hugh 的个人博客

redis是一个著名的key-value存储系统，而作为其官方推荐的java版客户端jedis也非常强大和稳定，支持事务、管道及有jedis自身实现的分布式。 

在这里对jedis关于事务、管道和分布式的调用方式做一个简单的介绍和对比： 
一、普通同步方式 

最简单和基础的调用方式， 

@Test 
public void test1Normal() { 
    Jedis jedis = new Jedis("localhost"); 
    long start = System.currentTimeMillis(); 
    for (int i = 0; i < 100000; i++) { 
        String result = jedis.set("n" + i, "n" + i); 
    } 
    long end = System.currentTimeMillis(); 
    System.out.println("Simple SET: " + ((end - start)/1000.0) + " seconds"); 
    jedis.disconnect(); 
} 

很简单吧，每次set之后都可以返回结果，标记是否成功。 
二、事务方式(Transactions) 

redis的事务很简单，他主要目的是保障，一个client发起的事务中的命令可以连续的执行，而中间不会插入其他client的命令。 

看下面例子： 

@Test 
public void test2Trans() { 
    Jedis jedis = new Jedis("localhost"); 
    long start = System.currentTimeMillis(); 
    Transaction tx = jedis.multi(); 
    for (int i = 0; i < 100000; i++) { 
        tx.set("t" + i, "t" + i); 
    } 
    List<Object> results = tx.exec(); 
    long end = System.currentTimeMillis(); 
    System.out.println("Transaction SET: " + ((end - start)/1000.0) + " seconds"); 
    jedis.disconnect(); 
} 

我们调用jedis.watch(…)方法来监控key，如果调用后key值发生变化，则整个事务会执行失败。另外，事务中某个操作失败，并不会回滚其他操作。这一点需要注意。还有，我们可以使用discard()方法来取消事务。 
三、管道(Pipelining) 

有时，我们需要采用异步方式，一次发送多个指令，不同步等待其返回结果。这样可以取得非常好的执行效率。这就是管道，调用方法如下： 

@Test 
public void test3Pipelined() { 
    Jedis jedis = new Jedis("localhost"); 
    Pipeline pipeline = jedis.pipelined(); 
    long start = System.currentTimeMillis(); 
    for (int i = 0; i < 100000; i++) { 
        pipeline.set("p" + i, "p" + i); 
    } 
    List<Object> results = pipeline.syncAndReturnAll(); 
    long end = System.currentTimeMillis(); 
    System.out.println("Pipelined SET: " + ((end - start)/1000.0) + " seconds"); 
    jedis.disconnect(); 
} 

四、管道中调用事务 

就Jedis提供的方法而言，是可以做到在管道中使用事务，其代码如下： 

@Test 
public void test4combPipelineTrans() { 
    jedis = new Jedis("localhost"); 
    long start = System.currentTimeMillis(); 
    Pipeline pipeline = jedis.pipelined(); 
    pipeline.multi(); 
    for (int i = 0; i < 100000; i++) { 
        pipeline.set("" + i, "" + i); 
    } 
    pipeline.exec(); 
    List<Object> results = pipeline.syncAndReturnAll(); 
    long end = System.currentTimeMillis(); 
    System.out.println("Pipelined transaction: " + ((end - start)/1000.0) + " seconds"); 
    jedis.disconnect(); 
} 

但是经测试（见本文后续部分），发现其效率和单独使用事务差不多，甚至还略微差点。 
五、分布式直连同步调用 

@Test 
public void test5shardNormal() { 
    List<JedisShardInfo> shards = Arrays.asList( 
            new JedisShardInfo("localhost",6379), 
            new JedisShardInfo("localhost",6380)); 

    ShardedJedis sharding = new ShardedJedis(shards); 

    long start = System.currentTimeMillis(); 
    for (int i = 0; i < 100000; i++) { 
        String result = sharding.set("sn" + i, "n" + i); 
    } 
    long end = System.currentTimeMillis(); 
    System.out.println("Simple@Sharing SET: " + ((end - start)/1000.0) + " seconds"); 

    sharding.disconnect(); 
} 

这个是分布式直接连接，并且是同步调用，每步执行都返回执行结果。类似地，还有异步管道调用。 
六、分布式直连异步调用 

@Test 
public void test6shardpipelined() { 
    List<JedisShardInfo> shards = Arrays.asList( 
            new JedisShardInfo("localhost",6379), 
            new JedisShardInfo("localhost",6380)); 

    ShardedJedis sharding = new ShardedJedis(shards); 

    ShardedJedisPipeline pipeline = sharding.pipelined(); 
    long start = System.currentTimeMillis(); 
    for (int i = 0; i < 100000; i++) { 
        pipeline.set("sp" + i, "p" + i); 
    } 
    List<Object> results = pipeline.syncAndReturnAll(); 
    long end = System.currentTimeMillis(); 
    System.out.println("Pipelined@Sharing SET: " + ((end - start)/1000.0) + " seconds"); 

    sharding.disconnect(); 
} 

七、分布式连接池同步调用 （适用于2.2及以下版本）

如果，你的分布式调用代码是运行在线程中，那么上面两个直连调用方式就不合适了，因为直连方式是非线程安全的，这个时候，你就必须选择连接池调用。 

@Test 
public void test7shardSimplePool() { 
    List<JedisShardInfo> shards = Arrays.asList( 
            new JedisShardInfo("localhost",6379), 
            new JedisShardInfo("localhost",6380)); 

    ShardedJedisPool pool = new ShardedJedisPool(new JedisPoolConfig(), shards); 

    ShardedJedis one = pool.getResource(); 

    long start = System.currentTimeMillis(); 
    for (int i = 0; i < 100000; i++) { 
        String result = one.set("spn" + i, "n" + i); 
    } 
    long end = System.currentTimeMillis(); 
    pool.returnResource(one); 
    System.out.println("Simple@Pool SET: " + ((end - start)/1000.0) + " seconds"); 

    pool.destroy(); 
} 

上面是同步方式，当然还有异步方式。 
八、分布式连接池异步调用 （适用于2.2及以下版本）

@Test 
public void test8shardPipelinedPool() { 
    List<JedisShardInfo> shards = Arrays.asList( 
            new JedisShardInfo("localhost",6379), 
            new JedisShardInfo("localhost",6380)); 

    ShardedJedisPool pool = new ShardedJedisPool(new JedisPoolConfig(), shards); 

    ShardedJedis one = pool.getResource(); 

    ShardedJedisPipeline pipeline = one.pipelined(); 

    long start = System.currentTimeMillis(); 
    for (int i = 0; i < 100000; i++) { 
        pipeline.set("sppn" + i, "n" + i); 
    } 
    List<Object> results = pipeline.syncAndReturnAll(); 
    long end = System.currentTimeMillis(); 
    pool.returnResource(one); 
    System.out.println("Pipelined@Pool SET: " + ((end - start)/1000.0) + " seconds"); 
    pool.destroy(); 
} 

九、需要注意的地方 

    事务和管道都是异步模式。在事务和管道中不能同步查询结果。比如下面两个调用，都是不允许的： 

     Transaction tx = jedis.multi(); 
     for (int i = 0; i < 100000; i++) { 
         tx.set("t" + i, "t" + i); 
     } 
     System.out.println(tx.get("t1000").get());  //不允许 

     List<Object> results = tx.exec(); 

     … 
     … 

     Pipeline pipeline = jedis.pipelined(); 
     long start = System.currentTimeMillis(); 
     for (int i = 0; i < 100000; i++) { 
         pipeline.set("p" + i, "p" + i); 
     } 
     System.out.println(pipeline.get("p1000").get()); //不允许 

     List<Object> results = pipeline.syncAndReturnAll(); 

    事务和管道都是异步的，个人感觉，在管道中再进行事务调用，没有必要，不如直接进行事务模式。 

    分布式中，连接池的性能比直连的性能略好(见后续测试部分)。 

    分布式调用中不支持事务。 

    因为事务是在服务器端实现，而在分布式中，每批次的调用对象都可能访问不同的机器，所以，没法进行事务。 

十、测试 

运行上面的代码，进行测试，其结果如下： 

Simple SET: 5.227 seconds 

Transaction SET: 0.5 seconds 
Pipelined SET: 0.353 seconds 
Pipelined transaction: 0.509 seconds 

Simple@Sharing SET: 5.289 seconds 
Pipelined@Sharing SET: 0.348 seconds 

Simple@Pool SET: 5.039 seconds 
Pipelined@Pool SET: 0.401 seconds 

另外，经测试分布式中用到的机器越多，调用会越慢。上面是2片，下面是5片： 

Simple@Sharing SET: 5.494 seconds 
Pipelined@Sharing SET: 0.51 seconds 
Simple@Pool SET: 5.223 seconds 
Pipelined@Pool SET: 0.518 seconds 

下面是10片： 

Simple@Sharing SET: 5.9 seconds 
Pipelined@Sharing SET: 0.794 seconds 
Simple@Pool SET: 5.624 seconds 
Pipelined@Pool SET: 0.762 seconds 

下面是100片： 

Simple@Sharing SET: 14.055 seconds 
Pipelined@Sharing SET: 8.185 seconds 
Simple@Pool SET: 13.29 seconds 
Pipelined@Pool SET: 7.767 seconds 

分布式中，连接池方式调用不但线程安全外，根据上面的测试数据，也可以看出连接池比直连的效率更好。 
十一、完整的测试代码 

package com.example.nosqlclient; 

import java.util.Arrays; 
import java.util.List; 

import org.junit.AfterClass; 
import org.junit.BeforeClass; 
import org.junit.Test; 

import redis.clients.jedis.Jedis; 
import redis.clients.jedis.JedisPoolConfig; 
import redis.clients.jedis.JedisShardInfo; 
import redis.clients.jedis.Pipeline; 
import redis.clients.jedis.ShardedJedis; 
import redis.clients.jedis.ShardedJedisPipeline; 
import redis.clients.jedis.ShardedJedisPool; 
import redis.clients.jedis.Transaction; 

import org.junit.FixMethodOrder; 
import org.junit.runners.MethodSorters; 

@FixMethodOrder(MethodSorters.NAME_ASCENDING) 
public class TestJedis { 

    private static Jedis jedis; 
    private static ShardedJedis sharding; 
    private static ShardedJedisPool pool; 

    @BeforeClass 
    public static void setUpBeforeClass() throws Exception { 
        List<JedisShardInfo> shards = Arrays.asList( 
                new JedisShardInfo("localhost",6379), 
                new JedisShardInfo("localhost",6379)); //使用相同的ip:port,仅作测试 


        jedis = new Jedis("localhost"); 
        sharding = new ShardedJedis(shards); 

        pool = new ShardedJedisPool(new JedisPoolConfig(), shards); 
    } 

    @AfterClass 
    public static void tearDownAfterClass() throws Exception { 
        jedis.disconnect(); 
        sharding.disconnect(); 
        pool.destroy(); 
    } 

    @Test 
    public void test1Normal() { 
        long start = System.currentTimeMillis(); 
        for (int i = 0; i < 100000; i++) { 
            String result = jedis.set("n" + i, "n" + i); 
        } 
        long end = System.currentTimeMillis(); 
        System.out.println("Simple SET: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test2Trans() { 
        long start = System.currentTimeMillis(); 
        Transaction tx = jedis.multi(); 
        for (int i = 0; i < 100000; i++) { 
            tx.set("t" + i, "t" + i); 
        } 
        //System.out.println(tx.get("t1000").get()); 

        List<Object> results = tx.exec(); 
        long end = System.currentTimeMillis(); 
        System.out.println("Transaction SET: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test3Pipelined() { 
        Pipeline pipeline = jedis.pipelined(); 
        long start = System.currentTimeMillis(); 
        for (int i = 0; i < 100000; i++) { 
            pipeline.set("p" + i, "p" + i); 
        } 
        //System.out.println(pipeline.get("p1000").get()); 
        List<Object> results = pipeline.syncAndReturnAll(); 
        long end = System.currentTimeMillis(); 
        System.out.println("Pipelined SET: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test4combPipelineTrans() { 
        long start = System.currentTimeMillis(); 
        Pipeline pipeline = jedis.pipelined(); 
        pipeline.multi(); 
        for (int i = 0; i < 100000; i++) { 
            pipeline.set("" + i, "" + i); 
        } 
        pipeline.exec(); 
        List<Object> results = pipeline.syncAndReturnAll(); 
        long end = System.currentTimeMillis(); 
        System.out.println("Pipelined transaction: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test5shardNormal() { 
        long start = System.currentTimeMillis(); 
        for (int i = 0; i < 100000; i++) { 
            String result = sharding.set("sn" + i, "n" + i); 
        } 
        long end = System.currentTimeMillis(); 
        System.out.println("Simple@Sharing SET: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test6shardpipelined() { 
        ShardedJedisPipeline pipeline = sharding.pipelined(); 
        long start = System.currentTimeMillis(); 
        for (int i = 0; i < 100000; i++) { 
            pipeline.set("sp" + i, "p" + i); 
        } 
        List<Object> results = pipeline.syncAndReturnAll(); 
        long end = System.currentTimeMillis(); 
        System.out.println("Pipelined@Sharing SET: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test7shardSimplePool() { 
        ShardedJedis one = pool.getResource(); 

        long start = System.currentTimeMillis(); 
        for (int i = 0; i < 100000; i++) { 
            String result = one.set("spn" + i, "n" + i); 
        } 
        long end = System.currentTimeMillis(); 
        pool.returnResource(one); 
        System.out.println("Simple@Pool SET: " + ((end - start)/1000.0) + " seconds"); 
    } 

    @Test 
    public void test8shardPipelinedPool() { 
        ShardedJedis one = pool.getResource(); 

        ShardedJedisPipeline pipeline = one.pipelined(); 

        long start = System.currentTimeMillis(); 
        for (int i = 0; i < 100000; i++) { 
            pipeline.set("sppn" + i, "n" + i); 
        } 
        List<Object> results = pipeline.syncAndReturnAll(); 
        long end = System.currentTimeMillis(); 
        pool.returnResource(one); 
        System.out.println("Pipelined@Pool SET: " + ((end - start)/1000.0) + " seconds"); 
    } 
}