RxNetty重用连接

时间:2018-08-15 10:20:23

标签: java rx-java netty netflix-ribbon

我想在没有Spring Cloud的情况下使用Netflix-Ribbon作为TCP客户端负载平衡器,并且编写测试代码。

public class App  implements Runnable
{
    public static String msg = "hello world";
    public BaseLoadBalancer lb;
    public RxClient<ByteBuf, ByteBuf >  client;
    public Server echo;

    App(){
        lb = new BaseLoadBalancer();
        echo = new Server("localhost", 8000);
        lb.setServersList(Lists.newArrayList(echo));
        DefaultClientConfigImpl impl = DefaultClientConfigImpl.getClientConfigWithDefaultValues();
        client = RibbonTransport.newTcpClient(lb, impl);
    }
    public static void main( String[] args ) throws Exception
    {
        for( int i = 40; i > 0; i--)
        {
            Thread t = new Thread(new App());
            t.start();
            t.join();
        }
        System.out.println("Main thread is finished");
    }
    public String sendAndRecvByRibbon(final String data) 
    {
        String response = "";
        try {
            response = client.connect().flatMap(new Func1<ObservableConnection<ByteBuf, ByteBuf>,
                    Observable<ByteBuf>>() {
                public Observable<ByteBuf> call(ObservableConnection<ByteBuf, ByteBuf> connection) {

                    connection.writeStringAndFlush(data);
                    return connection.getInput();
                }
            }).timeout(1, TimeUnit.SECONDS).retry(1).take(1)
                    .map(new Func1<ByteBuf, String>() {
                        public String call(ByteBuf ByteBuf) {
                            return ByteBuf.toString(Charset.defaultCharset());
                        }
                    })
                    .toBlocking()
                    .first();
        }
        catch (Exception e) {
            System.out.println(((LoadBalancingRxClientWithPoolOptions) client).getMaxConcurrentRequests());
            System.out.println(lb.getLoadBalancerStats());
        }
        return response;
    }
    public void run() {
        for (int i = 0; i < 200; i++) {
            sendAndRecvByRibbon(msg);
        }
    }

}

即使sendAndRecvByRibbon设置为true,我也会发现每次调用poolEnabled时都会创建一个新的套接字。所以,这让我感到困惑,我想念什么吗? 并且没有配置池大小的选项,但是有PoolMaxThreadsMaxConnectionsPerHost

我的问题是如何在我的简单代码中使用连接池,而我的sendAndRecvByRibbon有什么问题,它打开一个套接字然后仅使用一次,我该如何重用连接?感谢您的时间。

该服务器只是用pyhton3编写的简单回显服务器,我将conn.close()注释掉,因为我想使用长连接。

import socket
import threading
import time
import socketserver
class ThreadedTCPRequestHandler(socketserver.BaseRequestHandler):
    def handle(self):
        conn = self.request
        while True:
            client_data = conn.recv(1024)
            if not client_data:
                time.sleep(5)
            conn.sendall(client_data)
#            conn.close()

class ThreadedTCPServer(socketserver.ThreadingMixIn, socketserver.TCPServer):
    pass

if __name__ == "__main__":
    HOST, PORT = "localhost", 8000
    server = ThreadedTCPServer((HOST, PORT), ThreadedTCPRequestHandler)
    ip, port = server.server_address
    server_thread = threading.Thread(target=server.serve_forever)
    server_thread.daemon = True
    server_thread.start()
    server.serve_forever()

和mevan的pom,我只是在IED的自动生成的POM中添加了两个依赖项。

<dependency>
    <groupId>commons-configuration</groupId>
    <artifactId>commons-configuration</artifactId>
    <version>1.6</version>
</dependency>
<dependency>
    <groupId>com.netflix.ribbon</groupId>
    <artifactId>ribbon</artifactId>
    <version>2.2.2</version>
</dependency>

用于打印src_port的代码

@Sharable
public class InHandle extends ChannelInboundHandlerAdapter {
    public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
        System.out.println(ctx.channel().localAddress());
        super.channelRead(ctx, msg);
    }
}

public class Pipeline implements PipelineConfigurator<ByteBuf, ByteBuf> {
    public InHandle handler;
    Pipeline() {
        handler = new InHandle();
    }
    public void configureNewPipeline(ChannelPipeline pipeline) {
        pipeline.addFirst(handler);
    }
}

并将client = RibbonTransport.newTcpClient(lb, impl);更改为Pipeline pipe = new Pipeline();client = RibbonTransport.newTcpClient(lb, pipe, impl, new DefaultLoadBalancerRetryHandler(impl));

1 个答案:

答案 0 :(得分:2)

因此,您的App()构造函数将对lb / client / etc进行初始化。

然后,通过在第一个for循环中调用new App(),使用40个不同的RxClient实例(默认情况下每个实例都有自己的池)启动40个不同的线程。为了清楚起见,在这里生成多个RxClient实例的方式不允许它们共享任何公共池。尝试改用一个RxClient实例。

如果您像下面那样更改主要方法,它会停止创建额外的套接字吗?

    public static void main( String[] args ) throws Exception
    {
        App app = new App() // Create things just once
        for( int i = 40; i > 0; i--)
        {
            Thread t = new Thread(()->app.run()); // pass the run()
            t.start();
            t.join();
        }
        System.out.println("Main thread is finished");
    }

如果上述方法不能完全解决问题(至少它将减少40倍的已创建套接字数)-请您说明如何确定:

  

我发现每次调用sendAndRecvByRibbon都会创建一个新的套接字

以及用此行更新构造函数后的测量结果是什么

    DefaultClientConfigImpl impl = DefaultClientConfigImpl.getClientConfigWithDefaultValues();
    impl.set(CommonClientConfigKey.PoolMaxThreads,1); //Add this one and test

更新

是的,看着sendAndRecvByRibbon似乎没必要通过调用close来将Python: How to turn a dictionary of Dataframes into one big dataframe with column names being the key of the previous dict?标记为PooledConnection,一旦您不希望再读到它。

只要您期望唯一的一次读取事件,只需更改此行

connection.getInput()

return connection.getInput().zipWith(Observable.just(connection), new Func2<ByteBuf, ObservableConnection<ByteBuf, ByteBuf>, ByteBuf>() {
                        @Override
                        public ByteBuf call(ByteBuf byteBuf, ObservableConnection<ByteBuf, ByteBuf> conn) {
                            conn.close();
                            return byteBuf; 
                        }
                    });

请注意,如果您要设计基于TCP的更复杂的协议,则可以分析输入的字节缓冲区以查找您的特定“通信结束”符号,该符号指示可以将连接返回到池中。