UpdateRoot in the callback handler seems not work

[joaoCrulhas]

Hi.
The reason that I'm asking this question is because I'm facing an issue during the development.

I'm using the stack (Langchain, langfuse, and as model OpenAI).
So what I'm trying to do is, after invoking my chain, be able to bundle all process in a root trace, if some specific information (traceName, traceMetadata and user), to be more accessible to run the evaluations and other things.

Here is my code:

 const lf = new Langfuse();

    const trace = lf.trace({
      metadata: {
        type: 'DocumentType',
      },
      name: 'Task_One',
      userId: 'userId',
    });
const langfuseHandlerTrace = new CallbackHandler({ root: trace, updateRoot: true });
    const prompt = await this.langfuse.getPrompt('prompt1');
    const parsers = StructuredOutputParser.fromZodSchema(
      z.object({
         // my Parser
      }),
    );
    const promptTemplate = await PromptTemplate.fromTemplate(prompt.getLangchainPrompt()).partial({
      schema: parsers.getFormatInstructions(),
    });


    const extractJsonLambda = new RunnableLambda({
      func: extractJson<ConsumptionRecord>,
    });
    const postProcessing = new RunnableLambda({
      func: postProcessingConsumptionInvoice,
    });

    const chain = RunnableSequence.from(
      [promptTemplate, model, extractJsonLambda, postProcessing],
      'TaskOne',
    );

await chain.invoke(
      {
        query: UserInput,
      },
      {
        callbacks: [langfuseHandlerTrace],
      },
    );

The prompt is working fine, and the result is the expected result.
Info as userId, metadata is being stored in the trace, but I'd like to update the name because, currently what is being shown is
RunnableLambda, but I'd like to show the name to make it easy to filter the traces by name.

Thank you in advance.

[dosubot[bot]]

@joaoCrulhas Olá! Sou um bot que pode te ajudar a resolver bugs, responder perguntas e se tornar um colaborador. Estou aqui para te ajudar enquanto aguardamos um mantenedor humano.

The issue you're facing with the updateRoot option in the CallbackHandler class is a known limitation. The updateRoot option is not concurrency-safe and is marked for deprecation in future versions of the Langfuse JavaScript SDK ^[1].

To achieve your goal of updating the trace name to make it more descriptive and easier to filter, you can manually update the trace after its creation. Here is an example of how you can do this:

const lf = new Langfuse();

const trace = lf.trace({
  metadata: {
    type: 'DocumentType',
  },
  name: 'Task_One',
  userId: 'userId',
});

const langfuseHandlerTrace = new CallbackHandler({ root: trace, updateRoot: false });

const prompt = await this.langfuse.getPrompt('prompt1');
const parsers = StructuredOutputParser.fromZodSchema(
  z.object({
    // my Parser
  }),
);
const promptTemplate = await PromptTemplate.fromTemplate(prompt.getLangchainPrompt()).partial({
  schema: parsers.getFormatInstructions(),
});

const extractJsonLambda = new RunnableLambda({
  func: extractJson<ConsumptionRecord>,
});
const postProcessing = new RunnableLambda({
  func: postProcessingConsumptionInvoice,
});

const chain = RunnableSequence.from(
  [promptTemplate, model, extractJsonLambda, postProcessing],
  'TaskOne',
);

await chain.invoke(
  {
    query: UserInput,
  },
  {
    callbacks: [langfuseHandlerTrace],
  },
);

// Update the trace name after invocation
trace.update({
  name: 'Updated_Task_One',
});

await lf.flushAsync();

In this example, the updateRoot option is set to false, and the trace name is updated manually after the chain invocation. This ensures that the trace name is set to a more descriptive value, making it easier to filter and manage in the Langfuse UI ^[2][3].

To continue talking to Dosu, mention @dosu.

[joaoCrulhas]

Just one notice, doing this, the input and output will be null, to fix it should add

trace.update({
  input: input,
  output: output,
  name: 'Updated_Task_One',
});

[marcklingen]

glad that this works for you!

// Update the trace name after invocation
trace.update({
  name: 'Updated_Task_One',
});

can you check why the input/output will be null in this case? This should work as expected. You could enable debug logs in Langfuse SDK to see the raw events created.