Failed to export spans. Server responded with HTTP status code 403.

[kaikai-hangzhou]

Describe your question

Hi,our project integrated Langfuse according to the docs(using the endpoint https://us.cloud.langfuse.com/api/public/otel` with Basic Auth authentication). The integration previously worked as expected, and trace data was successfully ingested.
Howere, starting at an unknown point in time, the following error began to appear:
[OkHttp https://us.cloud.langfuse.com/...] i.o.e.internal.http.HttpExporter : Failed to export spans. Server responded with HTTP status code 403. Error message: Unable to parse response body, HTTP status message:
We have verified that no langfuse-related configuration changes were made during this period. Additionally, when accessing https://us.cloud.langfuse.com/api/public/otel directly, the endpoint currently returns an HTTP 404 status.
We are trying to understand what might have caused this behavior change, and what steps should be taken to diagnose and resolve the issue.
Any insights or suggestions would be appreciated. Thanks.

Langfuse Cloud or Self-Hosted?

Langfuse Cloud

If Self-Hosted

No response

If Langfuse Cloud

yes

SDK and integration versions

No response

Pre-Submission Checklist

• I have checked for existing issues/discussions and consulted Langfuse AI.

[inkeep-ai-assistant[bot]]

The HTTP 403 error you're experiencing indicates an authentication issue with your OTLP endpoint configuration. Based on the documentation, here are the key points to verify:

Correct Endpoint Configuration

For Langfuse Cloud US region, set your OTLP endpoint to(1):

OTEL_EXPORTER_OTLP_ENDPOINT="https://us.cloud.langfuse.com/api/public/otel"

(1)

The endpoint should not include /v1/traces in the OTEL_EXPORTER_OTLP_ENDPOINT variable, as the exporter appends this automatically(1). If you need signal-specific configuration, use(1):

OTEL_EXPORTER_OTLP_TRACES_ENDPOINT="https://us.cloud.langfuse.com/api/public/otel/v1/traces"

(1)

Authentication Setup

Langfuse uses Basic Auth for the OTLP endpoint(1). You must base64-encode your public and secret keys joined by a colon(2)(3):

import base64
auth = base64.b64encode(b"pk-...:sk-...").decode()

(2)

Then set the header(2)(3):

OTEL_EXPORTER_OTLP_HEADERS="Authorization=Basic ${AUTH_STRING}"

(2)(3)

Important Notes

• Langfuse does not support gRPC for the OpenTelemetry endpoint; use HTTP/protobuf instead(1)
• Set all OTLP-related environment variables before importing Langfuse or OpenTelemetry libraries, as they are read only once at import/init time(2)(3)
• The 404 response when accessing the endpoint directly is expected behavior; the endpoint requires proper authentication and payload(4)

Troubleshooting Steps

1. Verify your API keys are correct and haven't expired
2. Ensure the base64 encoding of your credentials is formatted correctly
3. Check that environment variables are set before library initialization
4. Confirm you're using http/protobuf protocol, not gRPC(1)

📚 Sources:

1. LLM Observability via OpenTelemetry
2. Getting Failed to export span batch code: 404, reason for a self hosted Langfuse
3. Getting Failed to export span batch code: 404, reason for a self hosted Langfuse
4. Traces aren't seen in the web ui, OTEL error: Failed to export batch code: 404, Not Found - /api/public/otel/v1/traces

---

Have another question? Just tag @inkeep.

[jannikmaierhoefer]

Hi @kaikai-hangzhou, was inkeep able to help you here?

[kaikai-hangzhou]

Thank you for the response. To clarify, the integration was working correctly before, and trace data had been successfully ingested. The 403 error started appearing at some point without a clearly identifiable trigger.
Based on the troubleshooting steps, we have already verified the following:
1.We have verified that our API keys are correct and have not expired
2.have ensured that the Base64 encoding of the Basic Auth credentials is formatted correctly
3.the exporter is using the HTTP/protobuf protocol, not gRPC
4.We have checked that all OTLP-related environment variables are set before the OpenTelemetry / Langfuse libraries are initialized. Additionally, we also tried configuring the same values via JVM system properties, but the issue still persists Despite completing all of the above checks, the exporter continues to fail with an HTTP 403 error.
Any insights or suggestions would be appreciated. Thanks again.

[jannikmaierhoefer]

Thanks @kaikai-hangzhou, how are the Otel events generated? Is it an external library that does this in your case?

[kaikai-hangzhou]

Thanks, in our project, the OTel events are generated automatically through Spring Boot 3's Observation API and Micrometer Tracing. We are not manually creating Spans; instead, we rely on the native instrumentation provided by the Spring ecosystem.Specifically, here is our technical stack:
Event Generation: We use spring-ai, which has built-in support for Micrometer Observations. It automatically records events and metrics for LLM calls.
Custom Metadata: We use a custom ObservationFilter to enrich the spans with high-cardinality data (like specific AI prompt/completion content) before they are converted to OTel Spans.
Bridge & Export:
We use micrometer-tracing-bridge-otel to bridge the Micrometer Observations into standard OpenTelemetry Spans.
We use the standard opentelemetry-exporter-otlp (Java SDK) to export these spans to Langfuse via the HTTP/protobuf protocol.