bug: Langchain Python Dataset run has null trace I/O

[nimbusaeta]

Describe the bug

In the VUI, in the Dataset run view, the inputs of each view appear as null, but in the Dataset Item view, the input is populated with the expected text.

To reproduce

The run was created following the Datasets Cookbook (the Langchain application part).

Additional information

No response

[marcklingen]

Thanks for creating this issue. We'll try to push a fix soon.

More context:

• item.get_langchain_handler() creates a trace under the hood
• if the handler is then passed to a langchain agent/chain, it does not update the trace (this is generally the desired behavior as multiple runs could be grouped into the same trace).

[pckim93]

Is this still an issue? I'm running into this as well (with some paths adjusted for privacy):

.../eval.py 
Traceback (most recent call last):
  File ".../eval.py", line 191, in <module>
    main()
  File ".../eval.py", line 100, in main
    handler = item.get_langchain_handler(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.11/site-packages/langfuse/client.py", line 2553, in get_langchain_handler
    self.link(
  File ".venv/lib/python3.11/site-packages/langfuse/client.py", line 2518, in link
    self.langfuse.client.dataset_run_items.create(
  File ".venv/lib/python3.11/site-packages/langfuse/api/resources/dataset_run_items/client.py", line 127, in create
    raise NotFoundError(pydantic_v1.parse_obj_as(typing.Any, _response.json()))  # type: ignore
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
langfuse.api.resources.commons.errors.not_found_error.NotFoundError: status_code: 404, body: {'message': 'Trace not found'}