Add Enhanced Jaeger Tracing Support (Updated)

[ajeetraina]

Enhanced Jaeger Tracing Implementation (Updated)

This PR adds comprehensive Jaeger tracing support to the GenAI application, providing detailed insights into request handling and model inference.

Note: This PR has been updated to be compatible with the build fixes in PR #14. Please merge PR #14 first.

Key Changes

1. Enhanced Tracing Package:

   • Improved error handling and span management
   • Added support for child spans, events, and detailed attributes
   • Fixed issues with the existing AddAttribute function

2. New Tracing Middleware:

   • Added HTTP middleware to automatically trace all requests
   • Captures HTTP method, URL, status code, and other request details
   • Properly propagates trace context through the application

3. Model Inference Tracing:

   • Added a TracedModelInference struct to track model performance
   • Detailed spans for each phase of model inference
   • Records metrics like Time To First Token (TTFT)
   • Creates events for significant milestones (first token, completion)

4. Metrics Middleware:

   • Refactored metrics collection into a middleware
   • Ensures consistent metrics for all requests

5. Fixed Package Structure:

   • Moved responseWriterWrapper to a dedicated file
   • Removed duplicate type definitions

Usage

The tracing is automatically enabled when the TRACING_ENABLED environment variable is set to true and uses the Jaeger endpoint specified in OTLP_ENDPOINT (defaults to jaeger:4318).

Benefits

• Detailed Performance Insights: Identify bottlenecks in model loading, inference, and response handling
• Error Tracing: Quickly pinpoint where errors occur in the request processing pipeline
• End-to-End Visibility: Track a request from frontend to model and back
• Token Metrics: See token generation rates and timing information

This implementation follows OpenTelemetry best practices and is designed to work seamlessly with Jaeger UI.