一个开发者的 AI 实战：用最少的提示词解决最头疼的报错

org.springframework.beans.factory.BeanCreationException:
Error creating bean with name 'orderController': Unsatisfied dependency expressed through field 'orderService';
nested exception is org.springframework.beans.factory.BeanCreationException:
Error creating bean with name 'orderServiceImpl' defined in file [.../OrderServiceImpl.class]:
Bean instantiation via constructor failed;
nested exception is org.springframework.beans.BeanInstantiationException:
Failed to instantiate [...OrderServiceImpl]: Constructor threw exception;
nested exception is java.net.ConnectException: Connection refused

【第一段】我原本想让它发生什么
我启动 Spring Boot 服务，期望正常启动并注册 Bean。

【第二段】实际发生了什么（原文，别转述）
<把完整堆栈原文贴进来，一行不删>

【第三段】我已经试过什么
1. 检查过 OrderServiceImpl 的注解，@Service 是有的
2. 确认过 risk.internal 域名在预发容器里 ping 不通
3. 本地启动是好的

@Service
public class OrderServiceImpl implements OrderService {

    private final RiskClient riskClient;
    private final OrderMapper orderMapper;
    private Map<String, Rule> ruleCache;

    public OrderServiceImpl(RiskClient riskClient,
                            OrderMapper orderMapper,
                            @Value("${risk.endpoint}") String endpoint) {
        this.riskClient = riskClient;
        this.orderMapper = orderMapper;
        // 启动时先把规则全量拉下来，省得运行时再查
        this.ruleCache = riskClient.fetchRules(endpoint);
    }

    @Override
    public OrderDetail getDetail(String orderId) {
        Rule rule = ruleCache.get(orderId);
        return orderMapper.selectById(orderId, rule);
    }
}

根因不在依赖注入，在最内层异常：
nested exception is java.net.ConnectException: Connection refused

BeanCreationException 只是外层包装。真正的链路是：
OrderServiceImpl 构造器里调用了 riskClient.fetchRules(endpoint)，
这是一个启动时的同步远程调用，下游连不上，构造器抛异常，
Spring 把它包装成了 Bean 创建失败。

你试过的第 2 条已经印证了这一点（预发 ping 不通 risk.internal）。

@Service
public class OrderServiceImpl implements OrderService {

    private final RiskClient riskClient;
    private final OrderMapper orderMapper;

    @Value("${risk.endpoint}")
    private String endpoint;

    private volatile Map<String, Rule> ruleCache = Collections.emptyMap();

    /** 启动后异步预热，失败不阻塞启动 */
    @PostConstruct
    public void warmUp() {
        try {
            this.ruleCache = riskClient.fetchRules(endpoint);
            log.info("风控规则预热完成，共 {} 条", ruleCache.size());
        } catch (Exception e) {
            log.warn("风控规则预热失败，启动继续，运行时降级为实时查询", e);
        }
    }

    @Override
    public OrderDetail getDetail(String orderId) {
        Rule rule = ruleCache.get(orderId);
        if (rule == null) {
            rule = riskClient.fetchOne(endpoint, orderId); // 兜底：实时查
        }
        return orderMapper.selectById(orderId, rule);
    }
}

Error: socket hang up
    at connResetException (node:internal/errors:704:14)
    at Socket.socketOnEnd (node:_http_client:518:23)
    at TLSSocket.emit (node:events:525:35)

const http = require('http');

const agent = new http.Agent({ keepAlive: true, maxSockets: 100 });

async function callDownstream(path) {
  return new Promise((resolve, reject) => {
    const req = http.request(
      { host: 'risk.internal', port: 8080, path, method: 'GET', agent },
      (res) => {
        let body = '';
        res.on('data', (chunk) => (body += chunk));
        res.on('end', () => resolve(body));
      }
    );
    req.on('error', reject);
    req.end();
  });
}

已经试过：
1. 给 req 加了 setTimeout(5000)，没用
2. 加了三次重试，错误率下降但还是偶发
3. 把 maxSockets 从 100 调到 20，没用

你三个方案都在"客户端单侧"打转，但这个错误的典型成因是连接复用竞态：

下游 nginx 的 keepalive_timeout 通常是 15s 或更短。
你的 agent 开了 keepAlive，但没设空闲回收时间，
于是客户端认为这条连接还活着，继续往里发请求，
可服务端早就把这条连接关了 —— 请求发出去，连接已断，就是 socket hang up。

特征正好对上你说的：低 QPS 时不出现，QPS 一上来就偶发，本地复现不了。

const agent = new http.Agent({
  keepAlive: true,
  maxSockets: 100,
  keepAliveMsecs: 5000,   // 心跳间隔要明显小于下游的 idle timeout
  timeout: 10000,         // socket 空闲 10s 后客户端主动回收，抢在 Nginx 关之前
});

// 兜底：被对端关掉的连接，主动从池里摘掉，别留给下一次复用
agent.on('free', (socket) => {
  if (socket.destroyed || !socket.writable) {
    socket.destroy();
  }
});

java.lang.NoSuchMethodError: com.google.common.collect.Sets.newConcurrentHashSet()Ljava/util/Set;
    at com.xxx.cache.RuleCache.init(RuleCache.java:41)
    at com.xxx.cache.RuleCache$$EnhancerBySpringCGLIB$$8f2c1a0.<init>(<generated>)
    ... 30 more

@Component
public class RuleCache {
    private final Set<String> keys = Sets.newConcurrentHashSet();

    public void init() {
        keys.add("default");
    }
}

升级 Guava 无效，说明运行时 classpath 里根本不是你声明的那个版本。

NoSuchMethodError 99% 是"编译时高版本、运行时低版本"，
不是"你没引入"。去查谁把老版本带进来了：

mvn dependency:tree -Dincludes=com.google.guava:guava

$ mvn dependency:tree -Dincludes=com.google.guava:guava

[INFO] com.xxx:order-service:jar:1.0.0
[INFO] +- com.google.guava:guava:jar:30.1-jre:compile
[INFO] \- com.xxx:legacy-common:jar:2.3.1:compile
[INFO]    \- com.third:sdk-report:jar:1.4.0:compile
[INFO]       \- com.google.guava:guava:jar:11.0.2:compile

<dependency>
    <groupId>com.xxx</groupId>
    <artifactId>legacy-common</artifactId>
    <version>2.3.1</version>
    <exclusions>
        <exclusion>
            <groupId>com.google.guava</groupId>
            <artifactId>guava</artifactId>
        </exclusion>
    </exclusions>
</dependency>