[Bug] Improper closing for task delegation when use streaming mode - OpenTelemetry affected

[Harry]

Description

There are two issues here:
First is with Agno Task Delegation: break Iterator and AsyncIterator without closing proper
Second is with openinference-instrumentation-agno. I realised that even after I fixed the first issue, the Agent in the team execution is the same level and the delegate_task_to_member

Steps to Reproduce

1. Enable Langfuse tracing opentelemetry
2. Run a team with task delegation
3. Checking terminal logs and Langfuse tracing logs.

Agent Configuration (if applicable)

import os
import asyncio
import base64

from uuid import uuid4

from dotenv import load_dotenv


from agno.agent import RunEvent
from agno.agent.agent import Agent


from agno.team import Team, TeamRunEvent

from agno.tools.hackernews import HackerNewsTools
from agno.tools.duckduckgo import DuckDuckGoTools


from agno.models.azure.openai_chat import AzureOpenAI

# Agno Langfuse setup
from openinference.instrumentation.agno import AgnoInstrumentor

from opentelemetry import trace as trace_api
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter


langfuse_public_key = ""
langfuse_secret_key = ""

LANGFUSE_AUTH = base64.b64encode(
    f"{langfuse_public_key}:{langfuse_secret_key}".encode()
).decode()
os.environ["OTEL_EXPORTER_OTLP_ENDPOINT"] = (
    "https://us.cloud.langfuse.com/api/public/otel"
)
os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"Authorization=Basic {LANGFUSE_AUTH}"
# Configure the tracer provider
tracer_provider = TracerProvider()
tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter()))
trace_api.set_tracer_provider(tracer_provider=tracer_provider)

# Start instrumenting agno
AgnoInstrumentor().instrument()

load_dotenv()


wikipedia_agent = Agent(
    id="hacker-news-agent",
    name="Hacker News Agent",
    role="Search Hacker News for information",
    model=AzureOpenAI(id="gpt-4o-2024-11-20"),
    tools=[HackerNewsTools()],
    instructions=[
        "Find articles about the company in the Hacker News",
    ],
)

website_agent = Agent(
    id="website-agent",
    name="Website Agent",
    role="Search the website for information",
    model=AzureOpenAI(id="gpt-4o-2024-11-20"),
    tools=[DuckDuckGoTools()],
    instructions=[
        "Search the website for information",
    ],
)

user_id = str(uuid4())
id = str(uuid4())

company_info_team = Team(
    name="Company Info Team",
    id=id,
    user_id=user_id,
    model=AzureOpenAI(id="gpt-4o-2024-11-20"),
    members=[
        wikipedia_agent,
        website_agent,
    ],
    markdown=True,
    instructions=[
        "You are a team that finds information about a company.",
        "First search the web and wikipedia for information about the company.",
        "If you can find the company's website URL, then scrape the homepage and the about page.",
    ],
    show_members_responses=True,
)


async def run_team_with_events(prompt: str):
    content_started = False
    async for run_output_event in company_info_team.arun(
        prompt,
        stream=True,
        stream_events=True,
    ):
        if run_output_event.event in [
            TeamRunEvent.run_started,
            TeamRunEvent.run_completed,
        ]:
            print(f"\nTEAM EVENT: {run_output_event.event}")

        if run_output_event.event in [TeamRunEvent.tool_call_started]:
            print(f"\nTEAM EVENT: {run_output_event.event}")
            print(f"TOOL CALL: {run_output_event.tool.tool_name}")
            print(f"TOOL CALL ARGS: {run_output_event.tool.tool_args}")

        if run_output_event.event in [TeamRunEvent.tool_call_completed]:
            print(f"\nTEAM EVENT: {run_output_event.event}")
            print(f"TOOL CALL: {run_output_event.tool.tool_name}")
            print(f"TOOL CALL RESULT: {run_output_event.tool.result}")

        # Member events
        if run_output_event.event in [RunEvent.tool_call_started]:
            print(f"\nMEMBER EVENT: {run_output_event.event}")
            print(f"AGENT ID: {run_output_event.agent_id}")
            print(f"TOOL CALL: {run_output_event.tool.tool_name}")
            print(f"TOOL CALL ARGS: {run_output_event.tool.tool_args}")

        if run_output_event.event in [RunEvent.tool_call_completed]:
            print(f"\nMEMBER EVENT: {run_output_event.event}")
            print(f"AGENT ID: {run_output_event.agent_id}")
            print(f"TOOL CALL: {run_output_event.tool.tool_name}")
            print(f"TOOL CALL RESULT: {run_output_event.tool.result}")

        if run_output_event.event in [TeamRunEvent.run_content]:
            if not content_started:
                print("CONTENT")
                content_started = True
            else:
                print(run_output_event.content, end="")


if __name__ == "__main__":
    asyncio.run(
        run_team_with_events(
            "Write me a full report on everything you can find about Agno, the company building AI agent infrastructure.",
        )
    )

Expected Behavior

No errors should be displayed in terminal and response of task delegation to the agent should be captured.

Actual Behavior

Errors were displayed in terminal and response of task delegation to the agent was not be captured.

Screenshots or Logs (if applicable)

ERROR:opentelemetry.context:Failed to detach context
Traceback (most recent call last):
  File "/Users/harry.tran/Library/Caches/pypoetry/virtualenvs/apac-gaia-reg-openai-wrapper-service-3kiMfCOe-py3.10/lib/python3.10/site-packages/opentelemetry/trace/__init__.py", line 589, in use_span
    yield span
  File "/Users/harry.tran/Library/Caches/pypoetry/virtualenvs/apac-gaia-reg-openai-wrapper-service-3kiMfCOe-py3.10/lib/python3.10/site-packages/openinference/instrumentation/_tracers.py", line 141, in start_as_current_span
    yield cast(OpenInferenceSpan, current_span)
  File "/Users/harry.tran/Library/Caches/pypoetry/virtualenvs/apac-gaia-reg-openai-wrapper-service-3kiMfCOe-py3.10/lib/python3.10/site-packages/openinference/instrumentation/agno/_wrappers.py", line 499, in arun_stream
    yield response
GeneratorExit

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/Users/harry.tran/Library/Caches/pypoetry/virtualenvs/apac-gaia-reg-openai-wrapper-service-3kiMfCOe-py3.10/lib/python3.10/site-packages/opentelemetry/context/__init__.py", line 155, in detach
    _RUNTIME_CONTEXT.detach(token)
  File "/Users/harry.tran/Library/Caches/pypoetry/virtualenvs/apac-gaia-reg-openai-wrapper-service-3kiMfCOe-py3.10/lib/python3.10/site-packages/opentelemetry/context/contextvars_context.py", line 53, in detach
    self._current_context.reset(token)
ValueError: <Token var=<ContextVar name='current_context' default={} at 0x105d3d030> at 0x12e8fda80> was created in a different Context

Environment

- agno: 2.2.8
- openinference-instrumentation-agno: 0.1.22

Possible Solutions (optional)

agno
In delegate_task_to_member function in team.py We should not break when consuming the Iterator and AsyncIterator

for member_agent_run_output_event in member_agent_run_response_stream:
  # If we get the final response, we can break out of the loop
  if isinstance(member_agent_run_output_event, TeamRunOutput) or isinstance(
      member_agent_run_output_event, RunOutput
  ):
      member_agent_run_response = member_agent_run_output_event  # type: ignore
      break # To be removed
  
  # Check if the run is cancelled
  check_if_run_cancelled(member_agent_run_output_event)
  
  # Yield the member event directly
  member_agent_run_output_event.parent_run_id = (
      member_agent_run_output_event.parent_run_id or run_response.run_id
  )
  yield member_agent_run_output_event

Before removing break

After removing break

openinference-instrumentation-agno
I have no solution for this

Additional Context

No response

[Harry]

@dirkbrnd I noticed you are working on the openinference-instrumentation-agno. Could you help to confirm this behaviour is expected that the Agent in the team execution is the same level and the delegate_task_to_member