自定义追踪

自定义追踪

演示了高级追踪模式，包括自定义追踪、跨度和复杂多智能体系统的流程组织。

本演示所展示的内容

• 自定义追踪：在单一流程中对多个智能体运行进行分组
• 自定义跨度：添加业务逻辑监控点
• 层次化追踪：用于复杂操作的嵌套跨度
• 追踪元数据：通过分组和元数据组织追踪

快速入门

1. 安装 OpenAI Agents SDK：

   pip install openai-agents
   ​

2. 设置环境：

   cp ../env.example .env
   # Edit .env and add your OpenAI API key
   ​

3. 运行智能体：

   import asyncio
   from agent import main
   
   # Test custom tracing patterns
   asyncio.run(main())
   ​

关键概念

• trace() 上下文管理器：创建自定义流程分组
• custom_span()：添加业务逻辑监控
• 追踪元数据：流程命名与组织
• 层次结构：用于复杂操作的嵌套跨度

自定义追踪模式

多步骤流程追踪

with trace("Research and Analysis Workflow") as workflow_trace:
    # Step 1: Research
    research_result = await Runner.run(research_agent, "Research AI in healthcare")
    
    # Step 2: Analysis  
    analysis_result = await Runner.run(analysis_agent, f"Analyze: {research_result.final_output}")
    
    # Step 3: Summary
    summary_result = await Runner.run(analysis_agent, f"Summarize: {analysis_result.final_output}")
​

自定义业务逻辑跨度

with trace("Document Processing Workflow") as doc_trace:
    
    with custom_span("Data Preparation") as prep_span:
        # Your business logic here
        data = prepare_data()
        prep_span.add_event("Data loaded", {"records": 100})
        prep_span.add_event("Data validated", {"errors": 0})
        
    with custom_span("AI Processing") as ai_span:
        result = await Runner.run(agent, "Process the data")
        ai_span.add_event("Processing complete", {
            "output_length": len(result.final_output)
        })
​

层次化跨度

with trace("E-commerce Order Processing") as order_trace:
    
    with custom_span("Order Validation") as validation_span:
        
        # Nested span for inventory check
        with custom_span("Inventory Check") as inventory_span:
            inventory_span.add_event("Stock verified", {"available": True})
        
        # Nested span for payment validation
        with custom_span("Payment Validation") as payment_span:
            payment_span.add_event("Payment authorized", {"amount": 99.99})
​

高级功能

追踪元数据与分组

conversation_id = "conv_12345"

# First interaction in conversation
with trace(
    "Customer Support - Initial Inquiry",
    group_id=conversation_id,
    metadata={"customer_id": "cust_789", "priority": "high"}
) as trace1:
    result1 = await Runner.run(support_agent, "How do I reset my password?")

# Follow-up interaction in same conversation
with trace(
    "Customer Support - Follow-up",
    group_id=conversation_id,
    metadata={"customer_id": "cust_789", "interaction": 2}
) as trace2:
    result2 = await Runner.run(support_agent, f"Based on this context: {result1.final_output}")
​

事件追踪

with custom_span("Business Process") as span:
    span.add_event("Process started", {"timestamp": datetime.now()})
    # Business logic here
    span.add_event("Milestone reached", {"progress": "50%"})
    # More business logic
    span.add_event("Process completed", {"status": "success"})
​

自定义追踪的优势

流程组织

• 分组相关操作：多个智能体运行集中于单一追踪
• 业务逻辑可见性：同时监控自定义流程与人工智能
• 性能分析：追踪端到端流程性能

生产环境监控

• 错误关联：跨多个组件的故障链路
• 性能优化：识别复杂流程中的瓶颈
• 用户旅程追踪：跟踪跨交互的对话过程

调试与分析

• 复杂流程理解：可视化多步骤流程
• 上下文保留：保持相关操作之间的关系
• 元数据组织：按业务标准筛选与搜索追踪

下一步

• 默认追踪 - 基础追踪原理
• 教程 11：生产环境模式 - 实际部署应用

免责声明：
本文档采用基于机器的 AI 翻译服务进行翻译。尽管我们力求准确，但请注意，自动翻译可能存在错误或不准确之处。应以原文语言版本的文档作为权威依据。如需获取关键信息，建议使用专业的人工翻译。对于因使用本翻译而产生的任何误解或误读，我们概不负责。