Twitter的分布式自增ID雪花算法snowflake (Java版)

 

概述

分布式系统中,有一些需要使用全局唯一ID的场景,这种时候为了防止ID冲突可以使用36位的UUID,但是UUID有一些缺点,首先他相对比较长,另外UUID一般是无序的。

有些时候我们希望能使用一种简单一些的ID,并且希望ID能够按照时间有序生成。

twitter的snowflake解决了这种需求,最初Twitter把存储系统从MySQL迁移到Cassandra,因为Cassandra没有顺序ID生成机制,所以开发了这样一套全局唯一ID生成服务。

 

结构

snowflake的结构如下(每部分用-分开):

0 - 0000000000 0000000000 0000000000 0000000000 0 - 00000 - 00000 - 000000000000

第一位为未使用,接下来的41位为毫秒级时间(41位的长度可以使用69年),然后是5位datacenterId和5位workerId(10位的长度最多支持部署1024个节点) ,最后12位是毫秒内的计数(12位的计数顺序号支持每个节点每毫秒产生4096个ID序号)

一共加起来刚好64位,为一个Long型。(转换成字符串长度为18)

snowflake生成的ID整体上按照时间自增排序,并且整个分布式系统内不会产生ID碰撞(由datacenter和workerId作区分),并且效率较高。据说:snowflake每秒能够产生26万个ID。

 

  1.  
    /** Snowflake */
  2.  
    public class IdWorker {
  3.  
     
  4.  
    private final long twepoch = 1288834974657L;
  5.  
    private final long workerIdBits = 5L;
  6.  
    private final long datacenterIdBits = 5L;
  7.  
    private final long maxWorkerId = -1L ^ (-1L << workerIdBits);
  8.  
    private final long maxDatacenterId = -1L ^ (-1L << datacenterIdBits);
  9.  
    private final long sequenceBits = 12L;
  10.  
    private final long workerIdShift = sequenceBits;
  11.  
    private final long datacenterIdShift = sequenceBits + workerIdBits;
  12.  
    private final long timestampLeftShift = sequenceBits + workerIdBits + datacenterIdBits;
  13.  
    private final long sequenceMask = -1L ^ (-1L << sequenceBits);
  14.  
     
  15.  
    private long workerId;
  16.  
    private long datacenterId;
  17.  
    private long sequence = 0L;
  18.  
    private long lastTimestamp = -1L;
  19.  
     
  20.  
    public IdWorker(long workerId, long datacenterId) {
  21.  
    if (workerId > maxWorkerId || workerId < 0) {
  22.  
    throw new IllegalArgumentException(String.format("worker Id can't be greater than %d or less than 0", maxWorkerId));
  23.  
    }
  24.  
    if (datacenterId > maxDatacenterId || datacenterId < 0) {
  25.  
    throw new IllegalArgumentException(String.format("datacenter Id can't be greater than %d or less than 0", maxDatacenterId));
  26.  
    }
  27.  
    this.workerId = workerId;
  28.  
    this.datacenterId = datacenterId;
  29.  
    }
  30.  
     
  31.  
    public synchronized long nextId() {
  32.  
    long timestamp = timeGen();
  33.  
    if (timestamp < lastTimestamp) {
  34.  
    throw new RuntimeException(String.format("Clock moved backwards. Refusing to generate id for %d milliseconds", lastTimestamp - timestamp));
  35.  
    }
  36.  
    if (lastTimestamp == timestamp) {
  37.  
    sequence = (sequence + 1) & sequenceMask;
  38.  
    if (sequence == 0) {
  39.  
    timestamp = tilNextMillis(lastTimestamp);
  40.  
    }
  41.  
    } else {
  42.  
    sequence = 0L;
  43.  
    }
  44.  
     
  45.  
    lastTimestamp = timestamp;
  46.  
     
  47.  
    return ((timestamp - twepoch) << timestampLeftShift) | (datacenterId << datacenterIdShift) | (workerId << workerIdShift) | sequence;
  48.  
    }
  49.  
     
  50.  
    protected long tilNextMillis(long lastTimestamp) {
  51.  
    long timestamp = timeGen();
  52.  
    while (timestamp <= lastTimestamp) {
  53.  
    timestamp = timeGen();
  54.  
    }
  55.  
    return timestamp;
  56.  
    }
  57.  
     
  58.  
    protected long timeGen() {
  59.  
    return System.currentTimeMillis();
  60.  
    }
  61.  
     
  62.  
    public static void main(String[] args) {
  63.  
    IdWorker idWorker = new IdWorker(0, 0);
  64.  
    for (int i = 0; i < 1000; i++) {
  65.  
    long id = idWorker.nextId();
  66.  
    System.out.println(id);
  67.  
    }
  68.  
    }
  69.  
    }
posted @ 2019-03-22 14:19  网鱼  阅读(922)  评论(0编辑  收藏  举报