用Jetty和redis实现接入服务器adapter

传统的服务器端为若干个客户端提供服务，一般需要开启多个服务器端进程。为了进一步提升服务器端的处理能力，可以如下图所示将服务解耦为两部分（adapter与workers），它们之间通过消息队列传输数据，其中workers处理具体业务，adapter负责接入请求以及反馈结果，具体包含下面两个工作。

1，将所有客户端的请求发送到消息队列（进而传给后台处理）

2，后台处理完毕后将结果返回响应队列，client adapter获取到结果后返回给相应客户端。

流程图如下：

我们选择用Jetty（8），redis以及php简单实现这个场景，主要用到Jetty的continuation机制和redis的list先进先出数据结构

接入服务器：

A.1，先配置一个服务器如下，同时开启一个守护线程阻塞监听response queue（用到json lib库以及jedis库）（使用专用线程处理响应可以避免“惊群”现象，不影响业务线程）

package test;

import java.util.HashMap;
import java.util.List;

import org.eclipse.jetty.continuation.Continuation;
import org.eclipse.jetty.server.*;
import org.eclipse.jetty.server.nio.SelectChannelConnector;
import org.eclipse.jetty.servlet.ServletContextHandler;
import org.eclipse.jetty.servlet.ServletHolder;
import org.eclipse.jetty.util.thread.QueuedThreadPool;

import org.json.simple.*;

import redis.clients.jedis.Jedis;

public class PJetty{

    public static HashMap<String,Continuation>globalMap = new HashMap<String,Continuation>();

    // 用一个守护线程阻塞等待结果队列返回数据
    public static class DaemonThread extends Thread{

        private JSONObject obj = new JSONObject();

        private Jedis jedis = new Jedis("127.0.0.1",6379);
        private List<String> res;

        public void run(){

            while(true){
                // 阻塞等待响应队列
                res = jedis.brpop(0, "response_queue");

                // 获取结果信息
                String response = res.get(1);

                obj=(JSONObject) JSONValue.parse(response);
                String request_sid = obj.get("request_sid").toString();
                String result = obj.get("results").toString();

                if(request_sid == null){
                    continue;
                }

                // 通过消息中的连接的sessonid获取到响应的continuation实例，然后设置结果信息再唤醒实例
                Continuation con = globalMap.get(request_sid);
                if(con == null){continue;}
                globalMap.remove(request_sid);

                //客户端异常断开这里会抛错
                try{
                    con.setAttribute("results", result);
                    con.resume();
                } catch(Exception e){
                    continue;
                }
            }

        }
    }

    public static void main(String[] args) throws Exception {

        //开启守护线程去阻塞等待响应结果队列，唤醒请求
        DaemonThread dt = new DaemonThread();
        dt.start();

        //设置connectors
        SelectChannelConnector connector1 = new SelectChannelConnector();
        connector1.setPort(1987);
        connector1.setThreadPool(new QueuedThreadPool(5));

        Server server = new Server();
        server.setConnectors(new Connector[]{connector1});

        //使用servlet处理请求
        ServletContextHandler context = new ServletContextHandler(ServletContextHandler.SESSIONS);
        context.setContextPath("/");
        context.addServlet(new ServletHolder(new NonBlockingServlet()), "/fetch");
        server.setHandler(context);

        server.start();
        server.join();
    }
}

A.2，实现自定义的servlets接受前端client连接，将请求信息传入队列request queue

package test;

import java.io.IOException;

import javax.servlet.http.HttpServlet;
import javax.servlet.http.HttpServletRequest;
import javax.servlet.http.HttpServletResponse;
import javax.servlet.http.HttpSession;

import org.eclipse.jetty.continuation.Continuation;
import org.eclipse.jetty.continuation.ContinuationSupport;
import org.json.simple.JSONObject;

import redis.clients.jedis.Jedis;

public class NonBlockingServlet extends HttpServlet {

    /**
     * generated serialize number
     */
    private static final long serialVersionUID = 3313258432391586994L;

    protected void doGet(HttpServletRequest request, HttpServletResponse response) throws IOException
    {
        // 用sleeptime来模拟后台工作量
         String sleepTime = request.getParameter("st");
         if(sleepTime == null){
             sleepTime = "0";
         }

         // 查看结果队列是否返回此连接的请求结果
         Object results = request.getAttribute("results");
         if (results==null) // 如果异步处理尚未返回结果
          {
           Continuation continuation = ContinuationSupport.getContinuation(request);

           if(continuation.isInitial()){
               // 设置连接超时时间
                continuation.setTimeout(10000);
                   response.setContentType("text/plain");
                response.getWriter().flush();

                HttpSession session=request.getSession();
                String sid = session.getId();

                Jedis jedis = new Jedis("127.0.0.1",6379);
                //将请求连接sessionid以及请求内容encode后传到处理队列中
                JSONObject obj=new JSONObject();
                obj.put("request_sid",sid);
                obj.put("params",sleepTime);

                jedis.lpush("request_queue", obj.toJSONString());

                //将连接和continuation实例做个映射关系存到全局hashmap中，不确定这里是否应该加锁
                PJetty.globalMap.put(sid, continuation);
           }

           // 判断是否超时
           if (continuation.isExpired())
           {
             // 返回超时Response
             response.getWriter().println("timeout");
                response.getWriter().flush();
             return;
           }

           // 挂起HTTP连接
           continuation.suspend(); 

           return; // or continuation.undispatch();
         }

         // 连接恢复后返回结果
         response.getWriter().println("Got Result:\t" + results);
         response.getWriter().flush();
    }
}

业务服务器：

B，实现后端worker.php（可以自定义worker进程数，我这里设置为5个php进程，进程数多能获取更好的并发）（用到predis库）

#!/root/bin/php
<?php

require_once("lib/Predis/Autoloader.php");

function worker_thread()
{
        Predis\Autoloader::register();
        $redis = new Predis\Client('tcp://127.0.0.1:6379');

                while(true){
                        try{
                                $request = $redis->brpop("request_queue", 0);
                        } catch(Exception $e){
                                continue;
                        }
                        /** demo
                         array(2) {
                         [0]=>
                         string(13) "request_queue"
                         [1]=>
                         string(55) "{"request_sid":"q0muxazo8k1h1k3uw85wuayh","params":"4"}"
                         }
                         */
                        $request = json_decode($request[1], true);
                        // sleep represents work loads
                        sleep(intval($request["params"]));
                        $results = array("request_sid"=>$request["request_sid"], "results"=>$request["params"]);
                        $response = json_encode($results);
                        $redis->lpush("response_queue",$response);
                }
}

//开启多个worker进程提供服务
for ($worker_nbr = 0; $worker_nbr < 5; $worker_nbr++) {
        $pid = pcntl_fork();
        if ($pid == 0) {
                worker_thread();

                return;
        }
}

?>

运行结果如下：

这只是一个简单的demo，为了防止redis，workers进程挂掉或者客户端异常断开，还需要做些异常处理，比如设置请求超时，捕获一些空指针等，超时需要将continuation从globalMap中剔除，防止内存得不到释放。

root # for((i=10;i>=1;i--)) ; do lynx -dump http://127.0.0.1:1987/fetch?st=$i & done
[] 14112
[] 14113
[] 14114
[] 14115
[] 14116
[] 14117
[] 14118
[] 14119
[] 14120
[] 14121
root # Got Result:     3
Got Result:     4
Got Result:     2
Got Result:     7
Got Result:     1
Got Result:     9
Got Result:     6
timeout
timeout
timeout

[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]   Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]-  Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i
[]+  Done                    lynx -dump http://127.0.0.1:1987/fetch?st=$i

redis数据库中存储的内容如下，可以看出虽然经后台处理后顺序变化了，但是对应关系正确，接入服务器能够根据request_sid把结果返回给相应的用户：

redis 127.0.0.1:6379> lrange request_queue 0 15
 1) "{\"request_sid\":\"igiwkwnb715aphw8uvtfa6rj\",\"params\":\"3\"}"
 2) "{\"request_sid\":\"wsrglxa3h6ef19ik5i0nbiiys\",\"params\":\"2\"}"
 3) "{\"request_sid\":\"tyiqoj6awj5t16ddpqusftwc8\",\"params\":\"6\"}"
 4) "{\"request_sid\":\"1052tgkiyy7c31bmxjtbom7ca\",\"params\":\"5\"}"
 5) "{\"request_sid\":\"17jo1xwnnkd3h1mhcqcfplrl5k\",\"params\":\"8\"}"
 6) "{\"request_sid\":\"1xk521sq6vmmf6enxauwzduj9\",\"params\":\"4\"}"
 7) "{\"request_sid\":\"1cxnir1slgjiq1o2n3xwznh0kk\",\"params\":\"9\"}"
 8) "{\"request_sid\":\"961vf8hao3stsv4vt1qif3ws\",\"params\":\"7\"}"
 9) "{\"request_sid\":\"35pfn5au6p8qdbri17p636si\",\"params\":\"10\"}"
10) "{\"request_sid\":\"1ca4wy8qsfr7av0hwk8xtlqhp\",\"params\":\"1\"}"

redis 127.0.0.1:6379> lrange response_queue 0 15
 1) "{\"request_sid\":\"tyiqoj6awj5t16ddpqusftwc8\",\"results\":\"6\"}"
 2) "{\"request_sid\":\"igiwkwnb715aphw8uvtfa6rj\",\"results\":\"3\"}"
 3) "{\"request_sid\":\"wsrglxa3h6ef19ik5i0nbiiys\",\"results\":\"2\"}"
 4) "{\"request_sid\":\"35pfn5au6p8qdbri17p636si\",\"results\":\"10\"}"
 5) "{\"request_sid\":\"1052tgkiyy7c31bmxjtbom7ca\",\"results\":\"5\"}"
 6) "{\"request_sid\":\"1cxnir1slgjiq1o2n3xwznh0kk\",\"results\":\"9\"}"
 7) "{\"request_sid\":\"17jo1xwnnkd3h1mhcqcfplrl5k\",\"results\":\"8\"}"
 8) "{\"request_sid\":\"961vf8hao3stsv4vt1qif3ws\",\"results\":\"7\"}"
 9) "{\"request_sid\":\"1xk521sq6vmmf6enxauwzduj9\",\"results\":\"4\"}"
10) "{\"request_sid\":\"1ca4wy8qsfr7av0hwk8xtlqhp\",\"results\":\"1\"}"