Table Level Load Balancer

針對 自己搭建的hbase叢集可以採用如下配置，但是，未做測試：

針對表的region來均衡需要配置：

<property>
<name>hbase.master.loadbalance.bytable</name>
<value>true</value>
</property>

下面程式碼實現了表級別的region分配邏輯，在此給大家分享下。
 

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.hbase.*;
import org.apache.hadoop.hbase.client.*;
import org.apache.hadoop.hbase.util.Bytes;

import java.io.IOException;
import java.util.*;

/**
 *  @author sh04595
 *  Created by sh04595 on 2017/11/29.
 *  優化策略 : 在table級別，對多於平局值部分的region，採用最少region節點分配策略，替代 admin.balancer();
 
 *  最終效果 : 在table級別，每個regionServer的region個數都在平均值上下
 *  balance region on table level
 */
public class HbaseBalancer {
public static final String tableStr = "tablename";public static final String ZK_QUORUM = "zk-srv1:2181";

    public static final Integer BUCKETS_PER_NODE = 50;
    public static void 
 
main(String[] args) throws IOException {

        Configuration config = HBaseConfiguration.create();

        //config.set(HConstants.ZOOKEEPER_CLIENT_PORT, "2181" );
config.set(HConstants.ZOOKEEPER_QUORUM, ZK_QUORUM);
        Connection conn = ConnectionFactory.createConnection(config);

        Admin admin = conn.getAdmin();
        ClusterStatus clusterStatus = admin.getClusterStatus();
        Collection<ServerName> hServers = clusterStatus.getServers();

        System.out.println("region servers :");
        Map<String,RegionServer> allRegionServers = new HashMap<>(15);
        // 根據region server建立 hostname 和regionServer的對映，對allRegionServers進行初始化
for(ServerName server : hServers){

            RegionServer rs = new RegionServer();
            rs.setServerName(server);
            allRegionServers.put( server.getHostname(), rs ) ;

            String getHostAndPort = server.getHostAndPort();
            String getHostname = server.getHostname();

            Long startCode = server.getStartcode();
            System.out.println(startCode +" "+getHostname +" " +getHostAndPort);
            //List<HRegionInfo> regionInfos = admin.getOnlineRegions(server);
allRegionServers .put(server.getHostname(), rs);
        }

        Table table = conn.getTable(TableName.valueOf(tableStr ));
        // 獲取region的位置資訊
RegionLocator locator = conn.getRegionLocator(table.getName());
        List<HRegionLocation> hRegionLocations= locator.getAllRegionLocations();

        int avgCnt = (( int)hRegionLocations.size())/ hServers.size();
        List<HRegionLocation> toAssign = new ArrayList<>(); // 當一個region server 的region的數量大於平均值的時候，儲存需要進行重新分配的region
System.out.println("=============== get Region Location end =============== ");
        // 根據已有的regionLocation資訊進行最大程度的分配到各自節點上
for (HRegionLocation hRegionLocation: hRegionLocations) {
            String hostname =hRegionLocation.getHostname();
            //RegionServer rs = allRegionServers.getOrDefault(hostname , new RegionServer() );
RegionServer rs = allRegionServers.get(hostname); // 上面預先建立的allRegionServers,已經進行初始化，保證這裡不會取空值
if (rs.getRegions().size() == 0) {
                rs.setServerName( hRegionLocation.getServerName());
            }
            if (rs.getRegions().size() < avgCnt) {
                rs.addRegion(hRegionLocation.getRegionInfo().getRegionNameAsString());
            } else {
                toAssign.add(hRegionLocation);
            }
            allRegionServers.putIfAbsent(hostname,rs); // move to rs.add
System.out.println(" one of the" + hRegionLocation.toString());
        }
        System.out.println("=============== get Region Location end =============== ");

        // get all table regions which need to move
        // move to erery serve
System.out.println(" region reassign");
        Iterator<HRegionLocation> assign = toAssign.iterator();
        for (HRegionLocation assignRegion: toAssign) {
            System.out.println("all need to reassign region " + assignRegion.toString());
        }
        System.out.println("=============== region reassign began ===============");

        while (assign.hasNext()){
            HRegionLocation region = assign.next();
            ServerName sn = region.getServerName();

            HRegionInfo regionInfo = region.getRegionInfo();
            String getEncodedName = regionInfo.getEncodedName();
            String sourceHostname = region.getHostname();
            String sourceName = sn.getServerName();

            Random rand = new Random();
            //String destServerKey = allRegionServers.keySet().toArray()[rand .nextInt(toAssign.size())].toString();
String destServerKey = getMinRegionServer(allRegionServers);
            RegionServer rs = allRegionServers.get(destServerKey);
            if (rs.getRegions().size() > avgCnt ){
                // 當所有的regionServer中的region個數大於 平均個數的是停止分配，保證每個節點的region的個數儘可能的平均分配到各個節點上，
                // 不會導致最後每個regionServer的region 個數已經達到平均值，但是某些regionServer的region個數仍然> (avgCnt+ 1)
break;
            }
            System.out.println(" get region toAssign" + region);
            String destServerName = rs.getServerName().getServerName();
            admin.move(regionInfo.getEncodedNameAsBytes(),Bytes.toBytes(destServerName));
            System.out.println(" reassign to " + destServerName);
            rs.addRegion(regionInfo.getRegionNameAsString());
        }

        System.out.println("=============== region reassign end ===============");
    }

    /**
     * 從regionserver中遍歷得到最小的 region server 的hostname
     * @param allRegionServers
* @return region server host name
     */
public static String getMinRegionServer(Map<String,RegionServer> allRegionServers ){
        String key = "";
        Integer cnt = Integer.MAX_VALUE ;
        for (String hostname : allRegionServers.keySet() ) {
            if ( allRegionServers.get(hostname).getRegions().size() < cnt ){
                cnt = allRegionServers.get(hostname).getRegions().size();
                key = hostname;
            }
        }
        return  key;
    }
 }

import org.apache.hadoop.hbase.ServerName;

import java.util.ArrayList;
import java.util.List;

/**
 * Created by sh04595 on 2017/11/29.
 */
public class RegionServer {

    private ServerName serverName;

    private List<String> regions ;

    public RegionServer(){
        super();
        this.regions = new ArrayList<String>();
    }

    public ServerName getServerName() {
        return serverName;
    }

    public void setServerName(ServerName serverName) {
        this.serverName = serverName;
    }


    public List<String> getRegions() {
        return regions;
    }

    public void setRegions(List<String> regions) {
        this.regions = regions;
    }
     public void addRegion ( String region ) {
        this.regions.add(region);
    }

}