雪花算法

发表于 | 分类于

Twitter-Snowflake算法，为了满足每秒产生上万条不重复id并且保证一定的顺序的产生。

雪花算法生成id结构

|____|__|___|
1 41位 时间戳 10位机器id 12位序列号

最高位不用，加起来64位，可以用一个long表示。

41位时间戳表示毫秒,每毫秒的序列号都从0开始加，12bit序列号最大可以4095

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
//开始生成的时间戳，用当前时间戳-twepoch作为雪花算法的时间戳
private final long twepoch = 1539658610050L;
//服务器id占的位数
private final long serverIdBits = 10L;
//服务器id最大值
private final long maxServerId = 1 << serverIdBits -1;
//序列号占的位数
private final long sequenceBits = 12L;
//服务器id右移位数
private final long serverIdShift = sequenceBits;
//时间戳id右移位数
private final long timestampLeftShift = sequenceBits + serverIdBits;
//最大序列化
private final long maxSequence = 1<< sequenceBits-1;
@Value("${serverId}")
private long serverId;
private long sequence = 0L;
private long lastTimestamp = -1L;

public synchronized Long generatorKey() {
    long timestamp = System.currentTimeMillis();
    if (timestamp < lastTimestamp) {
        throw new RuntimeException(String.format("时间被调整,上次时间戳 %d>%d", lastTimestamp,timestamp));
    }
    if (lastTimestamp == timestamp) {
         sequence=(sequence+1)%maxSequence;
        if (sequence==0) {
            //如果这毫秒的序列号用完，自旋等待下以毫秒再生成
            while ((timestamp=System.currentTimeMillis())<=lastTimestamp){}
        }
    } else {
        sequence=0L;
    }
    lastTimestamp = timestamp;
    return ((timestamp - twepoch) << timestampLeftShift)  | (serverId << serverIdShift) |sequence;
}