HTTP Instrumentation Breaks Azure AI Agents SDK

[Shreyas-Microsoft]

Describe your environment

Environment

• OpenTelemetry Version: opentelemetry-auto-instrumentation (latest)
• Python Version: 3.11.9
• Azure Monitor OpenTelemetry: 1.7.0
• Azure AI Projects SDK: 1.0.0
• Operating System: Windows/Linux (Azure Container Apps)
• Deployment: Azure Container Apps with Application Insights

What happened?

Description

Expected Behavior

Azure AI Agents SDK should work normally with structured response formats when OpenTelemetry HTTP instrumentation is enabled.

Actual Behavior

When OpenTelemetry auto-instrumentation is enabled via Azure Monitor, Azure AI service calls fail with:

OpenAI APIException: Error code: 400 - {
  'error': {
    'message': 'Unknown response format <class \'dict\'>', 
    'type': 'invalid_request_error', 
    'param': None, 
    'code': None
  }
}

Root Cause Analysis

1. Azure AI Agents SDK creates ResponseFormatJsonSchemaType objects for structured responses
2. OpenTelemetry HTTP instrumentation intercepts and modifies HTTP requests containing these objects
3. The serialization/instrumentation process corrupts the response format specification
4. Azure AI service receives malformed request and returns the error

Steps to Reproduce

Reproduction Steps

1. Install Dependencies

pip install azure-monitor-opentelemetry==1.7.0
pip install azure-ai-projects==1.0.0
pip install opentelemetry-auto-instrumentation

2. Enable Auto-Instrumentation

from azure.monitor.opentelemetry import configure_azure_monitor

# This line enables the problematic instrumentation
configure_azure_monitor(
    connection_string="InstrumentationKey=your-key-here"
)

3. Create Azure AI Agent with Structured Response

from azure.ai.projects.models import (
    ResponseFormatJsonSchemaType, 
    ResponseFormatJsonSchema
)
from azure.ai.projects import AIProjectClient

# Create response format object
response_format_obj = ResponseFormatJsonSchemaType(
    json_schema=ResponseFormatJsonSchema(
        name="TestResponse",
        description="Test response format",
        schema={
            "type": "object", 
            "properties": {
                "result": {"type": "string"}
            }
        },
    )
)

# This call FAILS when telemetry is enabled
agent_definition = await ai_project_client.agents.create_agent(
    model="gpt-4o",
    name="test-agent",
    instructions="Test instructions",
    response_format=response_format_obj,  # ← CAUSES ERROR
)

4. Error Occurs

The create_agent call fails with the "Unknown response format" error when OpenTelemetry instrumentation is active.

Expected Result

Expected Behavior

Azure AI Agents SDK should work normally with structured response formats when OpenTelemetry HTTP instrumentation is enabled.

Actual Result

Actual Behavior

When OpenTelemetry auto-instrumentation is enabled via Azure Monitor, Azure AI service calls fail with:

OpenAI APIException: Error code: 400 - {
  'error': {
    'message': 'Unknown response format <class \'dict\'>', 
    'type': 'invalid_request_error', 
    'param': None, 
    'code': None
  }
}

Additional context

No response

Would you like to implement a fix?

None

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}