rpc框架之 thrift连接池实现

接前一篇rpc框架之HA/负载均衡构架设计 继续，写了一个简单的thrift 连接池：

先做点准备工作：

package yjmyzz;

public class ServerInfo {

    public String getHost() {
        return host;
    }

    public void setHost(String host) {
        this.host = host;
    }

    public int getPort() {
        return port;
    }

    public void setPort(int port) {
        this.port = port;
    }

    private String host;
    private int port;

    public ServerInfo(String host, int port) {
        this.host = host;
        this.port = port;
    }

    public String toString() {
        return "host:" + host + ",port:" + port;
    }
}

上面这个类，用来封装服务端的基本信息，主机名+端口号，连接时需要用到。

package yjmyzz;

import org.apache.thrift.transport.TTransport;

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

public class TransfortWrapper {

    private TTransport transport;

    /**
     * 是否正忙
     */
    private boolean isBusy = false;

    /**
     * 是否已经挂
     */
    private boolean isDead = false;

    /**
     * 最后使用时间
     */
    private Date lastUseTime;

    /**
     * 服务端Server主机名或IP
     */
    private String host;

    /**
     * 服务端Port
     */
    private int port;

    public TransfortWrapper(TTransport transport, String host, int port, boolean isOpen) {
        this.lastUseTime = new Date();
        this.transport = transport;
        this.host = host;
        this.port = port;
        if (isOpen) {
            try {
                transport.open();
            } catch (Exception e) {
                //e.printStackTrace();
                System.err.println(host + ":" + port + " " + e.getMessage());
                isDead = true;
            }
        }
    }

    public TransfortWrapper(TTransport transport, String host, int port) {
        this(transport, host, port, false);
    }


    public boolean isBusy() {
        return isBusy;
    }

    public void setIsBusy(boolean isBusy) {
        this.isBusy = isBusy;
    }

    public boolean isDead() {
        return isDead;
    }

    public void setIsDead(boolean isDead) {
        this.isDead = isDead;
    }

    public TTransport getTransport() {
        return transport;
    }

    public void setTransport(TTransport transport) {
        this.transport = transport;
    }

    /**
     * 当前transport是否可用
     *
     * @return
     */
    public boolean isAvailable() {
        return !isBusy && !isDead && transport.isOpen();
    }

    public Date getLastUseTime() {
        return lastUseTime;
    }

    public void setLastUseTime(Date lastUseTime) {
        this.lastUseTime = lastUseTime;
    }

    public String getHost() {
        return host;
    }

    public int getPort() {
        return port;
    }

    public String toString() {
        SimpleDateFormat format = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");
        return "hashCode:" + hashCode() + "," +
                host + ":" + port + ",isBusy:" + isBusy + ",isDead:" + isDead + ",isOpen:" +
                transport.isOpen() + ",isAvailable:" + isAvailable() + ",lastUseTime:" + format.format(lastUseTime);
    }
}

这是对TTransport的封装，主要增加了一些辅助信息，直接看代码注释即可。

下面才是连接池的主要内容：

package yjmyzz;

import org.apache.thrift.transport.TSocket;
import org.apache.thrift.transport.TTransport;

import java.util.Date;
import java.util.List;
import java.util.concurrent.Semaphore;
import java.util.concurrent.TimeUnit;

/**
 * Thrift连接池
 *
 * @author : 菩提树下的杨过(http://yjmyzz.cnblogs.com/)
 * @version : 0.1 BETA
 * @since : 2015-09-27（中秋）
 */
public class ThriftTransportPool {

    Semaphore access = null;
    TransfortWrapper[] pool = null;
    int poolSize = 1;//连接池大小
    int minSize = 1;//池中保持激活状态的最少连接个数
    int maxIdleSecond = 300;//最大空闲时间（秒），超过该时间的空闲时间的连接将被关闭
    int checkInvervalSecond = 60;//每隔多少秒，检测一次空闲连接（默认60秒）
    List<ServerInfo> serverInfos;
    boolean allowCheck = true;
    Thread checkTread = null;

    public int getCheckInvervalSecond() {
        return checkInvervalSecond;
    }

    public void setCheckInvervalSecond(int checkInvervalSecond) {
        this.checkInvervalSecond = checkInvervalSecond;
    }


    /**
     * 连接池构造函数
     *
     * @param poolSize            连接池大小
     * @param minSize             池中保持激活的最少连接数
     * @param maxIdleSecond       单个连接最大空闲时间，超过此值的连接将被断开
     * @param checkInvervalSecond 每隔多少秒检查一次空闲连接
     * @param serverList          服务器列表
     */
    public ThriftTransportPool(int poolSize, int minSize, int maxIdleSecond, int checkInvervalSecond, List<ServerInfo> serverList) {
        if (poolSize <= 0) {
            poolSize = 1;
        }
        if (minSize > poolSize) {
            minSize = poolSize;
        }
        if (minSize <= 0) {
            minSize = 0;
        }
        this.maxIdleSecond = maxIdleSecond;
        this.minSize = minSize;
        this.poolSize = poolSize;
        this.serverInfos = serverList;
        this.allowCheck = true;
        this.checkInvervalSecond = checkInvervalSecond;
        init();
        check();
    }

    /**
     * 连接池构造函数（默认最大空闲时间300秒）
     *
     * @param poolSize   连接池大小
     * @param minSize    池中保持激活的最少连接数
     * @param serverList 服务器列表
     */
    public ThriftTransportPool(int poolSize, int minSize, List<ServerInfo> serverList) {
        this(poolSize, minSize, 300, 60, serverList);
    }


    public ThriftTransportPool(int poolSize, List<ServerInfo> serverList) {
        this(poolSize, 1, 300, 60, serverList);
    }

    public ThriftTransportPool(List<ServerInfo> serverList) {
        this(serverList.size(), 1, 300, 60, serverList);
    }


    /**
     * 检查空闲连接
     */
    private void check() {
        checkTread =
                new Thread(new Runnable() {
                    public void run() {
                        while (allowCheck) {
                            //System.out.println("--------------");
                            System.out.println("开始检测空闲连接...");
                            for (int i = 0; i < pool.length; i++) {
                                //if (pool[i] == null) {
                                //    System.out.println("pool[" + i + "]为null");
                                //}
                                //if (pool[i].getTransport() == null) {
                                //    System.out.println("pool[" + i + "].getTransport()为null");
                                //}
                                if (pool[i].isAvailable() && pool[i].getLastUseTime() != null) {
                                    long idleTime = new Date().getTime() - pool[i].getLastUseTime().getTime();
                                    //超过空闲阀值的连接，主动断开，以减少资源消耗
                                    if (idleTime > maxIdleSecond * 1000) {
                                        if (getActiveCount() > minSize) {
                                            pool[i].getTransport().close();
                                            pool[i].setIsBusy(false);
                                            System.out.println(pool[i].hashCode() + "," + pool[i].getHost() + ":" + pool[i].getPort() + " 超过空闲时间阀值被断开！");
                                        }
                                    }
                                }
                            }
                            System.out.println("当前活动连接数：" + getActiveCount());
                            try {
                                Thread.sleep(checkInvervalSecond * 1000);
                            } catch (Exception e) {
                                e.printStackTrace();
                            }
                        }
                    }
                });
        checkTread.start();
    }

    /**
     * 连接池初始化
     */
    private void init() {
        access = new Semaphore(poolSize);
        pool = new TransfortWrapper[poolSize];

        for (int i = 0; i < pool.length; i++) {
            int j = i % serverInfos.size();
            TSocket socket = new TSocket(serverInfos.get(j).getHost(),
                    serverInfos.get(j).getPort());
            if (i < minSize) {
                pool[i] = new TransfortWrapper(socket, serverInfos.get(j).getHost(), serverInfos.get(j).getPort(), true);
            } else {
                pool[i] = new TransfortWrapper(socket, serverInfos.get(j).getHost(), serverInfos.get(j).getPort());
            }
        }
    }


    /**
     * 从池中取一个可用连接
     * @return
     */
    public TTransport get() {
        try {
            if (access.tryAcquire(3, TimeUnit.SECONDS)) {
                synchronized (this) {
                    for (int i = 0; i < pool.length; i++) {
                        if (pool[i].isAvailable()) {
                            pool[i].setIsBusy(true);
                            pool[i].setLastUseTime(new Date());
                            return pool[i].getTransport();
                        }
                    }
                    //尝试激活更多连接
                    for (int i = 0; i < pool.length; i++) {
                        if (!pool[i].isBusy() && !pool[i].isDead()
                                && !pool[i].getTransport().isOpen()) {
                            try {
                                pool[i].getTransport().open();
                                pool[i].setIsBusy(true);
                                pool[i].setLastUseTime(new Date());
                                return pool[i].getTransport();
                            } catch (Exception e) {
                                //e.printStackTrace();
                                System.err.println(pool[i].getHost() + ":" + pool[i].getPort() + " " + e.getMessage());
                                pool[i].setIsDead(true);
                            }
                        }
                    }
                }
            }
        } catch (Exception e) {
            e.printStackTrace();
            throw new RuntimeException("can not get available client");

        }
        throw new RuntimeException("all client is too busy");
    }

    /**
     * 客户端调用完成后，必须手动调用此方法，将TTransport恢复为可用状态
     *
     * @param client
     */
    public void release(TTransport client) {
        boolean released = false;
        synchronized (this) {
            for (int i = 0; i < pool.length; i++) {
                if (client == pool[i].getTransport() && pool[i].isBusy()) {
                    pool[i].setIsBusy(false);
                    released = true;
                    break;
                }
            }
        }
        if (released) {
            access.release();
        }
    }


    public void destory() {
        if (pool != null) {
            for (int i = 0; i < pool.length; i++) {
                pool[i].getTransport().close();
            }
        }
        allowCheck = false;
        checkTread = null;
        System.out.print("连接池被销毁！");
    }

    /**
     * 获取当前已经激活的连接数
     *
     * @return
     */
    public int getActiveCount() {
        int result = 0;
        for (int i = 0; i < pool.length; i++) {
            if (!pool[i].isDead() && pool[i].getTransport().isOpen()) {
                result += 1;
            }
        }
        return result;
    }

    /**
     * 获取当前繁忙状态的连接数
     *
     * @return
     */
    public int getBusyCount() {
        int result = 0;
        for (int i = 0; i < pool.length; i++) {
            if (!pool[i].isDead() && pool[i].isBusy()) {
                result += 1;
            }
        }
        return result;
    }

    /**
     * 获取当前已"挂"掉连接数
     *
     * @return
     */
    public int getDeadCount() {
        int result = 0;
        for (int i = 0; i < pool.length; i++) {
            if (pool[i].isDead()) {
                result += 1;
            }
        }
        return result;
    }

    public String toString() {
        return "poolsize:" + pool.length +
                ",minSize:" + minSize +
                ",maxIdleSecond:" + maxIdleSecond +
                ",checkInvervalSecond:" + checkInvervalSecond +
                ",active:" + getActiveCount() +
                ",busy:" + getBusyCount() +
                ",dead:" + getDeadCount();
    }

    public String getWrapperInfo(TTransport client) {
        for (int i = 0; i < pool.length; i++) {
            if (pool[i].getTransport() == client) {
                return pool[i].toString();
            }
        }
        return "";
    }
}

主要思路：

1.构造器里，传入 连接池大小，最小连接数，连接最大空闲时间，空间连接检测时间间隔，服务端列表等基本信息

2.然后调用init方法进行初始化，初始化时把pool[]数组填满，不过在填充的时候，要根据minsize决定激活多少连接（换句话讲，连接实例都都建好了，只是连不连的问题)，另外初始化的时候，还要考虑到某个服务器宕机的可能，如果服务端挂了，将对应的实例设置为isDead=true的状态

3.新开一个线程定时检查是否有空闲连接，如果空闲时间太长，主动断开，以节省开销。

4.get()方法从数组中捞一个可用的连接出来,取的时候要考虑到唤醒"沉睡"连接的情况，即如果当前池中只有2个活动连接，这时又来了请求，没有活动连接了，要从池中把断开的连接叫醒一个。

5.要控制并发控制，多个线程同时调用get()想从池中取可用连接时，可用Semaphore+Lock的机制来加以控制，可参考上一篇内容。

测试：

package yjmyzz;

import org.apache.thrift.transport.TSocket;

import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.Callable;
import java.util.concurrent.FutureTask;


public class PoolTest {

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

        //初始化一个连接池（poolsize=15,minsize=1,maxIdleSecond=5,checkInvervalSecond=10）
        final ThriftTransportPool pool = new ThriftTransportPool(15, 1, 5, 10, getServers());


        //模拟客户端调用
        createClients(pool);

        //等候清理空闲连接
        Thread.sleep(30000);

        //再模拟一批客户端，验证连接是否会重新增加
        createClients(pool);

        System.out.println("输入任意键退出...");
        System.in.read();
        //销毁连接池
        pool.destory();

    }


    private static void createClients(final ThriftTransportPool pool) throws Exception {

        //模拟5个client端
        int clientCount = 5;

        Thread thread[] = new Thread[clientCount];
        FutureTask<String> task[] = new FutureTask[clientCount];

        for (int i = 0; i < clientCount; i++) {
            task[i] = new FutureTask<String>(new Callable<String>() {
                public String call() throws Exception {
                    TSocket scoket = (TSocket) pool.get();//从池中取一个可用连接
                    //模拟调用RPC会持续一段时间
                    System.out.println(Thread.currentThread().getName() + " => " + pool.getWrapperInfo(scoket));
                    try {
                        Thread.sleep(1000);
                    } catch (InterruptedException e) {
                        e.printStackTrace();
                    }
                    pool.release(scoket);//记得每次用完，要将连接释放（恢复可用状态）
                    return Thread.currentThread().getName() + " done.";
                }
            });
            thread[i] = new Thread(task[i], "Thread" + i);
        }

        //启用所有client线程
        for (int i = 0; i < clientCount; i++) {
            thread[i].start();
            Thread.sleep(10);
        }

        System.out.println("--------------");

        //等待所有client调用完成
        for (int i = 0; i < clientCount; i++) {
            System.out.println(task[i].get());
            System.out.println(pool);
            System.out.println("--------------");
            thread[i] = null;
        }
    }

    private static List<ServerInfo> getServers() {
        List<ServerInfo> servers = new ArrayList<ServerInfo>();
        servers.add(new ServerInfo("localhost", 9000));
        servers.add(new ServerInfo("localhost", 9001));
        servers.add(new ServerInfo("localhost", 1002));//这一个故意写错的，模拟服务器挂了，连接不上的情景
        return servers;
    }
}

输出：

****************************
开始检测空闲连接...
当前活动连接数：1
Thread1 => hashCode:919192718,localhost:9001,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:09:59
Thread0 => hashCode:1510835162,localhost:9000,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:09:59
localhost:1002 java.net.ConnectException: Connection refused
Thread2 => hashCode:1466719669,localhost:9000,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:00
Thread3 => hashCode:2080503518,localhost:9001,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:00
localhost:1002 java.net.ConnectException: Connection refused
Thread4 => hashCode:411724643,localhost:9000,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:00
--------------
Thread0 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:3,dead:2
--------------
Thread1 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:3,dead:2
--------------
Thread2 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:2,dead:2
--------------
Thread3 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:1,dead:2
--------------
Thread4 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:0,dead:2
--------------
开始检测空闲连接...
1510835162,localhost:9000 超过空闲时间阀值被断开！
919192718,localhost:9001 超过空闲时间阀值被断开！
1466719669,localhost:9000 超过空闲时间阀值被断开！
2080503518,localhost:9001 超过空闲时间阀值被断开！
当前活动连接数：1
开始检测空闲连接...
当前活动连接数：1
开始检测空闲连接...
当前活动连接数：1
Thread0 => hashCode:411724643,localhost:9000,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:31
Thread1 => hashCode:1510835162,localhost:9000,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:31
Thread2 => hashCode:919192718,localhost:9001,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:31
Thread3 => hashCode:1466719669,localhost:9000,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:31
Thread4 => hashCode:2080503518,localhost:9001,isBusy:true,isDead:false,isOpen:true,isAvailable:false,lastUseTime:2015-09-27 16:10:31
--------------
Thread0 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:4,dead:2
--------------
Thread1 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:3,dead:2
--------------
Thread2 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:2,dead:2
--------------
Thread3 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:1,dead:2
--------------
Thread4 done.
poolsize:15,minSize:1,maxIdleSecond:5,checkInvervalSecond:10,active:5,busy:0,dead:2
--------------
输入任意键退出...
q
连接池被销毁！

***********************

注意上面高亮颜色的部分，2080503518 连接创建后，后来被check方法主动检测到空闲断开，然后第二轮调用时，又重新激活。411724643 则幸免于难，一直战斗到最后。另外由于故意写错了一个server地址，池中始终有二个dead的实例。

值得改进的地方：

主要是公平性的问题，在初始化的时候，如果服务器有3台，而指定的连接池大小为4，目前的做法是，用4对3取模，所以第1、4个连接实例都是连接到服务器1，get取可用连接的时候也有类似情况，是按pool数组从前向后遍历的，捞到第1个可用的连接就完事了，这样永远是排在List前面的服务器压力会大一些，这样有点不太符合负载"均衡"的本意。

不过，这个问题也很好解决，有一个很简单有效的技巧，实际应用中，服务器列表是从zk上取回来的，取回来后，先对数组做随机排序，这样整体看来下，多个连接池总体的连接分布情况就比较平均了。