'Span' object has no attribute '_context_manager_state'

[Afstkla]

How do you use Sentry?

Sentry Saas (sentry.io)

Version

2.35.0

Steps to Reproduce

Run

import asyncio

import sentry_sdk
from agents import Agent, Runner
from sentry_sdk.integrations.openai_agents import OpenAIAgentsIntegration

sentry_sdk.init(
    dsn="...",
    send_default_pii=True,
    traces_sample_rate=1.0,
    integrations=[
        OpenAIAgentsIntegration(),
    ],
)


main_agent = Agent(
    name="main_agent",
    model="gpt-5",
)


async def run_agent() -> None:
    runner = await Runner.run(
        starting_agent=main_agent,
        input="How are you?",
    )
    print(runner.final_output)


async def main() -> None:
    await asyncio.gather(*[run_agent() for _ in range(2)])


if __name__ == "__main__":
    asyncio.run(main())

Then, the agent that finishes 2nd will fail with:

{
  "error": "'Span' object has no attribute '_context_manager_state'",
  "tool_name": "<agentname>_agent"
}

Expected Result

1. This error not to happen
2. That most (if not all) parts of the Sentry SDK are wrapped in try catch blocks, so that if for some reason like now a Span cannot be closed, it doesn't interfere with the normal flow

Actual Result

The agent crashes, and instead of the result that was generated by the nested agent being available to the main agent, the main agent only sees the error message and cannot continue.

[linear]

PY-1815 'Span' object has no attribute '_context_manager_state'

[Afstkla]

(probably duplicate of #4698, sorry, just found that one)

[sentrivana]

Hey @Afstkla, thanks for reporting.

This does look similar to what I wanted to patch with #4698, but it's not the same since in your case, it's the Span complaining about the missing attribute, and in the PR it was the profiler.

All in all though it looks like there's a more fundamental issue with tracing (and profiling, by extension) in the OpenAIAgentsIntegration. But at the very least we should, of course, not error out like this. I'll see if I can find the issue; otherwise I'll at least patch our code so that we don't throw an exception.

An example to reproduce (even if not consistently) would be super helpful -- in my tests I haven't been able to trigger this so far.