全局唯一订单号生成方法(参考snowflake)

backgroud

Snowflake is a network service for generating unique ID numbers at high scale with some simple guarantees.

简介

对于一个较大的订购业务场景,我们往往需要能够生成一个全局的唯一的订单号,如何在多个集群,多个节点高效生成唯一订单号?我们参考了Twitter的snowflake算法。

snowflake最初由Twitter开发,用的scala,对于Twitter而言,必须满足每秒上万条消息的请求,并且每条消息能够分配一个全局唯一的ID,因此,ID生成服务要求必须满足高性能(>10K ids/s)、低延迟(<2ms)、高可用的特性,同时生成的ID还可以进行大致的排序,以方便客户端的排序。

Snowflake满足了以上的需求。Snowflake生成的每一个ID都是64位的整型数,它的核心算法也比较简单高效,结构如下:

  • 41位的时间序列,精确到毫秒级,41位的长度可以使用69年。时间位还有一个很重要的作用是可以根据时间进行排序。

  • 10位的机器标识,10位的长度最多支持部署1024个节点。

  • 12位的计数序列号,序列号即一系列的自增id,可以支持同一节点同一毫秒生成多个ID序号,12位的计数序列号支持每个节点每毫秒产生4096个ID序号。

  • 最高位是符号位,始终为0,不可用。

原生算法java实现


/** 
* 摘自网上某blog,记不得地址了。。 
* @Project concurrency 
* Created by wgy on 16/7/19. 
*/ 
public class IdGen { 
private long workerId; 
private long datacenterId; 
private long sequence = 0L; 
private long twepoch = 1288834974657L; //Thu, 04 Nov 2010 01:42:54 GMT 
private long workerIdBits = 5L; //节点ID长度 
private long datacenterIdBits = 5L; //数据中心ID长度 
private long maxWorkerId = -1L ^ (-1L << workerIdBits); //最大支持机器节点数0~31,一共32个 
private long maxDatacenterId = -1L ^ (-1L << datacenterIdBits); //最大支持数据中心节点数0~31,一共32个 
private long sequenceBits = 12L; //序列号12位 
private long workerIdShift = sequenceBits; //机器节点左移12位 
private long datacenterIdShift = sequenceBits + workerIdBits; //数据中心节点左移17位 
private long timestampLeftShift = sequenceBits + workerIdBits + datacenterIdBits; //时间毫秒数左移22位 
private long sequenceMask = -1L ^ (-1L << sequenceBits); //4095 
private long lastTimestamp = -1L; 
private static class IdGenHolder { 
private static final IdGen instance = new IdGen(); 

public static IdGen get(){ 
return IdGenHolder.instance; 

public IdGen() { 
this(0L, 0L); 

public IdGen(long workerId, long datacenterId) { 
if (workerId > maxWorkerId || workerId < 0) { 
throw new IllegalArgumentException(String.format("worker Id can't be greater than %d or less than 0", maxWorkerId)); 

if (datacenterId > maxDatacenterId || datacenterId < 0) { 
throw new IllegalArgumentException(String.format("datacenter Id can't be greater than %d or less than 0", maxDatacenterId)); 

this.workerId = workerId; 
this.datacenterId = datacenterId; 

public synchronized long nextId() { 
long timestamp = timeGen(); //获取当前毫秒数 
//如果服务器时间有问题(时钟后退) 报错。 
if (timestamp < lastTimestamp) { 
throw new RuntimeException(String.format( 
"Clock moved backwards. Refusing to generate id for %d milliseconds", lastTimestamp - timestamp)); 

//如果上次生成时间和当前时间相同,在同一毫秒内 
if (lastTimestamp == timestamp) { 
//sequence自增,因为sequence只有12bit,所以和sequenceMask相与一下,去掉高位 
sequence = (sequence + 1) & sequenceMask; 
//判断是否溢出,也就是每毫秒内超过4095,当为4096时,与sequenceMask相与,sequence就等于0 
if (sequence == 0) { 
timestamp = tilNextMillis(lastTimestamp); //自旋等待到下一毫秒 

} else { 
sequence = 0L; //如果和上次生成时间不同,重置sequence,就是下一毫秒开始,sequence计数重新从0开始累加 

lastTimestamp = timestamp; 
// 最后按照规则拼出ID。 
// 000000000000000000000000000000000000000000 00000 00000 000000000000 
// time datacenterId workerId sequence 
return ((timestamp - twepoch) << timestampLeftShift) | (datacenterId << datacenterIdShift) 
| (workerId << workerIdShift) | sequence; 

protected long tilNextMillis(long lastTimestamp) { 
long timestamp = timeGen(); 
while (timestamp <= lastTimestamp) { 
timestamp = timeGen(); 

return timestamp; 

protected long timeGen() { 
return System.currentTimeMillis(); 


注释已经写的比较详细了,不做特别的说明。

订购业务唯一订单号实现

对于订购业务而言,虽然可以记录订单的创建时间,但是一般都需要带有显示的时间戳属性。因此,一个long型已无法满足实际的需求,将输出修改为String类型,前17位用于存储yyyyMMddHHMMssSSS格式的时间,后面用于记录所在集群,节点,以及自增量。


import org.apache.commons.lang.time.DateFormatUtils;

import java.net.InetAddress; 
import java.net.UnknownHostException; 
import java.util.Date;

/** 
* 与snowflake算法区别,返回字符串id,占用更多字节,但直观从id中看出生成时间 

* @Project concurrency 
* Created by wgy on 16/7/19. 
*/ 
public enum IdGenerator {

INSTANCE;

private long workerId;   //用ip地址最后几个字节标示
private long datacenterId = 0L; //可配置在properties中,启动时加载,此处默认先写成0
private long sequence = 0L;
private long workerIdBits = 8L; //节点ID长度
private long datacenterIdBits = 2L; //数据中心ID长度,可根据时间情况设定位数
private long sequenceBits = 12L; //序列号12位
private long workerIdShift = sequenceBits; //机器节点左移12位
private long datacenterIdShift = sequenceBits + workerIdBits; //数据中心节点左移14位
private long sequenceMask = -1L ^ (-1L << sequenceBits); //4095
private long lastTimestamp = -1L;

IdGenerator(){
    workerId = 0x000000FF & getLastIP();
}


public synchronized String nextId() {
    long timestamp = timeGen(); //获取当前毫秒数
    //如果服务器时间有问题(时钟后退) 报错。
    if (timestamp < lastTimestamp) {
        throw new RuntimeException(String.format(
                "Clock moved backwards.  Refusing to generate id for %d milliseconds", lastTimestamp - timestamp));
    }
    //如果上次生成时间和当前时间相同,在同一毫秒内
    if (lastTimestamp == timestamp) {
        //sequence自增,因为sequence只有12bit,所以和sequenceMask相与一下,去掉高位
        sequence = (sequence + 1) & sequenceMask;
        //判断是否溢出,也就是每毫秒内超过4095,当为4096时,与sequenceMask相与,sequence就等于0
        if (sequence == 0) {
            timestamp = tilNextMillis(lastTimestamp); //自旋等待到下一毫秒
        }
    } else {
        sequence = 0L; //如果和上次生成时间不同,重置sequence,就是下一毫秒开始,sequence计数重新从0开始累加
    }
    lastTimestamp = timestamp;


    long suffix = (datacenterId << datacenterIdShift) | (workerId << workerIdShift) | sequence;

    String datePrefix = DateFormatUtils.format(timestamp, "yyyyMMddHHMMssSSS");

    return datePrefix + suffix;
}

protected long tilNextMillis(long lastTimestamp) {
    long timestamp = timeGen();
    while (timestamp <= lastTimestamp) {
        timestamp = timeGen();
    }
    return timestamp;
}

protected long timeGen() {
    return System.currentTimeMillis();
}

private byte getLastIP(){
    byte lastip = 0;
    try{
        InetAddress ip = InetAddress.getLocalHost();
        byte[] ipByte = ip.getAddress();
        lastip = ipByte[ipByte.length - 1];
    } catch (UnknownHostException e) {
        e.printStackTrace();
    }
    return lastip;
}

}

 

测试

测试环境

  • macbook Pro 2.4 GHz Intel Core i5 4 GB 1600 MHz DDR3
  • 10个线程,每个线程生成5w个

    需2000ms左右,测试代码如下:

测试代码


@Test 
public void testNextId() throws Exception { 
final IdGenerator idg = IdGenerator.INSTANCE; 
ExecutorService es = Executors.newFixedThreadPool(10); 
final HashSet idSet = new HashSet(); 
Collections.synchronizedCollection(idSet); 
long start = System.currentTimeMillis(); 
System.out.println(" start generate id *"); 
for (int i = 0; i < 10; i++) 
es.execute(new Runnable() { 
public void run() { 
for (int j = 0; j < 50000; j++) { 
String id= idg.nextId(); 
synchronized (idSet){ 
idSet.add(id); 



}); 
es.shutdown(); 
es.awaitTermination(10, TimeUnit.SECONDS); 
long end = System.currentTimeMillis(); 
System.out.println(" end generate id "); 
System.out.println("* cost " + (end-start) + " ms!"); 
Assert.assertEquals(10 * 50000, idSet.size()); 

测试结果

start generate id * 
end generate id * 
* cost 2091 ms!

posted @ 2016-09-20 13:40  费曼带我飞  阅读(6703)  评论(0编辑  收藏  举报