Too many DubboClientHandler threads in consumer-消费者中建立了太多DubboClientHandler 线程

[Jaskey]

Dubbo Version: 2.5.3

I have a large provider cluster with more than 30 instances. And I found that in client side, the consumer uses a stand alone thread pools for each instances, so there are 30 thread pool to consume the service, which I can find out from the jstack that I have many diffrent thread groups named like DubboClientHandler-IP1-XXX.

Question1:
I guess the consumer uses stand alone cached thread pool for each provider instances but with the max pool size of "default.actives", am I right?

Question2:
But the problem is that, I hope to maximize the tps , so I provide a very large cluster, but now the consumer will suffer from high cpu consumption due to large thread counts.

Can I change the thread model in consumer side? Like I use a shared thread pool so that I can control the maximum thread size.

---

我有一个非常大的provider 集群（30个实例）。然后我们发现在客户端，为每一个实例，都准备了一个独立线程池，所以我们看到有30个线程池以去发起消费请求，这个从jstack堆栈上可以证明，因为里面很多线程组的名字是类似DubboClientHandler-IP1-XXX，其中ip都是不一样的。

问题1：

按照我的猜测，每个实例是应该会有一个cached的线程池（可回收），然后其中的最大线程数是default.actives，是否如此？

问题2：

之所以我们提供如此大的服务集群，是因为我们希望最大化TPS，但现在集群是大了，但是消费者又因此导致巨大的线程消耗（高CPU）。我们能否通过配置改变这个线程的模型，例如用一个共享的线程池去处理所有的服务实例，这样最大的线程数便可控了。

[chickenlj]

Duplicate of #2013