Litellm integration is broken

[enniofilicicchia]

It does not even enter the function
Opentelemetry._handle_sucess in litellm.integrations.opentelemetry.py

litellm.callbacks = ["otel"]
os.environ["LMNR_PROJECT_API_KEY"] = (
    "yyy"
)
os.environ["OTEL_EXPORTER"] = "otlp_grpc"
os.environ["OTEL_ENDPOINT"] = "http://mydomain.com:8000"
os.environ["OTEL_HEADERS"] = (
    f"authorization=Bearer {os.environ["LMNR_PROJECT_API_KEY"]}"
)
os.environ["DEBUG_OTEL"] = "true"

I followed the debug run and no logging method of Opentelemetry ever enters.

1. LiteLLM is very popular, it would be cool for laminar to have a working reliable integration!
2. In the meantime I would be happy with manually sending the POST request on the traces endpoint, but I cannot find anything about /api/v1/traces in the docs or the OpenAPI specs.
3. As a side note it's unclear if the endpoint should include "/api", "/api/v1" or "/api/v1/traces", but this is not the main problem.

[dinmukhamedm]

Hey, thanks for opening the issue! I think this is to do with how litellm updated their dependency management. Can you try adding the proxy extra of litellm, e.g.

pip install `litellm[proxy]`
# or uv add litellm --extra proxy
# or poetry add litellm --extra proxy

Also be sure to have the following packages installed: opentelemetry-api opentelemetry-sdk opentelemetry-exporter-otlp

Please let us know, if this resolves your issue. I will update the docs both on our side and LiteLLM side

---

To your questions number 2 and 3, it must be /v1/traces, but it accepts the protobuf encoded payload (see more in our docs), so I don't think an API spec will help much. In addition, HTTP connection is less reliable, and so we advice against it, so the decision to not document that bit is half-concious

[dinmukhamedm]

Also, for reference: BerriAI/litellm#8169