DOC: <Issue related to /observability/how_to_guides/tracing/trace_with_opentelemetry>

[mindful-time]

Issue: Unauthorized Access When Integrating Traceloop SDK with LangSmith with LLama-index

Description

When attempting to integrate Traceloop SDK with LangSmith for tracing LlamaIndex operations, receiving consistent 401 Unauthorized errors. The trace export attempts are failing with the following error:

ERROR:
opentelemetry.exporter.otlp.proto.http.trace_exporter: Failed to export batch 
code: 401, 
reason: {"error":"Unauthorized"}

Current Setup

• Using Traceloop SDK to send traces to LangSmith's OpenTelemetry endpoint
• Environment configured with Azure OpenAI for LlamaIndex operations
• Traceloop initialization includes:
  • API endpoint: https://api.smith.langchain.com/otel
  • Headers with LangSmith API key and content type
  • Batch disable flag set to true

Steps to Reproduce

1. Set up environment with LangSmith API key
2. Initialize Traceloop SDK with LangSmith endpoint
3. Run any LlamaIndex operation (in this case, a simple document query)
4. Observe trace export failures in logs

Expected Behavior

• Successful authentication with LangSmith
• Proper export of OpenTelemetry traces to LangSmith

Actual Behavior

• 401 Unauthorized errors when attempting to export traces
• No traces being recorded in LangSmith

Potential Investigation Done

1. Verifed LangSmith API key validity and permissions
2. Confirmed correct environment variable name (LANGCHAIN_API_KEY)
3. Validate LangSmith endpoint URL and authentication requirements
4. Check Traceloop SDK configuration for LangSmith compatibility

Related Documentation

• OpenTelemetry Tracing Guide for Langsmith
• OpenTelemetry Tracing Guide for Llama Index
• OpenTelemetry Tracing Guide for Traceloop

full implementation

# Initialize Traceloop SDK and LlamaIndex with Azure OpenAI
# This script sets up tracing and testing of the LlamaIndex integration with Azure OpenAI

import os
from traceloop.sdk import Traceloop

# Get LangSmith API key from environment variables for tracing
LANGSMITH_API_KEY = os.getenv("LANGCHAIN_API_KEY")

# Initialize Traceloop with LangSmith endpoint and authentication
# - api_endpoint: LangSmith OTEL endpoint for trace collection
# - headers: Authentication and content type headers
# - disable_batch: Send traces immediately without batching
# - app_name: Name of the application for trace identification
Traceloop.init(api_endpoint="https://api.smith.langchain.com/otel",
               headers=
               {
                "x-api-key": LANGSMITH_API_KEY, 
                "content-type": "application/protobuf"},
               disable_batch=True,
               app_name="test"
               )

# Import required LlamaIndex components
from llama_index.core import VectorStoreIndex, Document
from llama_index.llms.azure_openai import AzureOpenAI
from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
from llama_index.core import Settings
from app.core.config import get_settings

# Load application settings
settings = get_settings()

# Define required Azure OpenAI settings to validate configuration
required_settings = [
        ("Azure OpenAI API Key", settings.azure_openai_api_key),
        ("Azure OpenAI Endpoint", settings.azure_openai_endpoint),
        ("Azure OpenAI Deployment Name", settings.azure_openai_deployment_name),
        ("Azure OpenAI API Version", settings.azure_openai_api_version),
        ("Azure OpenAI Embeddings Name", settings.azure_openai_embeddings_name),
        ("Azure OpenAI Embeddings Endpoint", settings.azure_openai_embeddings_endpoint),
    ]

# Configure LlamaIndex to use Azure OpenAI for text generation
Settings.llm = AzureOpenAI(
            model=settings.text_model,
            engine=settings.azure_openai_deployment_name,
            deployment_name=settings.azure_openai_deployment_name,
            api_key=settings.azure_openai_api_key,
            azure_endpoint=settings.azure_openai_endpoint,
            api_version=settings.azure_openai_api_version,
        )

# Configure LlamaIndex to use Azure OpenAI for embeddings
Settings.embed_model = AzureOpenAIEmbedding(
            model=settings.azure_openai_embeddings_model,
            deployment_name=settings.azure_openai_embeddings_name,
            api_key=settings.azure_openai_api_key,
            azure_endpoint=settings.azure_openai_embeddings_endpoint,
            api_version=settings.azure_openai_embeddings_api_version,
        )

# Test the setup with a sample document and query
try:
    # Create test index with example document
    documents = [Document.example()]
    index = VectorStoreIndex.from_documents(documents)
    query_engine = index.as_query_engine()
    
    # Run test query
    response = query_engine.query("What is this document about?")
    print(f"Query Response: {response}")
except Exception as e:
    print(f"Error occurred: {str(e)}")
    

HTTP Request is working

[angus-langchain]

Hi @mindful-time, Can you confirm that content encoding is set to proto. Also, could you confirm that if you create a new API key in the app and re-run this code it still doesn't work?

[mindful-time]

Hi @angus-langchain sorry was busy, can we make this open. The issue still persists

[angus-langchain]

Hi @mindful-time, I'm unable to repro with the code you shared. Are you able to trace to LangSmith normally?

https://smith.langchain.com/public/277dcdd3-c0d1-412e-9cd9-ae7abf4b8262/r

[mindful-time]

@angus-langchain do you mean with llama-index or in general with langchain/langgraph ? If it is general with langchain and langgrpah i am , but the llama-index isn't working based on the from traceloop.sdk

[mindful-time]

@angus-langchain changed the code to following seemed to work !

`# initialize Traceloop with LangSmith endpoint and authentication environment variables os.environ["TRACELOOP_BASE_URL"] = "https://api.smith.langchain.com/otel" os.environ["TRACELOOP_HEADERS"] = "x-api-key=" + langsmith_settings.api_key + ",Langsmith-Project=" + langsmith_settings.project

Traceloop.init()`

Changed code: updated code repo langsmith-llamaindex Btw any idea why all the trace aren't coming as one run and three

I think we should use the create_global_handler in llama_index https://docs.llamaindex.ai/en/stable/module_guides/observability/

Found here : https://github.com/run-llama/llama_index/blob/main/llama-index-core/llama_index/core/callbacks/global_handlers.py

PHOENIX_API_KEY = "<PHOENIX_API_KEY>" os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"api_key={PHOENIX_API_KEY}" llama_index.core.set_global_handler( "arize_phoenix", endpoint="https://llamatrace.com/v1/traces" )