我假设您的意思是当您启用keep-alive时，Tomcat会因为套接字太多而过载。Tomcat有两个设置来限制keep-alive会话的数量。第一个是keep-alive会话的最大数量，默认值为100，但您可以使用system属性降低该值：

server.tomcat.max-keep-alive-requests

将此值设置为小于100的值，直到服务器不再使用过多的保持活动套接字。将此值设置为正确的最大值后，应查看：

server.tomcat.keep-alive-timeout

• 它决定保持活动状态会话的保留时间。缩短此时间（以毫秒为单位，而不是以秒为单位）将使频繁发送请求的客户端优先于不频繁发送请求的客户端。默认值为60000（1分钟）。

PS你还没有说你使用的是哪个Web服务器，但是默认的是Tomcat，所以我假设你使用的是Tomcat，否则有类似的设置，例如Jetty。
以上方法应该可以解决您的问题，但如果不能，您仍然可以按如下所示配置故障转移逻辑：

// 5 attempts, connect timeout 5 seconds, receive timeout 20 seconds (per attempt)
  // Should take about 30 seconds to fail if the server is completely down
  final RetryStrategy retryStrategy = new RetryStrategy();
  retryStrategy.setMaxNumberOfRetries(5);
  retryStrategy.setDelayBetweenRetries(1000);
  FailoverFeature failoverFeature = new FailoverFeature();
  failoverFeature.setStrategy(retryStrategy);
  failoverFeature.setTargetSelector(new FailoverTargetSelector(address));
  jaxWsProxyFactoryBean.getFeatures().add(failoverFeature);
  serviceBean = jaxWsProxyFactoryBean.create();

  HTTPConduitConfigurer httpConduitConfigurer = (name, address, c) ->
  {
     HTTPClientPolicy httpClientPolicy = new HTTPClientPolicy();
     httpClientPolicy.setConnectionTimeout(5000);
     httpClientPolicy.setReceiveTimeout(20000);
     c.setClient(httpClientPolicy);
  };

  Client client = ClientProxy.getClient(serviceBean);
  client.getBus().setExtension(httpConduitConfigurer, HTTPConduitConfigurer.class);