google_genai reads package metadata using importlib.metadata.version().

Environment

• langchain-google-genai: 2.1.12
• langgraph: 0.6.8
• langgraph-api: 0.4.37
• langgraph-cli: 0.4.2
• Python: 3.13.5

Minimal Reproducible Example

# minimal_example.py
from langchain_google_genai import ChatGoogleGenerativeAI
from langgraph.graph import StateGraph, MessagesState


async def my_node(state: MessagesState):
    llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", api_key="YOUR_API_KEY")
    response = await llm.ainvoke(state["messages"])
    return {"messages": [response]}


graph = StateGraph(MessagesState)
graph.add_node("chat", my_node)
graph.set_entry_point("chat")
graph.set_finish_point("chat")

minimal_graph = graph.compile()

// langgraph.json
{
  "dependencies": ["."],
  "graphs": {
    "minimal": "./minimal_example.py:minimal_graph"
  },
  "env": ".env"
}

Steps to reproduce:

1. Run langgraph dev
2. Invoke the graph

Stack Trace

Traceback (most recent call last):
File "./.venv/lib/python3.13/site-packages/langgraph_api/worker.py", line 142, in wrap_user_errors
await consume(
    stream, run_id, resumable, stream_modes, thread_id=run["thread_id"]
)
File "./.venv/lib/python3.13/site-packages/langgraph_api/stream.py", line 501, in consume
raise e
File "./.venv/lib/python3.13/site-packages/langgraph_api/stream.py", line 484, in consume
async for mode, payload in stream:
    ...<6 lines>...
)
File "./.venv/lib/python3.13/site-packages/langgraph_api/stream.py", line 369, in astream_state
event = await wait_if_not_done(anext(stream, sentinel), done)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        File "./.venv/lib/python3.13/site-packages/langgraph_api/asyncio.py", line 89, in wait_if_not_done
raise e.exceptions[0] from None
File "./.venv/lib/python3.13/site-packages/langgraph/pregel/main.py", line 2976, in astream
async for _ in runner.atick(
        ...<13 lines>...
yield o
File "./.venv/lib/python3.13/site-packages/langgraph/pregel/_runner.py", line 401, in atick
_panic_or_proceed(
~~~~~~~~~~~~~~~~~^
futures.done.union(f for f, t in futures.items() if t is not None),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
timeout_exc_cls=asyncio.TimeoutError,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
panic=reraise,
^^^^^^^^^^^^^^
)
^
File "./.venv/lib/python3.13/site-packages/langgraph/pregel/_runner.py", line 511, in _panic_or_proceed
raise exc
File "./.venv/lib/python3.13/site-packages/langgraph/pregel/_retry.py", line 137, in arun_with_retry
return await task.proc.ainvoke(task.input, config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/langgraph/_internal/_runnable.py", line 706, in ainvoke
input = await asyncio.create_task(
        ^^^^^^^^^^^^^^^^^^^^^^^^^^
        step.ainvoke(input, config, **kwargs), context=context
                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "./.venv/lib/python3.13/site-packages/langgraph/_internal/_runnable.py", line 474, in ainvoke
ret = await self.afunc(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "./.claude/tmp/minimal_example.py", line 7, in my_node
response = await llm.ainvoke(state["messages"])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 417, in ainvoke
llm_result = await self.agenerate_prompt(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
             ...<8 lines>...
)
^
File "./.venv/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 1036, in agenerate_prompt
return await self.agenerate(
       ^^^^^^^^^^^^^^^^^^^^^
       prompt_messages, stop=stop, callbacks=callbacks, **kwargs
                                                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "./.venv/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 994, in agenerate
raise exceptions[0]
File "./.venv/lib/python3.13/site-packages/langchain_core/language_models/chat_models.py", line 1153, in _agenerate_with_cache
async for chunk in self._astream(messages, stop=stop, **kwargs):
    ...<7 lines>...
chunks.append(chunk)
File "./.venv/lib/python3.13/site-packages/langchain_google_genai/chat_models.py", line 1921, in _astream
if not self.async_client:
    ^^^^^^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/langchain_google_genai/chat_models.py", line 1624, in async_client
client_info=get_client_info(f"ChatGoogleGenerativeAI:{self.model}"),
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/langchain_google_genai/_common.py", line 164, in get_client_info
client_library_version, user_agent = get_user_agent(module)
~~~~~~~~~~~~~~^^^^^^^^
File "./.venv/lib/python3.13/site-packages/langchain_google_genai/_common.py", line 143, in get_user_agent
langchain_version = metadata.version("langchain-google-genai")
File "/opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/importlib/metadata/__init__.py", line 987, in version
return distribution(distribution_name).version
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/importlib_metadata/__init__.py", line 557, in version
return md_none(self.metadata)['Version']
^^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/importlib_metadata/__init__.py", line 527, in metadata
self.read_text('METADATA')
~~~~~~~~~~~~~~^^^^^^^^^^^^
File "./.venv/lib/python3.13/site-packages/importlib_metadata/__init__.py", line 998, in read_text
return self._path.joinpath(filename).read_text(encoding='utf-8')
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
File "/opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/pathlib/_local.py", line 546, in read_text
return PathBase.read_text(self, encoding, errors, newline)
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/homebrew/Cellar/python@3.13/3.13.5/Frameworks/Python.framework/Versions/3.13/lib/python3.13/pathlib/_abc.py", line 633, in read_text
return f.read()
~~~~~~^^
File "./.venv/lib/python3.13/site-packages/blockbuster/blockbuster.py", line 109, in wrapper
raise BlockingError(func_name)

blockbuster.blockbuster.BlockingError: Blocking call to io.TextIOWrapper.read

What I've Tried

1. ✗ Passing API key directly instead of loading from file - still blocks on metadata read

Related Issues

• Blocking operation inside ChatVertexAI #873 - Similar blocking operation issue with ChatVertexAI

[ati-pave]

+1, also running into this issue :(

[tpegl]

+1, also getting this now

[acuzman]

Here is a quick fix:
Fork and replace line 164 from [...genai/langchain_google_genai/_common.py‎]:

client_library_version, user_agent = get_user_agent(module)

with:

import threading
    def run_in_new_thread(func, *args, **kwargs):
        """Run a function in a new thread and return its result."""

        result = {}
        def wrapper():
            result['value'] = func(*args, **kwargs)

        thread = threading.Thread(target=wrapper)
        thread.start()
        thread.join()
        return result.get('value')

    client_library_version, user_agent = run_in_new_thread(get_user_agent, module)

[IanAtDazed]

Would be great to have a fix for this. Moving to a separate thread:

    # Wrap the agent call in asyncio.to_thread to avoid blocking I/O
    result = await asyncio.to_thread(
        agent.invoke,
        {"messages": messages}
    )

gets rid of the error, but it's unwieldy and inneficient when you want the ability to easily switch between model providers, especially in multi-agent workflows.

[smart-cau]

I've created a pull request to fix this issue: #1264

The fix caches the package version at module import time instead of reading it on every call, which eliminates the blocking I/O that was causing BlockingError in async contexts.

Changes:

• Added module-level version caching in both langchain-google-genai and langchain-google-vertexai
• Added tests to verify the caching behavior and async safety
• All existing tests pass

I confirmed the fix works by testing it locally with the same setup as the Minimal Reproducible Example. If anyone is interested in this issue, please review the PR.