usage.tool_calls undercounts when running tools in parallel

[certainly-param]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

Hey, I think there's a race condition in the usage tracking. When multiple tools run concurrently, usage.tool_calls doesn't get incremented correctly.

The increment happens in _tool_manager.py (line ~350):
python
usage.tool_calls += 1

But in _agent_graph.py, tools are launched in parallel with asyncio.create_task, and they all share the same usage object. So the increment isn't atomic and the count gets messed up.

I tested with 3 tools and sometimes get tool_calls=1, sometimes 2, rarely 3.

This affects usage limits and monitoring. Should I add a lock around the increment or is there a better approach?

Example Code

import asyncio
from pydantic_ai import Agent

agent = Agent('openai:gpt-4o')

@agent.tool
async def fetch_user(user_id: int) -> str:
    await asyncio.sleep(0.1)
    return f"User {user_id}"

@agent.tool
async def fetch_orders(user_id: int) -> str:
    await asyncio.sleep(0.1)
    return f"Orders for {user_id}"

@agent.tool
async def fetch_preferences(user_id: int) -> str:
    await asyncio.sleep(0.1)
    return f"Preferences for {user_id}"

async def main():
    result = await agent.run("Get user data, orders, and preferences for user 123")
    print(f"Expected tool_calls: 3")
    print(f"Actual tool_calls: {result.usage.tool_calls}")
    
    # Run multiple times to see the race
    for i in range(5):
        r = await agent.run("Use all three tools")
        print(f"Run {i+1}: {r.usage.tool_calls}")

asyncio.run(main())

Python, Pydantic AI & LLM client version

Python: 3.12.7
PydanticAI: 1.0.16 (or main branch commit d2a39f61)
OS: macOS (or whatever you're using)

[DouweM]

@certainly-param Good catch, can you please add a lock around the increment?