LangGraph状态机：在生产环境中管理复杂的代理任务流程

什么是LangGraph？

LangGraph是一个专为LLM应用设计的工作流编排框架。其核心原则是：

将复杂任务分解为状态和转换
管理状态转换逻辑
处理任务执行过程中的各种异常

想象一下购物过程：浏览 → 加入购物车 → 结账 → 支付。LangGraph帮助我们高效地管理这样的工作流。

核心概念

1. 状态

状态就像任务执行中的检查点：

from typing import TypedDict, List

class ShoppingState(TypedDict):
    # 当前状态
    current_step: str
    # 购物车商品
    cart_items: List[str]
    # 总金额
    total_amount: float
    # 用户输入
    user_input: str

class ShoppingGraph(StateGraph):
    def __init__(self):
        super().__init__()

        # 定义状态
        self.add_node("browse", self.browse_products)
        self.add_node("add_to_cart", self.add_to_cart)
        self.add_node("checkout", self.checkout)
        self.add_node("payment", self.payment)

2. 状态转换

状态转换定义了任务流程的“路线图”：

class ShoppingController:
    def define_transitions(self):
        # 添加转换规则
        self.graph.add_edge("browse", "add_to_cart")
        self.graph.add_edge("add_to_cart", "browse")
        self.graph.add_edge("add_to_cart", "checkout")
        self.graph.add_edge("checkout", "payment")

    def should_move_to_cart(self, state: ShoppingState) -> bool:
        """确定是否应转换到购物车状态"""
        return "add to cart" in state["user_input"].lower()

3. 状态持久化

为了确保系统可靠性，我们需要持久化状态信息：

class StateManager:
    def __init__(self):
        self.redis_client = redis.Redis()

    def save_state(self, session_id: str, state: dict):
        """将状态保存到Redis"""
        self.redis_client.set(
            f"shopping_state:{session_id}",
            json.dumps(state),
            ex=3600  # 1小时过期
        )

    def load_state(self, session_id: str) -> dict:
        """从Redis加载状态"""
        state_data = self.redis_client.get(f"shopping_state:{session_id}")
        return json.loads(state_data) if state_data else None
  4. 错误恢复机制

任何步骤都可能失败，我们需要优雅地处理这些情况：

class ErrorHandler:
    def __init__(self):
        self.max_retries = 3

    async def with_retry(self, func, state: dict):
        """带重试机制的函数执行"""
        retries = 0
        while retries < self.max_retries:
            try:
                return await func(state)
            except Exception as e:
                retries += 1
                if retries == self.max_retries:
                    return self.handle_final_error(e, state)
                await self.handle_retry(e, state, retries)

    def handle_final_error(self, error, state: dict):
        """处理最终错误"""
        # 保存错误状态
        state["error"] = str(error)
        # 回滚到最后一个稳定状态
        return self.rollback_to_last_stable_state(state)

实际案例：智能客服系统

让我们来看一个实际例子——智能客服系统：

from langgraph.graph import StateGraph, State

class CustomerServiceState(TypedDict):
    conversation_history: List[str]
    current_intent: str
    user_info: dict
    resolved: bool

class CustomerServiceGraph(StateGraph):
    def __init__(self):
        super().__init__()

        # 初始化状态
        self.add_node("greeting", self.greet_customer)
        self.add_node("understand_intent", self.analyze_intent)
        self.add_node("handle_query", self.process_query)
        self.add_node("confirm_resolution", self.check_resolution)

    async def greet_customer(self, state: State):
        """问候客户"""
        response = await self.llm.generate(
            prompt=f"""
            Conversation history: {state['conversation_history']}
            Task: Generate appropriate greeting
            Requirements:
            1. Maintain professional friendliness
            2. Acknowledge returning customers
            3. Ask how to help
            """
        )
        state['conversation_history'].append(f"Assistant: {response}")
        return state

    async def analyze_intent(self, state: State):
        """理解用户意图"""
        response = await self.llm.generate(
            prompt=f"""
            Conversation history: {state['conversation_history']}
            Task: Analyze user intent
            Output format:
            {{
                "intent": "refund/inquiry/complaint/other",
                "confidence": 0.95,
                "details": "specific description"
            }}
            """
        )
        state['current_intent'] = json.loads(response)
        return state
  用法

# 初始化系统
graph = CustomerServiceGraph()
state_manager = StateManager()
error_handler = ErrorHandler()

async def handle_customer_query(user_id: str, message: str):
    # 加载或创建状态
    state = state_manager.load_state(user_id) or {
        "conversation_history": [],
        "current_intent": None,
        "user_info": {},
        "resolved": False
    }

    # 添加用户消息
    state["conversation_history"].append(f"User: {message}")

    # 执行状态机流程
    try:
        result = await graph.run(state)
        # 保存状态
        state_manager.save_state(user_id, result)
        return result["conversation_history"][-1]
    except Exception as e:
        return await error_handler.with_retry(
            graph.run,
            state
        )