Langfuse loses trace context when OpenTelemetry middleware is enabled

[stefanshipinkoski]

Hey Langfuse team 👋,

I’m using Langfuse together with Honeycomb in a FastAPI app, both relying on OpenTelemetry OTLP exports for tracing.

Honeycomb setup: uses TracerProvider + OTLPSpanExporter and adds OpenTelemetryMiddleware via

app.add_middleware(OpenTelemetryMiddleware)

which works fine alone.

Langfuse setup: creates its own isolated TracerProvider and initializes the client as:

Langfuse(tracer_provider=TracerProvider(), ...)

and the @observe decorator to track LLM flows and functions.

• Python: 3.13
• langfuse-sdk: 3.0.1
• opentelemetry-sdk: 1.37.0

Problem:
When OpenTelemetryMiddleware is enabled, all Langfuse traces lose context — the dashboard shows disconnected or incomplete traces. Disabling the middleware restores proper Langfuse behavior.

Question:
What’s the recommended way to integrate Langfuse with an existing OpenTelemetry setup (Honeycomb) that already defines a global TracerProvider and exports via OTLP?

I tried searching for similar issues, but they seem not to be related to what I experience. If I'm wrong, feel free to share it with me 🙏

Additionally, if needed, I can share screenshots showing the difference between Langfuse traces with and without the OpenTelemetry middleware enabled.

Thank you,
BR,
Stefan

[jannikmaierhoefer]

Hi @stefanshipinkoski, could you try to set up a separate OTel tracer for Langfuse, similar to the setup we documented here with Sentry?

https://langfuse.com/docs/observability/sdk/typescript/advanced-usage#setup-with-sentry

[stefanshipinkoski]

No problem!
No, at the moment I don't the use the blocked_instrumentation_scopes option.
This is my current Lnagfuse client initialization:

_langfuse_client = Langfuse(
                public_key=settings.langfuse_public_key,
                secret_key=settings.langfuse_secret_key,
                host=settings.langfuse_host,
                environment=settings.environment,
                tracer_provider=langfuse_provider,
)

Let me know if more info is needed!

[hassiebp]

Hi @stefanshipinkoski - could you please share trace links for the broken Langfuse Traces?

This is most likely due to the HoneyComb middleware interfering with the OTEL context in unexpected ways

[stefanshipinkoski]

Hey @hassiebp, hmm yeah makes sense.
Unfortunately I can't share those exact traces, but here is a sample trace where the issue persists link
Please let me know if there is a way, to run both service without any middleware interfering? 🙏

[stefanshipinkoski]

Hey @hassiebp, circling back to this
can you advise on any next steps, or what can I do to resolve conflicts between HoneyComb and Langfuse?
Any suggestion would be really appreciated 🙏

[hassiebp]

@stefanshipinkoski - this is most likely on HoneyComb not playing nicely with the OpenTelemetry context. We have not yet looked closer into this issue