Do another update call when we're updating an ended trace in update_current_trace

[yorickvP]

It's possible to have observations that 'outlive' their trace, for example when returning a StreamingResponse in FastAPI.

Currently, calling update_current_trace in such cases only worked for the server_merged_attributes. We can work around this by sending the metadata in .trace() instead of storing it in the context.

Failing use case:

from fastapi import router, Response, StreamingResponse

@router.post("/test")
@observe()
async def test() -> Response:
    response_gen = test_inner()
    return StreamingResponse(response_gen, media_type="text/event-stream")

@observe()
async def test_inner() -> AsyncGenerator[str, None]:
    langfuse_context.update_current_trace(
        user_id="unauth"
    )
    yield "test"

---

Important

Fix update_current_trace to handle updates for ended traces by checking trace_already_ended and sending appropriate update events.

• Behavior:
  • In update_current_trace, handle cases where traces end before observations by checking trace_already_ended.
  • If trace_already_ended, send an update event for all attributes, not just server_merged_attributes.
  • Avoid updating _observation_params_context if trace_already_ended.

^{This description was created by for a8f972e. It will automatically update as commits are pushed.}

Greptile Summary

Disclaimer: Experimental PR review

Modified the update_current_trace function to handle cases where observations outlive their traces, particularly in FastAPI StreamingResponse scenarios.

• Added check in langfuse/decorators/langfuse_decorator.py to detect if trace has already ended by checking observation params context
• Modified behavior to send all parameters directly to server instead of storing in context when trace has ended
• Fixed issue with metadata and tags not being properly updated for ended traces
• Added test coverage in tests/test_decorators.py to verify async streaming scenarios where observations continue after trace completion

_{💡 (1/5) You can manually trigger the bot by mentioning @greptileai in a comment!}

[greptile-apps]

LGTM

_{1 file(s) reviewed, no comment(s)
Edit PR Review Bot Settings | Greptile}

[greptile-apps]

LGTM

_{1 file(s) reviewed, no comment(s)
Edit PR Review Bot Settings | Greptile}

[CLAassistant]

All committers have signed the CLA.