Span created with Langchain not showing up in traces.

[HardKothari]

When I am using Langchain code to trace the invocation to chain/graph the generation or retrieval span is not showing up and not expanding. Its not showing tokens or model costs as well.

It used to show this before:

I dont know what went missing. I did not change anything in the code from previous working phase.

These are library versions installed in my python virtual environment.
langchain==0.2.13
langfuse==2.39.3
langchain_core==0.2.30
langchain_openai==0.1.2

Using the latest image for self hosting langfuse on docker updated 6 days ago.

This is my codeblock:

              langfuse_handler = CallbackHandler(
                  public_key=settings.LANGFUSE_PUBLIC_KEY,
                  secret_key=settings.LANGFUSE_SECRET_KEY,
                  host=settings.LANGFUSE_HOST,
                  session_id="OuterHandler-" + uuid4().__str__(),
              )
              callbacks.append(langfuse_handler)

          response = await graph.ainvoke({"data": data},
                                         config={
              "callbacks": callbacks
          }

Help on this is appreciated.

The traces are logged but information is not proper and does not show everything up.

[dosubot[bot]]

Hey there, @HardKothari! I'm here to help you with any bugs, questions, or contributions you have. Let's get to the bottom of this together!

The issue with spans not showing up or expanding in the Langchain traces, and the tokens or model costs not being displayed, could be due to the CallbackHandler not being properly instantiated or configured. Here are a few steps to ensure everything is set up correctly:

1. Ensure Proper Instantiation and Configuration:
   Make sure the CallbackHandler is correctly set up with the necessary parameters and that the flushAsync method is called to send the data.

   from langfuse.client import Langfuse
   from langfuse.callback.langchain import LangchainCallbackHandler
   from uuid import uuid4
   
   langfuse_handler = LangchainCallbackHandler(
       public_key=settings.LANGFUSE_PUBLIC_KEY,
       secret_key=settings.LANGFUSE_SECRET_KEY,
       host=settings.LANGFUSE_HOST,
       session_id="OuterHandler-" + str(uuid4()),
   )
   callbacks.append(langfuse_handler)
   
   response = await graph.ainvoke({"data": data}, config={"callbacks": callbacks})
   await langfuse_handler.flush_async()

2. Verify Environment Variables:
   Ensure that the environment variables LANGFUSE_PUBLIC_KEY, LANGFUSE_SECRET_KEY, and LANGFUSE_HOST are correctly set.

3. Check for Recent Updates:
   There have been recent updates in the Langfuse library (version 2.39.3) that could affect trace visualization or data capture. These updates include improved ingestion of traces, integration with Langchain, and the ability to add information on releases and versions to traces and observations ^[1].

4. Manual Flushing:
   Ensure that the flush_async method is called to send a batch immediately, especially for short-lived applications like serverless functions. Consider using the shutdown_async method to make sure all requests are flushed and pending requests are awaited before the process exits.

   await langfuse_handler.flush_async()
   await langfuse_handler.shutdown_async()

5. Check Docker Image:
   Since you are using the latest Docker image for self-hosting Langfuse, ensure that the image is correctly set up and there are no issues with the Docker configuration. Recent changes in the Langfuse Docker image could affect trace visualization or data capture ^[1].

6. Updated Documentation:
   Refer to the updated Langfuse documentation for new guidelines or best practices for setting up and using the CallbackHandler with Langchain. The methods getTraceId and getTraceUrl have been marked as deprecated, and there are new examples for setting up the CallbackHandler ^[2][3][4].

By following these steps, you should be able to resolve the issue with spans not showing up or expanding in the Langchain traces, and the tokens or model costs not being displayed.

To continue talking to Dosu, mention @dosu.

[marcklingen]

Which python version do you run on? (issue)

[HardKothari]

I am using 3.9

I did see the issue and it seems that may be the reason. The only option in that case for me would be to use 3.11

I used sync version before and just moved to async which explains the before and after.

Thanks a lot for sharing the information.