import java.io.UnsupportedEncodingException;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;
import java.security.MessageDigest;
import java.security.NoSuchAlgorithmException;
import java.util.*;
/**
* Created by IntelliJ IDEA.
* User: test
* Date: 12-5-24
* Time: 下午5:37
* To change this template use File | Settings | File Templates.
*/
public class ConsistencyHash {
private TreeMap<Long,Object> nodes = null;
//真实服务器节点信息
private List<Object> shards = new ArrayList();
//设置虚拟节点数目
private int VIRTUAL_NUM = 4;
/**
* 初始化一致环
*/
public void init() {
shards.add("192.168.0.0-服务器0");
shards.add("192.168.0.1-服务器1");
shards.add("192.168.0.2-服务器2");
shards.add("192.168.0.3-服务器3");
shards.add("192.168.0.4-服务器4");
nodes = new TreeMap<Long,Object>();
for(int i=0; i<shards.size(); i++) {//真实服务器
Object shardInfo = shards.get(i);
for(int j=0; j<VIRTUAL_NUM; j++) {
nodes.put(hash(computeMd5("SHARD-" + i + "-NODE-" + j),j), shardInfo);//每台真实服务器对应四台虚拟服务器
}
}
}
/**
* 根据key的hash值取得服务器节点信息
* @param hash
* @return
*/
public Object getShardInfo(long hash) {
Long key = hash;
SortedMap<Long, Object> tailMap=nodes.tailMap(key);//- treemap中的tailmap是返回大于或等于某个key的子map
if(tailMap.isEmpty()) {
key = nodes.firstKey();
} else {
key = tailMap.firstKey();
}
return nodes.get(key);
}
/**
* 打印圆环节点数据
*/
public void printMap() {
System.out.println(nodes);
}
/**
* 根据2^32把节点分布到圆环上面。
* @param digest
* @param nTime
* @return
*/
public long hash(byte[] digest, int nTime) {
long rv = ((long) (digest[3+nTime*4] & 0xFF) << 24)
| ((long) (digest[2+nTime*4] & 0xFF) << 16)
| ((long) (digest[1+nTime*4] & 0xFF) << 8)
| (digest[0+nTime*4] & 0xFF);
return rv & 0xffffffffL; /* Truncate to 32-bits */
}
/**
* Get the md5 of the given key.
* 计算MD5值
*/
public byte[] computeMd5(String k) {
MessageDigest md5;
try {
md5 = MessageDigest.getInstance("MD5");
} catch (NoSuchAlgorithmException e) {
throw new RuntimeException("MD5 not supported", e);
}
md5.reset();
byte[] keyBytes = null;
try {
keyBytes = k.getBytes("UTF-8");
} catch (UnsupportedEncodingException e) {
throw new RuntimeException("Unknown string :" + k, e);
}
md5.update(keyBytes);
return md5.digest();
}
public static void main(String[] args) {
Random ran = new Random();
ConsistencyHash ch = new ConsistencyHash();
ch.init();
ch.printMap();
//循环50次,是为了取50个数来测试效果,当然也可以用其他任何的数据来测试
for(int i=0; i<50; i++) {
System.out.println(ch.getShardInfo(ch.hash(ch.computeMd5(String.valueOf(i)),ran.nextInt(ch.VIRTUAL_NUM))));
}
}
}
===========================上面的hash生成算法存在问题,对于动态增删节点时================================
这几天看了几遍一致性哈希的文章,但是都没有比较完整的实现,因此试着实现了一下,这里我就不讲一致性哈希的原理了,网上很多,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;
- }
- }
- }