以一致性哈希用在负载均衡的实例来说,一致性哈希就是先把主机ip从小大到全部放到一个环内,然后客户端ip来连接的时候,把客户端ip连接到大小最接近客户端ip且大于客户端ip的主机。当然,这里的ip一般都是要先hash一下的。我的程序运行结果如下:
-
添加客户端,一开始有4个主机,分别为s1,s2,s3,s4,每个主机有100个虚拟主机:
-
101客户端(hash:-3872430075274208315)连接到主机->s2-192.168.1.2
-
102客户端(hash:-6461488502093916753)连接到主机->s1-192.168.1.1
-
103客户端(hash:-3272337528088901176)连接到主机->s3-192.168.1.3
-
104客户端(hash:7274050343425899995)连接到主机->s2-192.168.1.2
-
105客户端(hash:6218187750346216421)连接到主机->s1-192.168.1.1
-
106客户端(hash:-8497989778066313989)连接到主机->s2-192.168.1.2
-
107客户端(hash:2219601794372203979)连接到主机->s3-192.168.1.3
-
108客户端(hash:1903054837754071260)连接到主机->s3-192.168.1.3
-
109客户端(hash:-2425484502654523425)连接到主机->s1-192.168.1.1
-
删除主机s2-192.168.1.2的变化:
-
hash(-8497989778066313989)改变到->s4-192.168.1.4
-
hash(7274050343425899995)改变到->s2-192.168.1.2
-
hash(-3872430075274208315)改变到->s4-192.168.1.4
-
hash(7274050343425899995)改变到->s1-192.168.1.1
-
增加主机s5-192.168.1.5的变化:
-
hash(1903054837754071260)改变到->s5-192.168.1.5
-
hash(1903054837754071260)改变到->s5-192.168.1.5
-
hash(-3272337528088901176)改变到->s5-192.168.1.5
-
最后的客户端到主机的映射为:
-
hash(-8497989778066313989)连接到主机->s4-192.168.1.4
-
hash(-6461488502093916753)连接到主机->s1-192.168.1.1
-
hash(-3872430075274208315)连接到主机->s4-192.168.1.4
-
hash(-3272337528088901176)连接到主机->s5-192.168.1.5
-
hash(-2425484502654523425)连接到主机->s1-192.168.1.1
-
hash(1903054837754071260)连接到主机->s5-192.168.1.5
-
hash(2219601794372203979)连接到主机->s3-192.168.1.3
-
hash(6218187750346216421)连接到主机->s1-192.168.1.1
-
hash(7274050343425899995)连接到主机->s1-192.168.1.1
看结果可知:一开始添加到9个客户端,连接到主机s1,s2,s3,s4的客户端分别有3,3,3,0个,经过删除主机s2,添加主机s5,最后9个客户端分别连接到主机s1,s2,s3,s4,s5的个数为4,0,1,2,2.这里要说明一下删除主机s2的情况,hash尾号为9995的客户端先连接到s2,再连接到s1,为什么会出现这种情况呢?因为每一个真实主机有n个虚拟主机,删除s2却打印“hash(7274050343425899995)改变到->s2-192.168.1.2”是因为删除了s2的其中一个虚拟主机,跳转到另一个虚拟主机,但还是在s2上,当然,这里是打印中间情况,以便了解,真实的环境是删除了s2后,所有他的虚拟节点都会马上被删除,虚拟节点上的连接也会重新连接到另一个主机的虚拟节点,不会存在这种中间情况。
以下给出所有的实现代码,大家共同学习:
public class Shard<Node> { // S类封装了机器节点的信息 ,如name、password、ip、port等
static private TreeMap<Long, Node> nodes; // 虚拟节点到真实节点的映射
static private TreeMap<Long,Node> treeKey; //key到真实节点的映射
static private List<Node> shards = new ArrayList<Node>(); // 真实机器节点
private final int NODE_NUM = 100; // 每个机器节点关联的虚拟节点个数
boolean flag = false;
public Shard(List<Node> shards) {
super();
this.shards = shards;
init();
}
public static void main(String[] args) {
// System.out.println(hash("w222o1d"));
// System.out.println(Long.MIN_VALUE);
// System.out.println(Long.MAX_VALUE);
Node s1 = new Node("s1", "192.168.1.1");
Node s2 = new Node("s2", "192.168.1.2");
Node s3 = new Node("s3", "192.168.1.3");
Node s4 = new Node("s4", "192.168.1.4");
Node s5 = new Node("s5","192.168.1.5");
shards.add(s1);
shards.add(s2);
shards.add(s3);
shards.add(s4);
Shard<Node> sh = new Shard<Shard.Node>(shards);
System.out.println("添加客户端,一开始有4个主机,分别为s1,s2,s3,s4,每个主机有100个虚拟主机:");
sh.keyToNode("101客户端");
sh.keyToNode("102客户端");
sh.keyToNode("103客户端");
sh.keyToNode("104客户端");
sh.keyToNode("105客户端");
sh.keyToNode("106客户端");
sh.keyToNode("107客户端");
sh.keyToNode("108客户端");
sh.keyToNode("109客户端");
sh.deleteS(s2);
sh.addS(s5);
System.out.println("最后的客户端到主机的映射为:");
printKeyTree();
}
public static void printKeyTree(){
for(Iterator<Long> it = treeKey.keySet().iterator();it.hasNext();){
Long lo = it.next();
System.out.println("hash("+lo+")连接到主机->"+treeKey.get(lo));
}
}
private void init() { // 初始化一致性hash环
nodes = new TreeMap<Long, Node>();
treeKey = new TreeMap<Long, Node>();
for (int i = 0; i != shards.size(); ++i) { // 每个真实机器节点都需要关联虚拟节点
final Node shardInfo = shards.get(i);
for (int n = 0; n < NODE_NUM; n++)
// 一个真实机器节点关联NODE_NUM个虚拟节点
nodes.put(hash("SHARD-" + shardInfo.name + "-NODE-" + n), shardInfo);
}
}
//增加一个主机
private void addS(Node s) {
System.out.println("增加主机"+s+"的变化:");
for (int n = 0; n < NODE_NUM; n++)
addS(hash("SHARD-" + s.name + "-NODE-" + n), s);
}
//添加一个虚拟节点进环形结构,lg为虚拟节点的hash值
public void addS(Long lg,Node s){
SortedMap<Long, Node> tail = nodes.tailMap(lg);
SortedMap<Long,Node> head = nodes.headMap(lg);
Long begin = 0L;
Long end = 0L;
SortedMap<Long, Node> between;
if(head.size()==0){
between = treeKey.tailMap(nodes.lastKey());
flag = true;
}else{
begin = head.lastKey();
between = treeKey.subMap(begin, lg);
flag = false;
}
nodes.put(lg, s);
for(Iterator<Long> it=between.keySet().iterator();it.hasNext();){
Long lo = it.next();
if(flag){
treeKey.put(lo, nodes.get(lg));
System.out.println("hash("+lo+")改变到->"+tail.get(tail.firstKey()));
}else{
treeKey.put(lo, nodes.get(lg));
System.out.println("hash("+lo+")改变到->"+tail.get(tail.firstKey()));
}
}
}
//删除真实节点是s
public void deleteS(Node s){
if(s==null){
return;
}
System.out.println("删除主机"+s+"的变化:");
for(int i=0;i<NODE_NUM;i++){
//定位s节点的第i的虚拟节点的位置
SortedMap<Long, Node> tail = nodes.tailMap(hash("SHARD-" + s.name + "-NODE-" + i));
SortedMap<Long,Node> head = nodes.headMap(hash("SHARD-" + s.name + "-NODE-" + i));
Long begin = 0L;
Long end = 0L;
SortedMap<Long, Node> between;
if(head.size()==0){
between = treeKey.tailMap(nodes.lastKey());
end = tail.firstKey();
tail.remove(tail.firstKey());
nodes.remove(tail.firstKey());//从nodes中删除s节点的第i个虚拟节点
flag = true;
}else{
begin = head.lastKey();
end = tail.firstKey();
tail.remove(tail.firstKey());
between = treeKey.subMap(begin, end);//在s节点的第i个虚拟节点的所有key的集合
flag = false;
}
for(Iterator<Long> it = between.keySet().iterator();it.hasNext();){
Long lo = it.next();
if(flag){
treeKey.put(lo, tail.get(tail.firstKey()));
System.out.println("hash("+lo+")改变到->"+tail.get(tail.firstKey()));
}else{
treeKey.put(lo, tail.get(tail.firstKey()));
System.out.println("hash("+lo+")改变到->"+tail.get(tail.firstKey()));
}
}
}
}
//映射key到真实节点
public void keyToNode(String key){
SortedMap<Long, Node> tail = nodes.tailMap(hash(key)); // 沿环的顺时针找到一个虚拟节点
if (tail.size() == 0) {
return;
}
treeKey.put(hash(key), tail.get(tail.firstKey()));
System.out.println(key+"(hash:"+hash(key)+")连接到主机->"+tail.get(tail.firstKey()));
}
/**
* MurMurHash算法,是非加密HASH算法,性能很高,
* 比传统的CRC32,MD5,SHA-1(这两个算法都是加密HASH算法,复杂度本身就很高,带来的性能上的损害也不可避免)
* 等HASH算法要快很多,而且据说这个算法的碰撞率很低.
* http://murmurhash.googlepages.com/
*/
private static Long hash(String key) {
ByteBuffer buf = ByteBuffer.wrap(key.getBytes());
int seed = 0x1234ABCD;
ByteOrder byteOrder = buf.order();
buf.order(ByteOrder.LITTLE_ENDIAN);
long m = 0xc6a4a7935bd1e995L;
int r = 47;
long h = seed ^ (buf.remaining() * m);
long k;
while (buf.remaining() >= 8) {
k = buf.getLong();
k *= m;
k ^= k >>> r;
k *= m;
h ^= k;
h *= m;
}
if (buf.remaining() > 0) {
ByteBuffer finish = ByteBuffer.allocate(8).order(
ByteOrder.LITTLE_ENDIAN);
// for big-endian version, do this first:
// finish.position(8-buf.remaining());
finish.put(buf).rewind();
h ^= finish.getLong();
h *= m;
}
h ^= h >>> r;
h *= m;
h ^= h >>> r;
buf.order(byteOrder);
return h;
}
static class Node{
String name;
String ip;
public Node(String name,String ip) {
this.name = name;
this.ip = ip;
}
@Override
public String toString() {
return this.name+"-"+this.ip;
}
}
}