feat: Enhance streaming API timeout handling with mathematical modeling

[qizwiz]

PR: feat: Enhance streaming API timeout handling with mathematical modeling

Overview

This PR addresses GitHub issue #239 by implementing a comprehensive mathematical modeling approach to understand and solve the streaming API timeout issue that occurs after 64 seconds.

Problem

The streaming API setup was timing out after 64 seconds, causing user frustration and limiting the tool's effectiveness for large requests. The error message provided generic troubleshooting tips but didn't offer specific solutions based on the request characteristics.

Solution

We've implemented a comprehensive mathematical modeling approach to understand and solve this timeout issue:

1. Mathematical Modeling

We created a StreamingTimeoutModel that calculates expected streaming request times based on:

• Data size and complexity
• System load factors
• Processing rates
• Network latency

This allows us to predict when timeouts will occur and recommend appropriate solutions.

2. Adaptive Timeout Calculation

Instead of fixed timeouts, we now calculate adaptive timeouts based on request characteristics:

Adaptive Timeout = Base Timeout + 
                   (Data Size × 0.05) + 
                   (Complexity × 0.1) + 
                   (System Load × 20)

3. Enhanced Error Messaging

When timeouts occur, we now provide more specific troubleshooting guidance based on the request characteristics:

• For large requests: Suggestions to break into smaller chunks
• For complex requests: Recommendations for progressive summarization
• Configuration suggestions: Current vs. recommended timeout values

4. CLI Configuration Options

New CLI options allow users to configure timeout behavior:

• --openai-timeout: Set API timeout in milliseconds
• --openai-max-retries: Set maximum retry attempts

5. Configuration Recommendations

The system now provides configuration recommendations based on analysis of current settings.

Technical Implementation

Core Changes

1. Created StreamingTimeoutModel - A mathematical model for predicting and preventing timeouts
2. Enhanced OpenAIContentGenerator - Added adaptive timeout handling and improved error messaging
3. Added CLI options - New --openai-timeout and --openai-max-retries configuration options
4. Added configuration recommendations - Based on analysis of current settings

Files Modified

• packages/core/src/models/streamingTimeoutModel.ts - New mathematical model
• packages/core/src/models/streamingTimeoutModel.test.ts - Tests for the model
• packages/core/src/models/streamingTimeoutModel.verification.test.ts - Formal verification tests
• packages/core/src/core/openaiContentGenerator.ts - Enhanced timeout handling
• packages/cli/src/config/config.ts - Added CLI options

Usage Examples

CLI Usage

# Increase timeout for large requests
qwen --openai-timeout 300000 --prompt "Analyze this large codebase"

# Set retry policy
qwen --openai-max-retries 5 --prompt "Complex analysis task"

# Combine both for maximum reliability
qwen --openai-timeout 300000 --openai-max-retries 5 --prompt "Analyze large, complex codebase"

Configuration File

{
  "contentGenerator": {
    "timeout": 120000,
    "maxRetries": 3,
    "samplingParams": {
      "temperature": 0.7,
      "max_tokens": 2048
    }
  }
}

Testing

All tests pass, including new tests for the streaming timeout model:

• Unit tests for mathematical calculations
• Integration tests with the OpenAI content generator
• CLI configuration tests
• Formal verification tests proving deterministic behavior

Future Improvements

1. Machine Learning Approach: Use historical data to predict optimal timeouts
2. Dynamic Adjustment: Real-time adjustment of timeouts based on current performance
3. Progressive Enhancement: Start with conservative timeouts and increase based on success patterns

This solution transforms a frustrating timeout issue into an opportunity for intelligent, adaptive system behavior that improves the user experience for large and complex requests.

Fixes #239

[Xinlong-Wu]

Hi, @tanzhenxin

Could anyone review this patch? This issue has been bothering me for a long time.