本文是笔者学习过程中的简单笔记,日后会逐渐增加内容,主要参考资料是《Hbase The Definitive Guide》。
我们可以通过Filter来减少从Server到Client在网络上传输的数据总量,以提升效率。通过HBase的Coprocessor特性,我们甚至可以将计算(computation)移动到数据所在的节点。
Introduction to Coprocessors
coprocessor使你能够直接在每个region server上执行任意的代码。更精确地说,它提供一些通过事件触发的功能,以region为基础执行code;这很像关系型数据库系统中的procedures(存储过程)。
在使用coprocessor时,你需要基于特定的interface创建专门的类,以jar包的形式提供给region server (如:可以将jar包放到$HBASE_HOME/lib/目录下)。这些coprocessor类可以通过配置文件静态加载,也可以在程序代码中动态加载。
corpocessor 框架提供了两种coprocessor基类:
1.Observer
这种coprocessor跟触发器相像:当特定的时间发生时,回调函数就会执行。
RegionObserver
处理数据操纵事件(data manipulationevents),这种coprocessor是和表的region紧密相连的。可以看作DML Coprocessor
MasterObserver
处理数据管理事件,是cluster范围的coprocessor。可以看做DDL Coprocessor
WALObserver
处理 write-ahead log processing 事件
2.Endpoint
The Coprocessor Class
所有的coprocessor类必须实现org.apache.hadoop.hbase.Coprocessor接口。
1.属性
PRIORITY_HIGHEST,PRIORITY_SYSTEM,PRIORITY_USER,PRIORITY_LOWEST四个静态常量表示coprocessor的优先级.值越低优先级越高。
2.方法
start(env) stop(env) :这两个方法在coprocessor开始及退役的时候被调用(these two methods are called when the coprocessor class is started,and eventually when it is decommissioned)
evn参数用来保存coprocessor整个生命周期的状态。
package org.apache.hadoop.hbase;
import java.io.IOException;
/**
* Coprocess interface.
*/
public interface Coprocessor {
static final int VERSION = 1;
/** Highest installation priority */
static final int PRIORITY_HIGHEST = 0;
/** High (system) installation priority */
static final int PRIORITY_SYSTEM = Integer.MAX_VALUE / 4;
/** Default installation priority for user coprocessors */
static final int PRIORITY_USER = Integer.MAX_VALUE / 2;
/** Lowest installation priority */
static final int PRIORITY_LOWEST = Integer.MAX_VALUE;
/**
* Lifecycle state of a given coprocessor instance.
*/
public enum State {
UNINSTALLED,
INSTALLED,
STARTING,
ACTIVE,
STOPPING,
STOPPED
}
// Interface
void start(CoprocessorEnvironment env) throws IOException;
void stop(CoprocessorEnvironment env) throws IOException;
}
Coprocessor Loading 加载coprocessor
静态加载和动态加载。
静态加载:在hbase-site.xml中做类似下面的配置
<property>
<name>hbase.coprocessor.region.classes</name>
<value>coprocessor.RegionObserverExample,coprocessor.AnotherCoprocessor</value>
</property>
<property>
<name>hbase.coprocessor.master.classes</name>
<value>coprocessor.MasterObserverExample</value>
</property>
<property>
<name>hbase.coprocessor.wal.classes</name>
<value>coprocessor.WALObserverExample,bar.foo.MyWALObserver</value>
</property>
public class LoadWithTableDescriptorExample {
public static void main(String[] args) throws IOException
{
Configuration conf = HBaseConfiguration.create();
FileSystem fs = FileSystem.get(conf);
//coprocessor所在的jar包的存放路径
Path path = new Path(fs.getUri() + Path.SEPARATOR +"test/coprocessor/"+
"test.jar");
//HTableDescriptor
HTableDescriptor htd = new HTableDescriptor("testtable");
//addFamily
htd.addFamily(new HColumnDescriptor("colfam1"));
//
//设置要加载的corpocessor
htd.setValue("COPROCESSOR$1", path.toString() +
"|" + RegionObserverExample.class.getCanonicalName() +
"|" + Coprocessor.PRIORITY_USER);
//
HBaseAdmin admin = new HBaseAdmin(conf);
//创建表"testtable"
admin.createTable(htd);
System.out.println("end");
}
}
下面是RegionObserverExample类的实现, 编译通过后,将该类打包成test.jar,并上传到hdfs://master:9000/test/coprocessor目录下
package coprocessor;
import java.io.IOException;
import java.sql.Date;
import java.util.List;
import org.apache.commons.net.ntp.TimeStamp;
import org.apache.hadoop.hbase.KeyValue;
import org.apache.hadoop.hbase.client.Get;
import org.apache.hadoop.hbase.coprocessor.BaseRegionObserver;
import org.apache.hadoop.hbase.coprocessor.ObserverContext;
import org.apache.hadoop.hbase.coprocessor.RegionCoprocessorEnvironment;
import org.apache.hadoop.hbase.util.Bytes;
public class RegionObserverExample extends
BaseRegionObserver {
public static final byte[] FIXED_ROW =
Bytes.toBytes("@@@GETTIME@@@");
//实现功能:用get查询 "@@@GETTIME@@@"行时,以字节数组形式返回系统时间
@Override
public void preGet(
final ObserverContext<RegionCoprocessorEnvironment> e,
final Get get, final List<KeyValue> results) throws
IOException {
if (Bytes.equals(get.getRow(), FIXED_ROW)) {
KeyValue kv = new KeyValue(get.getRow(), FIXED_ROW,
FIXED_ROW,
Bytes.toBytes(System.currentTimeMillis()));
results.add(kv);
}
}
public static void main(String args[]){
System.out.println("complete!");
}
}
Endpoints
前面提到的RegionObserver例子通过已知的row key参数,将列计算功能添加到get请求期间。看起来这足以实现其他功能,比如恩能够返回所有给定列的value的和的聚合函数。然而,RegionObserver并不能实现上述功能,因为row key 决定了由哪个region处理request,这样就只能将计算请求(computation request)发送到单一的server上。
HBase为了克服上述RegionObserver的局限性,由coprocessor框架提供了一个动态调用实现(a dynamic call implementation),称作endpoint concept.
The CoprocessorProtocol interface
The BaseEndpointCoprocessor class
实现一个endpoint包括以下两个步骤
1.Extend the CoprocessorProtocol interface
2.Extend the BaseEndpointCoprocessor class
下面是一个小例子,实现功能:客户端通过远程调用检索每个region的行数和KeyValue的个数。
1.RowCountProtocol interface, code:
public interface RowCountProtocol extends CoprocessorProtocol {
//获取行数
long getRowCount() throws IOException;
//获取应用Filter后的结果集的行数
long getRowCount(Filter filter) throws IOException;
//获取KeyValue的个数
long getKeyValueCount() throws IOException;
}
2.RowCountEndPoint class, code:
public class RowCountEndPoint extends BaseEndpointCoprocessor implements
RowCountProtocol {
public RowCountEndPoint() {
// TODO Auto-generated constructor stub
}
@Override
public long getRowCount() throws IOException {
// TODO Auto-generated method stub
return this.getRowCount(new FirstKeyOnlyFilter());
}
@Override
public long getRowCount(Filter filter) throws IOException {
// TODO Auto-generated method stub
return this.getRowCount(filter,false);
}
@Override
public long getKeyValueCount() throws IOException {
// TODO Auto-generated method stub
return this.getRowCount(null,true);
}
public long getRowCount(Filter filter,boolean countKeyValue) throws IOException {
// TODO Auto-generated method stub
Scan scan =new Scan();
scan.setMaxVersions(1);
if(filter !=null){
scan.setFilter(filter);
}
RegionCoprocessorEnvironment environment=
(RegionCoprocessorEnvironment) this.getEnvironment();
//使用内部scanner做扫描。
InternalScanner scanner = environment.getRegion().getScanner(scan);
//
long result=0;
//计数
try{
boolean done=false;
List<KeyValue> curValue = new ArrayList<KeyValue>();
do{
curValue.clear();
done=scanner.next(curValue);
result+=countKeyValue?curValue.size():1;
}while(done);
}catch(Exception e){
e.printStackTrace();
}finally{
scanner.close();
}
return result;
}
/**
* @param args
*/
public static void main(String[] args) {
// TODO Auto-generated method stub
System.out.println("success!");
}
}
3.
3.1将上述类打包到my_coprocessor.jar, copy到各个RegionServer节点的 $HBASE_HOME/lib目录下;
3.2修改$HBASE_HOME/conf/hbase-site.xml配置文件,添加如下信息:
<property>
<name>hbase.coprocessor.region.classes</name>
<value>
coprocessor.RegionObserverExample,
coprocessor.RowCountEndPoint
</value>
</property>
3.3 重启HBase Cluster
4.通过客户端调用之前定义的EndPoint Coprocessor
public class EndPointExample {
/**
* @author mango_song
* @param args
* @throws IOException
*/
public static void main(String[] args) throws IOException {
// TODO Auto-generated method stub
Configuration conf = HBaseConfiguration.create();
HTable table =new HTable(conf,"test");
try {
//
/*table.coprocessorExec 函数的描述信息:
* <RowCountProtocol, Long> Map<byte[], Long> org.apache.hadoop.hbase.client.HTable.coprocessorExec(
* Class<RowCountProtocol> protocol,
* byte[] startKey, byte[] endKey,
* Call<RowCountProtocol, Long> callable)
* throws IOException, Throwable
Invoke the passed org.apache.hadoop.hbase.client.coprocessor.Batch.Call
against the CoprocessorProtocol instances running in the selected regions.
All regions beginning with the region containing the startKey row,
through to the region containing the endKey row (inclusive) will be used.
If startKey or endKey is null, the first and last regions in the table,
respectively, will be used in the range selection.
Specified by: coprocessorExec(...) in HTableInterface
Parameters:
protocol the CoprocessorProtocol implementation to call
startKey start region selection with region containing this row
endKey select regions up to and including the region containing this row
callable wraps the CoprocessorProtocol implementation method calls made per-region
Returns:
a Map of region names to org.apache.hadoop.hbase.client.coprocessor.Batch.Call.call(Object) return values
Throws:
IOException
Throwable
*/
Map<byte[], Long> results=table.coprocessorExec(
RowCountProtocol.class,
null,
null,
new Batch.Call<RowCountProtocol, Long>() {
@Override
public Long call(RowCountProtocol instance)
throws IOException {
// TODO Auto-generated method stub
return instance.getRowCount();
}
}
);
long total =0;
//打印出每个region的行数及总行数
for(Map.Entry<byte[], Long> entry:results.entrySet() ){
total += entry.getValue();
System.out.println("Region: "+Bytes.toString(entry.getKey()) +
", Count: "+entry.getValue());
}
System.out.println("Total Count: "+total);
} catch (Throwable e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
}
运行结果如下,可以看出test表共由三个region组成,每个region拥有的行数分别为9,13,78
13/01/26 18:59:53 INFO zookeeper.ClientCnxn: Opening socket connection to server master/172.21.15.21:2181. Will not attempt to authenticate using SASL (无法定位登录配置)
13/01/26 18:59:53 INFO zookeeper.ClientCnxn: Socket connection established to master/172.21.15.21:2181, initiating session
13/01/26 18:59:53 INFO zookeeper.ClientCnxn: Session establishment complete on server master/172.21.15.21:2181, sessionid = 0x13c6a82639f000c, negotiated timeout = 40000
Region: test,,1358337586380.f3e04b8b43d073a509e9a374f643277a., Count: 9
Region: test,209,1358337769870.be5a99319eca6f2881ccd73789bfafb0., Count: 13
Region: test,222,1358337769870.94685f417a95e91d0c9185a95974f866., Count: 78
Total Count: 100
Batch类提供了一个更方便的方法来获取远程endpoint, Batch.forMethod() ,你可以得到一个配置好的Batch.Call实例用来传递到远程的region servers. 下面对EndPointExample做了修改,看起来是不是好看多了~~
Batch.Call call =Batch.forMethod(RowCountEndPoint.class, "getKeyValueCount");
Map<byte[], Long> results=table.coprocessorExec(
RowCountProtocol.class,
null,
null,
call
);
然而,通过直接implementing Batch.Call 更加灵活和强大,(you can perform additional processing on the results ,implementing Batch.call directly will provide more power and flexibility.) 下面的例子,同时获取rowCount和keyvalueCount
Map<byte[],Pair<Long,Long>> results=table.coprocessorExec(
RowCountProtocol.class,
null,
null,
new Batch.Call<RowCountProtocol,Pair<Long,Long>>() {
@Override
public Pair<Long, Long> call(RowCountProtocol instance)
throws IOException {
// TODO Auto-generated method stub
return new Pair<Long, Long>(
instance.getRowCount(),
instance.getKeyValueCount()
);
}
}
);
//
long totalRows=0;
long totalKeyValues=0;
for(Map.Entry<byte[], Pair<Long,Long>> entry:results.entrySet() ){
totalRows+=entry.getValue().getFirst();
totalKeyValues+=entry.getValue().getSecond();
System.out.println("region="+Bytes.toString(entry.getKey())+
" , rowCount="+entry.getValue().getFirst()+
" , keyValueCount="+entry.getValue().getSecond());
}
System.out.println("totalRows="+totalRows+
",totalKeyValues="+totalKeyValues);
当然,我们也可以通过coprocessorProxy()方法获取endpoint的client-side 代理,通过该代理,可以在给定的row key所在的region做你想要的操作 (如果row key不存在,则该对应的region为rowkey范围包含该row key的region)。
RowCountProtocol protocol=table.coprocessorProxy(RowCountProtocol.class, Bytes.toBytes("202"));
long rowsInRegion = protocol.getRowCount();
System.out.println("Region Row Count: "+rowsInRegion);