高CPU业务场景下的任务分发方案Gearman搭建一览

　　

　   Gearman是当年LiveJournal用来做图片resize的，大家也明白图片resize是一个高CPU的操作，如果让web网站去做这个高CPU的功能，有可能会拖垮你的

web应用，那本篇我们来看看gearman是如何解决这个问题的，它的架构图类似下面这样：

从上面这张图，你应该会看到，Gearman是由三个部分组成：

1. Job Server

    这个就是Gearman的Job Server，通过它对Client 和 jobwork 进行桥接，是不是想起来了中介者模式。。。

2. Client

   Gearman提供了Client API 给客户端调用，Client只需要将一个高CPU的业务函数名丢给Job Server，然后等待JobServer的返回执行结果。

3. jobwork

   Gearman提供了work API 给work客户端进行调用。jobserver会根据后端的work集群的负载情况，分发给一个合适的work去执行，并等待结果。

说到这里，你应该就明白了，本质上它属于那种分布式的RPC调用，而且非常牛逼的地方在于Client 和 Work 可以用不同的语言实现。

一：安装部署

1.  下载地址：https://github.com/gearman/gearmand/releases

         目前gearman的JobServer 有C，JAVA，Perl三种语言实现，由于C版本的JobServer是最活跃的，所以这里采用目前最新的1.1.17版本的gearmand在CentOS

上进行安装部署。

2.  快速安装

    可以通过官网http://gearman.org/getting-started/中的getting-started进行快速安装。

 <1> 基础依赖库安装和gearmand下载

yum -y install boost-devel gperf libevent-devel libuuid-devel gcc44 gcc-c++ 
wget https://github.com/gearman/gearmand/releases/download/1.1.17/gearmand-1.1.17.tar.gz
cd gearmand-1.1.17.tar.gz
tar xzvf gearmand-1.1.17.tar.gz
cd gearmand-1.1.17
[root@localhost gearmand-1.1.17]# ls
aclocal.m4  build-aux     configure.ac  gear_config.in  libgearman-1.0     libhashkit-1.0  Makefile.am  rpm      THANKS
AUTHORS     ChangeLog     COPYING       gearmand        libgearmancore     libhostile      Makefile.in  scripts  util
benchmark   configmake.h  docs          HACKING         libgearman-server  libtest         man          support  version.m4
bin         configure     examples      libgearman      libhashkit         m4              NEWS         tests

<2> 然后就是常规的./configure --prefix=/usr/myapp/gearman && make && make install  这个过程超级慢，可以出去抽跟烟，

        顺便再去拉泡屎。。。

./configure --prefix=/usr/myapp/gearman && make && make install

<3> 若干年后，当你看到这个就算安装成功了。。。还是得恭喜一下。。。。至少没让你踩到缺少各种依赖包的界面。

See any operating system documentation about shared libraries for
more information, such as the ld(1) and ld.so(8) manual pages.
----------------------------------------------------------------------
 /usr/bin/mkdir -p '/usr/myapp/gearman/sbin'
 /usr/bin/install -c -m 644 man/gearman_worker_create.3 man/gearman_worker_define_function.3 man/gearman_worker_echo.3 man/gearman_worker_errno.3 man/gearman_worker_error.3 man/gearman_worker_free.3 man/gearman_worker_function_exist.3 man/gearman_worker_grab_job.3 man/gearman_worker_options.3 man/gearman_worker_register.3 man/gearman_worker_remove_options.3 man/gearman_worker_remove_servers.3 man/gearman_worker_set_context.3 man/gearman_worker_set_log_fn.3 man/gearman_worker_set_namespace.3 man/gearman_worker_set_options.3 man/gearman_worker_set_timeout.3 man/gearman_client_has_option.3 man/gearman_client_options_t.3 man/gearman_task_attr_init.3 man/gearman_task_attr_init_background.3 man/gearman_task_attr_init_epoch.3 man/gearman_task_attr_t.3 man/gearman_worker_set_identifier.3 man/gearman_worker_set_workload_free_fn.3 man/gearman_worker_set_workload_malloc_fn.3 man/gearman_worker_st.3 man/gearman_worker_timeout.3 man/gearman_worker_unregister.3 man/gearman_worker_unregister_all.3 man/gearman_worker_wait.3 man/gearman_worker_work.3 man/libgearman.3 '/usr/myapp/gearman/share/man/man3'
  /bin/sh ./libtool   --mode=install /usr/bin/install -c gearmand/gearmand '/usr/myapp/gearman/sbin'
libtool: install: /usr/bin/install -c gearmand/gearmand /usr/myapp/gearman/sbin/gearmand
 /usr/bin/mkdir -p '/usr/myapp/gearman/bin'
  /bin/sh ./libtool   --mode=install /usr/bin/install -c bin/gearman bin/gearadmin '/usr/myapp/gearman/bin'
libtool: install: /usr/bin/install -c bin/.libs/gearman /usr/myapp/gearman/bin/gearman
libtool: install: /usr/bin/install -c bin/gearadmin /usr/myapp/gearman/bin/gearadmin
make[3]: Leaving directory `/usr/myapp/gearmand-1.1.17'
make[2]: Leaving directory `/usr/myapp/gearmand-1.1.17'
make[1]: Leaving directory `/usr/myapp/gearmand-1.1.17'

<4> 启动gearmand，你也可以用 -d 开启后台运行的模式，这里加上DEBUG只是看一下实时的DEBUG信息，如下所示：

[root@localhost myapp]# cd /usr/myapp/gearman
[root@localhost gearman]# ls
bin  include  lib  sbin  share
[root@localhost gearman]# cd bin
[root@localhost bin]# ls
gearadmin  gearman
[root@localhost bin]# cd /usr/myapp/gearman
[root@localhost gearman]# cd sbin
[root@localhost sbin]# ls
gearmand
[root@localhost sbin]# ./gearmand --verbose DEBUG
./gearmand: Could not open log file "/usr/myapp/gearman/var/log/gearmand.log", from "/usr/myapp/gearman/sbin", switching to stderr. (No such file or directory)
  DEBUG 2017-08-29 02:31:10.796259 [  main ] THREADS: 4 -> libgearman-server/gearmand.cc:263
   INFO 2017-08-29 02:31:10.796374 [  main ] Initializing Gear on port 4730 with SSL: false
   INFO 2017-08-29 02:31:10.796487 [  main ] Starting up with pid 40299, verbose is set to DEBUG
  DEBUG 2017-08-29 02:31:10.796637 [  main ] Method for libevent: epoll -> libgearman-server/gearmand.cc:364
  DEBUG 2017-08-29 02:31:10.798874 [  main ] Trying to listen on 0.0.0.0:4730 -> libgearman-server/gearmand.cc:646
   INFO 2017-08-29 02:31:10.800151 [  main ] Listening on 0.0.0.0:4730 (8)
  DEBUG 2017-08-29 02:31:10.800175 [  main ] Trying to listen on :::4730 -> libgearman-server/gearmand.cc:646
   INFO 2017-08-29 02:31:10.800307 [  main ] Listening on :::4730 (9)
  DEBUG 2017-08-29 02:31:10.800333 [  main ] Creating wakeup pipe -> libgearman-server/gearmand.cc:915
  DEBUG 2017-08-29 02:31:10.800344 [  main ] Creating 4 threads -> libgearman-server/gearmand.cc:378
  DEBUG 2017-08-29 02:31:10.800357 [  main ] Initializing libevent for IO thread -> libgearman-server/gearmand_thread.cc:224
  DEBUG 2017-08-29 02:31:10.800406 [  main ] Creating IO thread wakeup pipe -> libgearman-server/gearmand_thread.cc:495
  DEBUG 2017-08-29 02:31:10.800467 [  main ] Thread 1 created -> libgearman-server/gearmand_thread.cc:273
  DEBUG 2017-08-29 02:31:10.800507 [  main ] Initializing libevent for IO thread -> libgearman-server/gearmand_thread.cc:224
  DEBUG 2017-08-29 02:31:10.800550 [  main ] Creating IO thread wakeup pipe -> libgearman-server/gearmand_thread.cc:495
  DEBUG 2017-08-29 02:31:10.800585 [  main ] Thread 2 created -> libgearman-server/gearmand_thread.cc:273
  DEBUG 2017-08-29 02:31:10.800594 [  main ] Initializing libevent for IO thread -> libgearman-server/gearmand_thread.cc:224
  DEBUG 2017-08-29 02:31:10.800632 [  main ] Creating IO thread wakeup pipe -> libgearman-server/gearmand_thread.cc:495
  DEBUG 2017-08-29 02:31:10.800669 [  main ] Thread 3 created -> libgearman-server/gearmand_thread.cc:273
  DEBUG 2017-08-29 02:31:10.800677 [  main ] Initializing libevent for IO thread -> libgearman-server/gearmand_thread.cc:224
  DEBUG 2017-08-29 02:31:10.800714 [  main ] Creating IO thread wakeup pipe -> libgearman-server/gearmand_thread.cc:495
  DEBUG 2017-08-29 02:31:10.800753 [  main ] Thread 4 created -> libgearman-server/gearmand_thread.cc:273
  DEBUG 2017-08-29 02:31:10.800761 [  main ] replaying queue: begin -> libgearman-server/gearmand.cc:391
  DEBUG 2017-08-29 02:31:10.800766 [  main ] __replay -> libgearman-server/plugins/queue/default/queue.cc:101
  DEBUG 2017-08-29 02:31:10.800774 [  main ] replaying queue: end -> libgearman-server/gearmand.cc:397
   INFO 2017-08-29 02:31:10.800780 [  main ] Adding event for listening socket (8)
   INFO 2017-08-29 02:31:10.800787 [  main ] Adding event for listening socket (9)
  DEBUG 2017-08-29 02:31:10.800794 [  main ] Adding event for wakeup pipe -> libgearman-server/gearmand.cc:966
  DEBUG 2017-08-29 02:31:10.800801 [  main ] Entering main event loop -> libgearman-server/gearmand.cc:406
  DEBUG 2017-08-29 02:31:10.801186 [     2 ] Entering thread event loop -> libgearman-server/gearmand_thread.cc:463
  DEBUG 2017-08-29 02:31:10.801277 [     3 ] Entering thread event loop -> libgearman-server/gearmand_thread.cc:463
  DEBUG 2017-08-29 02:31:10.801507 [  main ] staring up Epoch thread -> libgearman-server/timer.cc:61
  DEBUG 2017-08-29 02:31:10.801635 [     1 ] Entering thread event loop -> libgearman-server/gearmand_thread.cc:463
  DEBUG 2017-08-29 02:31:10.802426 [     4 ] Entering thread event loop -> libgearman-server/gearmand_thread.cc:463

<5> 最后通过netstat，lsof, ps -ef 三板斧可以找出来gearmand大概占用的端口号，就如你看到的默认占用的4370端口，

        当然你也可以在启动的时候用help命令也是能够知道的。

[root@localhost ~]# netstat -tln
Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address           Foreign Address         State      
tcp        0      0 192.168.122.1:53        0.0.0.0:*               LISTEN     
tcp        0      0 0.0.0.0:22              0.0.0.0:*               LISTEN     
tcp        0      0 127.0.0.1:631           0.0.0.0:*               LISTEN     
tcp        0      0 127.0.0.1:25            0.0.0.0:*               LISTEN     
tcp        0      0 0.0.0.0:4730            0.0.0.0:*               LISTEN     
tcp6       0      0 :::8009                 :::*                    LISTEN     
tcp6       0      0 :::8080                 :::*                    LISTEN     
tcp6       0      0 :::22                   :::*                    LISTEN     
tcp6       0      0 ::1:631                 :::*                    LISTEN     
tcp6       0      0 ::1:25                  :::*                    LISTEN     
tcp6       0      0 :::4730                 :::*                    LISTEN     
tcp6       0      0 127.0.0.1:8005          :::*                    LISTEN     
[root@localhost ~]# ps -ef | grep gearmand
root      40299  15869  0 02:31 pts/1    00:00:00 ./gearmand --verbose DEBUG
root      40364  40327  0 02:33 pts/2    00:00:00 grep --color=auto gearmand
[root@localhost ~]# lsof -i :4730
COMMAND    PID USER   FD   TYPE DEVICE SIZE/OFF NODE NAME
gearmand 40299 root    8u  IPv4 322550      0t0  TCP *:gearman (LISTEN)
gearmand 40299 root    9u  IPv6 322551      0t0  TCP *:gearman (LISTEN)
[root@localhost ~]# 

二：Java Driver 在 Gearman上的使用

       为了演示，我可以做一个简单的 “字符串.ToUpper”的业务逻辑来验证一下这个架构是否可以跑的起来。

1. java 充当 Gearman 的 work 

    首先需要在mvn仓库中拉一下jar包：http://www.mvnrepository.com/artifact/org.gearman/gearman-java/0.6。

 

<1> UpperFunction类，这个类用于定义work具体的业务逻辑：

package com.datamip.gearmanwork;

import java.text.SimpleDateFormat;
import java.util.Date;

import org.gearman.client.GearmanJobResult;
import org.gearman.client.GearmanJobResultImpl;
import org.gearman.util.ByteUtils;
import org.gearman.worker.AbstractGearmanFunction;

//字符串大写的业务Function
public class UpperFunction extends AbstractGearmanFunction {

    @Override
    public GearmanJobResult executeFunction() {

        String param = ByteUtils.fromUTF8Bytes((byte[]) this.data);

        byte[] mybytes = param.toUpperCase().getBytes();

        GearmanJobResultImpl result = new GearmanJobResultImpl(mybytes, true, mybytes, null, null, -1, -1);

        SimpleDateFormat formatter = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");

        String dateString = formatter.format(new Date());

        System.out.println(String.format("当前时间:%s, 过来的字符串:%s，返回的字符串:%s", dateString, param,new String(mybytes)));

        return result;
    }
}

<2>  将UpperFunction注册到gearmand中，从红色代码可以看到，其实是一个kv模式，这里的key="myUpperFunc”的对应执行业务就是new UpperFunction。

        这样Client只需要传递一个"myUpperFunc"，Gearmand就知道这个“字符串”对应是哪一个处理函数。。。

public class App {
    public static void main(String[] args) {

        GearmanWorker worker = new GearmanWorkerImpl();

        GearmanNIOJobServerConnection conn = new GearmanNIOJobServerConnection("192.168.23.170", 4730);
        worker.addServer(conn);

        // 将‘将转大写的函数注册’ 到gearmand中 
10         worker.registerFunctionFactory(new GearmanFunctionFactory() {
11 
12             public String getFunctionName() {
13                 return "myUpperFunc";
14             }
15 
16             public GearmanFunction getFunction() {
17                 return new UpperFunction();
18             }
19         });
        
        System.out.println("启动服务。。。。");

        worker.work();
    }
}

2. java 充当 Gearman 的 client

    <1> GearSubmit类【简单的一个包装类，随便定义】

package com.datamip.gearmanclient;

import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;

import org.gearman.client.GearmanClient;
import org.gearman.client.GearmanClientImpl;
import org.gearman.client.GearmanJob;
import org.gearman.client.GearmanJobImpl;
import org.gearman.client.GearmanJobResult;
import org.gearman.common.GearmanJobServerConnection;
import org.gearman.common.GearmanNIOJobServerConnection;
import org.gearman.util.ByteUtils;

public class Gearsubmit {

    public  void process() throws InterruptedException, ExecutionException {
    
        GearmanJobServerConnection conn = new GearmanNIOJobServerConnection("192.168.23.170", 4730);

        GearmanClient client = new GearmanClientImpl();

        client.addJobServer(conn); // 添加连接

        String functionName = "myUpperFunc";

        byte[] data = ByteUtils.toUTF8Bytes("hello,world");

        // 创建后台任务
        GearmanJob job = GearmanJobImpl.createJob(functionName, data, null);

        GearmanJobResult jobResult = null;

        Future<GearmanJobResult> gearmanJobResult = client.submit(job);

        jobResult = gearmanJobResult.get();

        byte[] resultBytes = jobResult.getResults();

        // 获取job的返回值
        String value = ByteUtils.fromUTF8Bytes(resultBytes);

        System.out.println(value);

        System.out.println("执行结束");

        client.shutdown();
    }
}

<2> 主程序，开多线程并发的去执行。

public class App {
    public static void main(String[] args) throws InterruptedException, ExecutionException, IOException {

        ExecutorService executorService = Executors.newFixedThreadPool(100);

        for (int i = 0; i < 10000; i++) {
            executorService.execute(new Runnable() {
                
                @Override
                public void run() {
                    Gearsubmit submit=new Gearsubmit();
                    
                    try {
                        submit.process();
                    } catch (InterruptedException | ExecutionException e) {
                        // TODO Auto-generated catch block
                        e.printStackTrace();
                    }
                }
            });
        }
        
        System.in.read();
    }
}

好了，一切都准备好了，接下来为了演示，演示就是解释，我用Jar2Exe把work程序导出成jar再转换成exe，如下图：

然后我把3.exe开成5个实例，client用100个线程的线程池并发调用，当然一切都是模拟。。。。可以看到，当我client启动的时候，5个work都在执行，

如果这个时候，你把某一个work停止了，jobserver也不再将任务丢给它，而是转给其他负载相对小的work继续执行。

好了，本篇就说到这里，希望对你有帮助。