[Debbug Spans are not appearing in the Dev environment]

[SahbiH]

Description:
I'm working on a chatbot using Chainlit and experiencing an issue where spans are not being generated in the production environment, despite working correctly locally.
Environment Details:

Local (working): Running on a dev branch- all spans appear correctly
Production (not working): Deployed website - no spans are generated

Expected Behavior:
Spans should be generated and visible in both local and production environments.
Actual Behavior:

✅ Local development: All spans are logged and visible
❌ Production website: No spans are generated or logged

Code Context:
I'm using Langfuse for observability with the following setup:

CallbackHandler for LLM tracing
Custom spans for various operations (RAG steps, retrieval, reranking, etc.)
Context managers like lf_client.start_as_current_observation()

Questions:

Are there known differences in how Chainlit handles spans between local and production deployments?
Could this be related to environment variables or configuration that needs to be set differently for production?
Are there any specific settings or middleware that might block span generation on the hosted website?
Should I check specific logs or enable certain debugging flags to diagnose this?

Additional Context:
The spans are created using Langfuse's start_as_current_observation() context managers throughout the codebase, and all Langfuse authentication checks pass successfully (lf_client.auth_check() returns True).

[inkeep-ai-assistant[bot]]

Based on the available documentation, here are several potential causes and debugging steps for missing spans in your production environment:

Environment Configuration

First, verify your Langfuse environment variables are correctly set in production(1):

LANGFUSE_SECRET_KEY = "sk-lf-..."
LANGFUSE_PUBLIC_KEY = "pk-lf-..."
LANGFUSE_BASE_URL = "https://cloud.langfuse.com" # 🇪🇺 EU region
# LANGFUSE_BASE_URL = "https://us.cloud.langfuse.com" # 🇺🇸 US region

(1)

Debug Logging

Enable debug logging to see what's happening with span creation(1):

import logging

langfuse_logger = logging.getLogger("langfuse")
langfuse_logger.setLevel(logging.DEBUG)

(1)

Alternatively, set debug=True when initializing the client or use the environment variable LANGFUSE_DEBUG="True"(1).

Data Flushing

In short-lived applications or serverless environments, spans might not be sent before the process terminates. Ensure you're calling flush() before your application exits(2):

from langfuse import get_client

langfuse = get_client()
# ... create traces and observations ...
langfuse.flush() # Ensures all pending data is sent

(2)

For proper cleanup, call shutdown() before exiting(2):

langfuse.shutdown()

(2)

Context Propagation Issues

If you're using ThreadPoolExecutors, ensure OpenTelemetry context is properly propagated(1):

from opentelemetry.instrumentation.threading import ThreadingInstrumentor

ThreadingInstrumentor().instrument()

(1)

Sampling Configuration

Check if sampling is accidentally enabled in production. If sample_rate is set too low, spans won't be sent(1):

from langfuse import Langfuse

# Sample approximately 20% of traces
langfuse_sampled = Langfuse(sample_rate=0.2)

(1)

Ensure your production environment doesn't have LANGFUSE_SAMPLE_RATE set to a low value(1).

Recommended Debugging Steps

1. Enable debug logging in production to see SDK activity(1)
2. Verify environment variables are correctly set(1)
3. Add explicit flush() calls after span creation, especially if using serverless/short-lived processes(2)
4. Check for sampling configuration that might be filtering out spans(1)
5. Verify auth_check() returns True in production (you mentioned this passes)(3)

The most common cause for missing spans in production environments is data not being flushed before the process terminates(2). Start by adding explicit flush() or shutdown() calls and enabling debug logging to see what's happening.

📚 Sources:

1. Advanced Usage
2. Instrumentation
3. LangChain & LangGraph Integration

---

Have another question? Just tag @inkeep.

[jannikmaierhoefer]

Hi @SahbiH, was inkeep able to help you here?

[raimondkempees-arc]

We have the exact same issue. Locally, with Langfuse debug logging on, spans are processed and sent to Langfuse. With the exact same configuration it does not show processed spans on an Azure AppService, even though we do see Langfuse doing something in the Azure logs:

2026-01-27T21:17:06.682Z [Langfuse SDK] [DEBUG] Chat model start with ID: 13c30c61-a347-**
2026-01-27T21:17:06.682Z [Langfuse SDK] [DEBUG] Generation start with ID: 13c30c61-a347-** and parentRunId a7896ce1-71**
2026-01-27T21:17:08.599Z [Langfuse SDK] [DEBUG] LLM end with ID: 13c30c**
2026-01-27T21:17:08.603Z [Langfuse SDK] [DEBUG] Chain end with ID: a7896ce1

Is there anything under the hood we need to look at? We've done all steps as described above. We're using NestJS as platform.

[raimondkempees-arc]

OK. After about 36 hours of pulling my hair out, on Azure this works:

            // Create the Langfuse span processor
            const spanProcessor = new LangfuseSpanProcessor({
                publicKey,
                secretKey,
                baseUrl,
                exportMode: "immediate",
            });

     // Create NodeTracerProvider with the span processor
            this.tracerProvider = new NodeTracerProvider({
                spanProcessors: [spanProcessor],
            });

            // Register globally AND set as Langfuse tracer provider
            this.tracerProvider.register();
            this.logger.log('NodeTracerProvider registered globally');

            // Set this provider directly as the Langfuse tracer provider
            setLangfuseTracerProvider(this.tracerProvider);

[jannikmaierhoefer]

Hi @raimondkempees-arc, thank you very much for posting your solution here!