redis是一個著名的key-value存儲系統,而作爲其官方推薦的java版客戶端jedis也非常強大和穩定,支持事務、管道及有jedis自身實現的分佈式。
在這裏對jedis關於事務、管道和分佈式的調用方式做一個簡單的介紹和對比:
一、普通同步方式
最簡單和基礎的調用方式,
@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();
}
很簡單吧,每次set之後都可以返回結果,標記是否成功。
二、事務方式(Transactions)
redis的事務很簡單,他主要目的是保障,一個client發起的事務中的命令可以連續的執行,而中間不會插入其他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();
}
我們調用jedis.watch(…)方法來監控key,如果調用後key值發生變化,則整個事務會執行失敗。另外,事務中某個操作失敗,並不會回滾其他操作。這一點需要注意。還有,我們可以使用discard()方法來取消事務。
三、管道(Pipelining)
有時,我們需要採用異步方式,一次發送多個指令,不同步等待其返回結果。這樣可以取得非常好的執行效率。這就是管道,調用方法如下:
@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、管道中調用事務
就Jedis提供的方法而言,是可以做到在管道中使用事務,其代碼如下:
@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
經測試(見本文後續部分),發現其效率和單獨使用事務差不多,甚至還略微差點。
五、分佈式直連同步調用
@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();
}
這個是分佈式直接連接,並且是同步調用,每步執行都返回執行結果。類似地,還有異步管道調用。
六、分佈式直連異步調用
@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、分佈式連接池同步調用
如果,你的分佈式調用代碼是運行在線程中,那麼上面兩個直連調用方式就不合適了,因爲直連方式是非線程安全的,這個時候,你就必須選擇連接池調用。
@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,當然還有異步方式。
八、分佈式連接池異步調用
@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
事務和管道都是異步模式。在事務和管道中不能同步查詢結果。比如下面兩個調用,都是不允許的:
Transaction tx = jedis.multi();
for (int i = 0; i < 100000; i++) {
tx.set("t" + i, "t" + i);
}
System.out.println(tx.get("t1000").get()); //不允許
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()); //不允許
List<Object> results = pipeline.syncAndReturnAll();
事務和管道都是異步的,個人感覺,在管道中再進行事務調用,沒有必要,不如直接進行事務模式。
分佈式中,連接池的性能比直連的性能略好(見後續測試部分)。
分佈式調用中不支持事務。
因爲事務是在服務器端實現,而在分佈式中,每批次的調用對象都可能訪問不同的機器,所以,沒法進行事務。
十、測試
運行上面的代碼,進行測試,其結果如下:
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,經測試分佈式中用到的機器越多,調用會越慢。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
分佈式中,連接池方式調用不但線程安全外,根據上面的測試數據,也可以看出連接池比直連的效率更好。
十一、完整的測試代碼
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,僅作測試
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");
}
}