hbase热点问题解决（预分区）

淡淡的烟草味﹌ 2023-01-19 09:54 143阅读 0赞

一、出现热点问题原因

1、hbase的中的数据是按照字典序排序的，当大量连续的rowkey集中写在个别的region，各个region之间数据分布不均衡；

2、创建表时没有提前预分区，创建的表默认只有一个region，大量的数据写入当前region；

3、创建表已经提前预分区，但是设计的rowkey没有规律可循，设计的rowkey应该由regionNo+messageId组成。

二、如何解决热点问题

解决这个问题，关键是要设计出可以让数据分布均匀的rowkey，与关系型数据库一样,rowkey是用来检索记录的主键。访问hbase table中的行，rowkey 可以是任意字符串(最大长度 是 64KB，实际应用中长度一般为 10-100bytes)，在hbase内部，rowkey保存为字节数组，存储时，数据按照rowkey的字典序排序存储。

创建表命令：

create 'testTable',\{NAME => 'cf', DATA\_BLOCK\_ENCODING => 'NONE', BLOOMFILTER => 'ROW', REPLICATION\_SCOPE=> '0', VERSIONS => '1', COMPRESSION => 'snappy', MIN\_VERSIONS =>'0', TTL => '15552000', KEEP\_DELETED\_CELLS => 'false', BLOCKSIZE =>'65536', IN\_MEMORY => 'false', BLOCKCACHE => 'true', METADATA =>\{'ENCODE\_ON\_DISK' => 'true'\}\},\{SPLITS\_FILE=>'/app/soft/test/region.txt'\}  
region.txt内容：

我这里预分10个region，执行命令之后，在hbase的console中可以看到以下信息，说明预分区ok了！！！

生成regionNo的工具类：RegionUtils

package com.cn.dl;  
   
import java.util.Random;  
   
/\*\*  
 \* Created by Tiger on 2018/4/18.  
 \*/  
public class RegionUtils \{  
   
    //十个预分区  
    private static final int REGION\_NUM = 10;  
   
    //存放regionNo：0001,0002,...0009,0010  
    private static final String\[\] REGION\_ARRAY = new String\[REGION\_NUM\];  
   
    static \{  
        initRegionArray();  
    \}  
    /\*\*  
     \* 生成regionNo  
     \* \*/  
    private static void initRegionArray()\{  
        for(int i=1; i<=REGION\_NUM; i++)\{  
            String regionNo = String.valueOf(i);  
            while (regionNo.length() < 4)\{  
                regionNo = "0" + regionNo;  
            \}  
            REGION\_ARRAY\[i-1\] = regionNo;  
        \}  
    \}  
    /\*\*  
     \* 随机获取regionNo  
     \* @return  regionNo  
     \* \*/  
    public static String getRegionNo()\{  
        Random random = new Random();  
        return REGION\_ARRAY\[random.nextInt(10)\];  
    \}  
   
    public static void main(String\[\] args) \{  
        int i= 0;  
        while (i < 100)\{  
            System.out.println(getRegionNo());  
            i++;  
        \}  
    \}  
\}  
  生成rowKey：

// TODO: 2018/12/18 只是一个生成rowKey的案例  
    public void execute(Tuple tuple) \{  
        try \{  
            JSONObject json = JSONObject.parseObject(tuple.getStringByField("messageSpout"));  
            String messageId = json.getString("messageId");  
            // TODO: 2018/12/18 生成rowKey:regionNo+时间戳+messageId ，加上时间戳在hbase中可以提高查询效率  
            String rowKey = RegionUtils.getRegionNo() + System.currentTimeMillis() + messageId;  
            json.put("rowKey",rowKey);  
            System.out.println(json.toJSONString());  
        \}catch (Exception e)\{  
            e.printStackTrace();  
        \}finally \{  
            collector.ack(tuple);  
        \}  
    \}  
  打印结果，rowkey=regionNo+时间戳+messageId，前缀是随机的

\{"name":"name2","messageId":"b998a8dfc05a4a819284213d4e727a85","age":12,"rowKey":"00041545105001972b998a8dfc05a4a819284213d4e727a85"\}  
\{"name":"name3","messageId":"799affbf346641e8a00bfeee78ffcdb4","age":13,"rowKey":"00031545105002973799affbf346641e8a00bfeee78ffcdb4"\}  
\{"name":"name4","messageId":"bbdf16b9a12b4fa09b060402f9522fed","age":14,"rowKey":"00051545105003973bbdf16b9a12b4fa09b060402f9522fed"\}  
\{"name":"name5","messageId":"03c119868cd742459464df53c3827147","age":15,"rowKey":"0009154510500497403c119868cd742459464df53c3827147"\}  
\{"name":"name6","messageId":"84c682681cdc4ac09ad3d270741074d3","age":16,"rowKey":"0002154510500597484c682681cdc4ac09ad3d270741074d3"\}  
\{"name":"name7","messageId":"aecbd65f3f434452ab4a924d8e42b947","age":17,"rowKey":"00091545105006976aecbd65f3f434452ab4a924d8e42b947"\}  
\{"name":"name8","messageId":"3bcb23e414e5450898b6b0eefff4d80a","age":18,"rowKey":"000315451050079783bcb23e414e5450898b6b0eefff4d80a"\}  
\{"name":"name9","messageId":"40be62bfcea24ea799ae6f191241c5e8","age":19,"rowKey":"0002154510500897840be62bfcea24ea799ae6f191241c5e8"\}  
\{"name":"name10","messageId":"94c220cd10d141c89cf08e61d0a48e7f","age":20,"rowKey":"0007154510500997894c220cd10d141c89cf08e61d0a48e7f"\}  
\{"name":"name11","messageId":"0796f735b1ba43beb7b15d63c4fd4ec8","age":21,"rowKey":"000515451050109780796f735b1ba43beb7b15d63c4fd4ec8"\}  
\{"name":"name12","messageId":"05ac8417e52443f48bf2c56879b3e2c6","age":22,"rowKey":"0009154510501197805ac8417e52443f48bf2c56879b3e2c6"\}  
\{"name":"name13","messageId":"b87484b633b747ba8320cdf69334459b","age":23,"rowKey":"00101545105012978b87484b633b747ba8320cdf69334459b"\}  
\{"name":"name14","messageId":"84c6daf1cdfd4c0f977a8742ee528977","age":24,"rowKey":"0005154510501397984c6daf1cdfd4c0f977a8742ee528977"\}  
\{"name":"name15","messageId":"8e01e5c53d024de18507ed2a98e38519","age":25,"rowKey":"000415451050149808e01e5c53d024de18507ed2a98e38519"\}  
\{"name":"name16","messageId":"48939394581946e881b91e797659d6ca","age":26,"rowKey":"0005154510501598248939394581946e881b91e797659d6ca"\}  
\{"name":"name17","messageId":"b5c2024c721642a5a2b4cc8682c7cd40","age":27,"rowKey":"00021545105016981b5c2024c721642a5a2b4cc8682c7cd40"\}  
\{"name":"name18","messageId":"4efb10fc351947f6a78761e7b3bf1783","age":28,"rowKey":"000215451050179824efb10fc351947f6a78761e7b3bf1783"\}  
\{"name":"name19","messageId":"a4a27bafd5f749d0b08d9eb832120737","age":29,"rowKey":"00041545105018983a4a27bafd5f749d0b08d9eb832120737"\}  
\{"name":"name20","messageId":"1d90758cbcc2495197b2ef0a05e58610","age":30,"rowKey":"000815451050199831d90758cbcc2495197b2ef0a05e58610"\}  
\{"name":"name21","messageId":"47ae3fb0e3914496b75445df92d0a133","age":31,"rowKey":"0002154510502098347ae3fb0e3914496b75445df92d0a133"\}  
\{"name":"name22","messageId":"419dccbeb2b74484997bd5373f8347af","age":32,"rowKey":"00071545105021982419dccbeb2b74484997bd5373f8347af"\}  
\{"name":"name23","messageId":"51e38da4c9c74542bcd38be704ff3fec","age":33,"rowKey":"0005154510502298351e38da4c9c74542bcd38be704ff3fec"\}  
\{"name":"name24","messageId":"e99e783220d14f63be1f967f82bd69fb","age":34,"rowKey":"00081545105023983e99e783220d14f63be1f967f82bd69fb"\}  
\{"name":"name25","messageId":"0e15f181464146598dfe44976676c706","age":35,"rowKey":"000915451050249840e15f181464146598dfe44976676c706"\}  
\{"name":"name26","messageId":"d0d8c0e939b44a688915714b85d61b1f","age":36,"rowKey":"00101545105025985d0d8c0e939b44a688915714b85d61b1f"\}  
\{"name":"name27","messageId":"4a7d3404871a4137b9db4e24209f1346","age":37,"rowKey":"000515451050269844a7d3404871a4137b9db4e24209f1346"\}  
\{"name":"name28","messageId":"51a5f39f032241fcaa6e5956a9ddb474","age":38,"rowKey":"0002154510502798651a5f39f032241fcaa6e5956a9ddb474"\}  
\{"name":"name29","messageId":"c54ec6f46fa047fdae6187afad55e8f0","age":39,"rowKey":"00011545105028986c54ec6f46fa047fdae6187afad55e8f0"\}  
\{"name":"name30","messageId":"076322780dae4748997006d53aa246c1","age":40,"rowKey":"00021545105029987076322780dae4748997006d53aa246c1"\}  
\{"name":"name31","messageId":"563c258840a84f12ad1a5341381b163c","age":41,"rowKey":"00101545105030986563c258840a84f12ad1a5341381b163c"\}  
\{"name":"name32","messageId":"377f3fbe63634fc4aa93b1737dd462da","age":42,"rowKey":"00051545105031988377f3fbe63634fc4aa93b1737dd462da"\}  
\{"name":"name33","messageId":"5c1af0f26cc94d0b85d0fbca617e57ad","age":43,"rowKey":"000715451050329895c1af0f26cc94d0b85d0fbca617e57ad"\}  
\{"name":"name34","messageId":"2d3050a0b3dd42df97ef87d45dbc0d26","age":44,"rowKey":"000715451050339902d3050a0b3dd42df97ef87d45dbc0d26"\}  
\{"name":"name35","messageId":"526401e1f5ab45aeb95421e0ac200e4b","age":45,"rowKey":"00041545105034990526401e1f5ab45aeb95421e0ac200e4b"\}  
总结：

这种设计的rowkey可以解决热点问题，但是要建立关联表，比如将rowkey保存到数据库或者nosql数据库中，因为前面的regionNo是随机的，不知道 对应数据在hbase的rowkey是多少；同一批数据，因为这个regionNo是随机的，所以要到多个region中get数据，不能使用startkey和endkey去get数据。

2、第二种设计rowkey的方式：通过messageId映射regionNo，这样既可以让数据均匀分布到各个region中，同时可以根据startkey和endkey可以get到同一批数据，messageId映射regionNo，使用一致性hash算法解决，一致性哈希算法在1997年由麻省理工学院的Karger等人在解决分布式Cache中提出的，设计目标是为了解决因特网中的热点(Hot spot)问题，

参考：

（https://baike.baidu.com/item/%E4%B8%80%E8%87%B4%E6%80%A7%E5%93%88%E5%B8%8C/2460889?fr=aladdin）（https://www.cnblogs.com/lpfuture/p/5796398.html）

public class ConsistentHash<T> implements Serializable\{  
    private static final long serialVersionUID = 1L;  
    private  final HashFunction hashFunction;  
    //每个regions的虚拟节点个数  
    private final int numberOfReplicas;  
    //存储虚拟节点的hash值到真实节点的映射  
    private final SortedMap<Long, String> circle = new TreeMap<Long, String>();  
    public ConsistentHash(HashFunction hashFunction, int numberOfReplicas, Collection<String> nodes) \{  
        this.hashFunction = hashFunction;  
        this.numberOfReplicas = numberOfReplicas;  
        for (String node : nodes)\{  
            add(node);  
        \}  
    \}  
    /\*\*  
     \* 添加节点  
     \* @param node  
     \* @see java.util.TreeMap  
     \* \*/  
    public void add(String node) \{  
        for (int i = 0; i < numberOfReplicas; i++)  
             /\*  
              \* 不同的虚拟节点(i不同)有不同的hash值,但都对应同一个实际机器node  
              \* 虚拟node一般是均衡分布在环上的,数据存储在顺时针方向的虚拟node上  
              \*/  
            circle.put(hashFunction.getHashValue(node.toString() + i), node);  
    \}  
    /\*\*  
     \* 移除节点  
     \* @param node  
     \* @see java.util.TreeMap  
     \* \*/  
    public void remove(String node) \{  
        for (int i = 0; i < numberOfReplicas; i++)  
            circle.remove(hashFunction.getHashValue(node.toString() + i));  
    \}  
    /\*\*  
     \* 获取对应key的hashcode值，然后根据hashcode获取当前数据储存的真实节点  
     \* \*/  
    public String get(Object key) \{  
        if (circle.isEmpty())  
            return null;  
        //获取对应key的hashcode值  
        long hash = hashFunction.getHashValue((String) key);  
        //数据映射在两台虚拟机器所在环之间,就需要按顺时针方向寻找机器  
        if (!circle.containsKey(hash)) \{  
            SortedMap<Long, String> tailMap = circle.tailMap(hash);  
            hash = tailMap.isEmpty() ? circle.firstKey() : tailMap.firstKey();  
        \}  
        return circle.get(hash);  
    \}  
    /\*\*  
     \* 获取hash环节点大小  
     \* @return  
     \* \*/  
    public long getSize() \{  
        return circle.size();  
    \}  
    /\*\*  
     \* 获取double类型数据的小数位后四位小数  
     \* @param num  
     \* @return  
     \* \*/  
    public String getDecimalPoint(double num)\{  
        DecimalFormat df = new DecimalFormat("0.0000");  
        return df.format(num);  
    \}  
\}  
public class HashFunction implements Serializable\{  
    private static final long serialVersionUID = 1L;  
    /\*\*  
     \* 获取对应字符串的hashCode值  
     \* @param key  
     \* @return  
     \* \*/  
    public  long getHashValue(String key) \{  
        final int p = 1677761999;  
        int hash = (int) 216613626111L;  
        for (int i = 0; i < key.length(); i++)  
            hash = (hash ^ key.charAt(i)) \* p;  
        hash += hash << 13;  
        hash ^= hash >> 8;  
        hash += hash << 3;  
        hash ^= hash >> 18;  
        hash += hash << 5;  
        // 如果算出来的值为负数则取其绝对值  
        if (hash < 0)  
            hash = Math.abs(hash);  
        return hash;  
    \}  
\}  
     我目前满意第二种方式，然后在es中建立关联表，get数据时，先在es中get到rowkey，然后在hbase中获取数据，这个根据自己的业务设计。  
  
原文链接：https://blog.csdn.net/qq\_31289187/article/details/80869906