Langfuse Connector Failing to Trace Pipeline Data

[LTOpen]

Describe the bug
I attempted to use Langfuse for tracing my pipeline, but it appears that nothing is being traced. Initially, I speculated that the issue might lie with the model. Although I ran the example code provided in the documentation and observed some functionality, it didn't meet expectations. Specifically, the pipeline fails to trace input and output values.

To diagnose the issue, I modified my pipelines multiple times. When I explicitly declare the input and output values, Langfuse traces only the final output value and still fails to capture intermediate values, particularly those related to the LLM model’s input and output.

After numerous tests, I'm still unable to determine the root of the problem. Attached is a screenshot for reference.

My Code

from haystack import Pipeline
from haystack.dataclasses import ChatMessage
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.components.builders.chat_prompt_builder import ChatPromptBuilder
from haystack.components.converters import OutputAdapter
from haystack_integrations.components.connectors.langfuse import LangfuseConnector
import os

os.environ["LANGFUSE_HOST"] = "https://cloud.langfuse.com"
os.environ["HAYSTACK_CONTENT_TRACING_ENABLED"] = "true"

tracer = LangfuseConnector('haystack-test')

messages = [
    ChatMessage.from_system("""You are a data generator, and you will generate data according to the given format.
---
User: 
Language: English 
Number of words: 1 
Part of speech: Noun 

Assistant: 
Result: Apple
---"""
),
    ChatMessage.from_user("""User:
Language: {{lang}}
Number of words: {{num}}
Part of speech: {{pos}}""")
]

prompt_builder = ChatPromptBuilder(template=messages)

llm = OpenAIChatGenerator(
    model='gpt-3.5-turbo'
)
adapter = OutputAdapter(
    template="""{{replies[0].content}}""",output_type=str
)

gen_pipe = Pipeline()

gen_pipe.add_component("tracer", tracer)
gen_pipe.add_component('prompt',prompt_builder)
gen_pipe.add_component('llm',llm)
gen_pipe.add_component('adapter',adapter)

gen_pipe.connect('prompt.prompt','llm.messages')
gen_pipe.connect('llm.replies','adapter')

response = gen_pipe.run(data={
    'prompt':{
        'template_variables': {
            'lang':'English',
            'num':1,
            'pos':'verb'
        }
    }
})
print(response)
print(response["tracer"]["trace_url"])

Picuture
example code in the documentation works:

tracing result in my code:

System:

• OS: Win10
• GPU/CPU: Intel/Nvidea
• Haystack version (commit or version number): 2.2.0
• Python version: 3.10.13

[Redna]

One thing you can try, is to set the environment variable before your actual import of LangfuseConnector. That's the docstring of the LangFuse tracer:

    import os

    os.environ["HAYSTACK_CONTENT_TRACING_ENABLED"] = "true"

    from haystack import Pipeline
    from haystack.components.builders import DynamicChatPromptBuilder
    from haystack.components.generators.chat import OpenAIChatGenerator
    from haystack.dataclasses import ChatMessage
    from haystack_integrations.components.connectors.langfuse import LangfuseConnector

#...

I am setting the variable in my run configuration and content tracing works most of the time fine.

[anakin87]

@Redna' suggestion seems to point in the right direction.

@LTOpen I am closing the issue. Feel free to reopen it if the problem persists.