🐛 Bug Report: Requirements for Langchain example

[damianoneill]

Which component is this bug for?

Langchain Instrumentation

📜 Description

When using the default pyproject.toml generated by

langchain app new ...

Langchain instrumentation does not occur.

👟 Reproduction steps

'''
langchain app new chat
'''

Results in the following pyproject.toml

[tool.poetry]
name = "chat"
version = "0.1.0"
description = ""
authors = ["Your Name <you@example.com>"]
readme = "README.md"
packages = [
    { include = "app" },
]

[tool.poetry.dependencies]
python = "^3.11"
uvicorn = "^0.23.2"
langserve = {extras = ["server"], version = ">=0.0.30"}
pydantic = "<2"


[tool.poetry.group.dev.dependencies]
langchain-cli = ">=0.0.15"

[build-system]
requires = ["poetry-core"]
build-backend = "poetry.core.masonry.api"

👍 Expected behavior

The docs should describe what else needs to be added to show full instrumentation for a langchain example.

👎 Actual Behavior with Screenshots

This results in only the LLM being traced.

🤖 Python Version

3.11

📃 Provide any additional context for the Bug.

It looks like the following modules need to be added to get the following trace.

langchain
opentelemetry-instrumentation-fastapi

In the langchain example described here - #1043 langchain is explicitly added, and opentelemetry-instrumentation-fastapi is pulled in by chromadb.

👀 Have you spent some time to check if this bug has been raised before?

• I checked and didn't find similar issue

Are you willing to submit PR?

None

[damianoneill]

@nirga let me know if you need anything else?

Also should I see the fastapi endpoint /invoke in the trace?

Thanks,
Damian.

[nirga]

Thanks @damianoneill! Will try to reproduce this.

For the fastapi - you'll need to add the FastAPI instrumentation. Are you using our SDK?

[damianoneill]

Morning @nirga I'm not sure what you mean about the SDK, is there something other than below that I should be doing?

try:
    Traceloop.init(
        app_name="Langchain Chatbot Application",
        api_endpoint="http://localhost:4318",  # http endpoint for opentelemetry (jaeger) collector
        disable_batch=True,
    )
except Exception as e:  # pylint: disable=broad-except
    logger.error("Failed to initialize Traceloop: %s", e)

[nirga]

No, I meant that we don't instrument FastAPI currently, so you should do it yourself after initializing Traceloop. This is really easy:

import fastapi
from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
from traceloop.sdk import Traceloop

app = fastapi.FastAPI()

try:
    Traceloop.init(
        app_name="Langchain Chatbot Application",
        api_endpoint="http://localhost:4318",  # http endpoint for opentelemetry (jaeger) collector
        disable_batch=True,
    )
    FastAPIInstrumentor.instrument_app(app)
except Exception as e:  # pylint: disable=broad-except
    logger.error("Failed to initialize Traceloop: %s", e)

@app.get("/foobar")
async def foobar():
    return {"message": "hello world"}

[asaf]

Hey @nirga ,

I started a local docker instance of jaegertracing/all-in-one:1.57 to check out openllmetry locally and I get Failed to export batch code: 404, reason: 404 page not found with the configuration suggested above.

I init Traceloop as you mentioned above:

Traceloop.init(
    app_name="Langchain Chatbot Application",
    api_endpoint="http://localhost:4318",
    disable_batch=True,
)

I also tried _api_endpoint="http://localhost:4318/v1/traces", same result.
(posting directly to Jaeger works)

I wonder if there's a way to debug Traceloop as im not sure it's sending the correct http call.

Best,
Asaf.

[nirga]

Hey @asaf, I'm going to delete this comment to avoid clutter in this issue as this is not the place to report new unrelated issues. I would love to assist you - you're welcome to send a message on slack, or open a separate issue / discussion here on GitHub.

Thanks!

[asaf]

Thanks @nirga since this is not a feature / bug rather more of a question i'll convert this into a discussion - thanks.