[Bug]: Logfire log not wrapped in request span

[ionflow]

What happened?

When making a request using LiteLLM in a fastapi logfire project logs are not wrapped in the span where the request is logged.

When you collapse the span you can see it is outside of it.

I would expect the Chat Completion log to be part of the span like other libraries (instructor/openai client).

Relevant log output

No response

Twitter / LinkedIn details

No response

[ishaan-jaff]

@elisalimli any chance you can take a look at this one / help us with this ?

[krrishdholakia]

is logfire otel compatible? wondering if we can just migrate to our native otel integration - https://docs.litellm.ai/docs/proxy/logging#logging-proxy-inputoutput-in-opentelemetry-format

Because Pydantic Logfire is built on OpenTelemetry, ...

it looks like it's otel compatible, but i can't find docs on logging to it via just otel

[elisalimli]

@ishaan-jaff Seems like you are starting a new thread for success_handler. https://github.com/BerriAI/litellm/blob/main/litellm/utils.py#L3867

Logfire groups the spans using the context managers, I am suspicious if the cause of this issue is the threading.

[elisalimli]

Replacing starting a new thread with calling the success_handler directly solved the issue. It looks like the started thread doesn't have the context managers.

[elisalimli]

is logfire otel compatible? wondering if we can just migrate to our native otel integration - https://docs.litellm.ai/docs/proxy/logging#logging-proxy-inputoutput-in-opentelemetry-format

Because Pydantic Logfire is built on OpenTelemetry, ...

it looks like it's otel compatible, but i can't find docs on logging to it via just otel

Would it solve the issue of the non-nested spans?

[ionflow]

Does this help: https://docs.pydantic.dev/logfire/guides/onboarding_checklist/integrate/#opentelemetry-instrumentation

[ishaan-jaff]

Okay I can confirm our OTEL logger works with logfire, looks a lot cleaner too :)

[ishaan-jaff]

Fix here: #4254

[ishaan-jaff]

This should be fixed on 1.40.16 @ionflow

can you try this
litellm.callbacks = ["logfire"]

https://docs.litellm.ai/docs/observability/logfire_integration