Hbase的Endpoint实现

原文地址：http://www.91baidu.ren/post/baiduren/443.html

protobuf在windows环境下的安装

 

     1、所需包下载地址：http://code.google.com/p/protobuf/downloads/list 。protobuf-x.x.x.tar.gz和protobuf-x.x.x-win32.zip 64位一样。

     2、将protobuf-x.x.x-win32.zip 解压到 System32 文件夹下

     3、解压protobuf-x.x.x-tar.gz到任意目录下，并复制protobuf-x.x.x-win32.zip的解压文件到src目录下

     4、使用dos窗口进入到protobuf-x.x.x-tar.gz的解压路径下的java路径中，使用maven命令： mvn package 编辑该包。

          生成protobuf-java-x.x.x.jar 

     5、使用自带的example测试环境是否安装成功，执行编译命令

               protoc -I=<.proto文件路径>  --java_out=<编译后的文件路径>  xxxx.proto

 

.proto文件简介

          /*全局属性定义，编译后的java类名，包路径*/

option java_package = "com.theta.hbase.coprocessor.protos"; 

option java_outer_classname = "CountProtos"; 

option java_generic_services = true; 

option java_generate_equals_and_hash = true; 

option optimize_for = SPEED; 

/*具体的消息

 *每个message域对应一个内部类，该内部类还包含一个Builder内部类

 *域内字段会生成对应的 setter和getter方法

 *使用Builder内部类来对字段赋值

 **/    

message CountRequest {

       required string startKey = 1;

       required string endKey = 2; 

} 

 

message CountResponse { 

       required int64 count = 1 [default = 0]; 

}

 /*提供服务的类

  *该类没有Builder内部类

  */

service CountService { 

       

       rpc count(CountRequest) 

            returns (CountResponse); 

}  

 

HbaseEndpoint的实现

     使用protobuf的message做为消息传递的格式，使用Rpc做为传输协议，一般会定义三个protobuf域，用于请求、响应、和业务实现。

    1、endpont端要实现protobuf的service内部类，实现service定义的具体方法。使用response返回给调用者。

     例子，实现count操作

        @Override

        public void count(RpcController controller, CountRequest request,

                     RpcCallback<CountResponse> done) {

              

              Scan scan = new Scan();

              scan.setFilter( new FirstKeyOnlyFilter());

              scan.setStartRow(Bytes. toBytes(request.getStartKey()));

              scan.setStopRow(Bytes. toBytes(request.getEndKey()));

              InternalScanner scanner = null;

               try {

                     scanner = env.getRegion().getScanner(scan);

                     List<Cell> results = new ArrayList<Cell>();

                      boolean hasMore = false;

                      byte[] lastRow = null;

                      long count = 0;

                      do {

                           hasMore = scanner.next(results);

                            for (Cell kv : results) {

                                   byte[] currentRow = CellUtil.cloneRow(kv);

                                   if (lastRow == null || !Bytes.equals(lastRow, currentRow)) {

                                         lastRow = currentRow;

                                         count++;

                                  }

                           }

                           results.clear();

                     } while (hasMore);

 

                     response = CountProtos.CountResponse.newBuilder().setCount(count).build();

 

              } catch (IOException ioe) {

                     ResponseConverter. setControllerException(controller, ioe);

              } finally {

                      if (scanner != null) {

                            try {

                                  scanner.close();

                           } catch (IOException ignored) {

                           }

                     }

              }

              done.run(response);

       }

 

2、client端使用HTable实例的 coprocessorService 方法发送Rpc请求,得到response结果

   例子：

             public long count(HTableInterface ht, String startKey, String endKey)

                      throws Throwable {

 

               final CountProtos.CountRequest request = CountProtos.CountRequest

                            . newBuilder().setStartKey(startKey).setEndKey(endKey).build();

              

              Map< byte[], Long> results = ht.coprocessorService(

                           CountProtos.CountService. class, null , null,

                            new Batch.Call<CountProtos.CountService, Long>() {

                                   public Long call(CountProtos.CountService counter)throws IOException {

                                         ServerRpcController controller = new ServerRpcController();

                                         BlockingRpcCallback<CountProtos.CountResponse> rpcCallback = new

                                                  BlockingRpcCallback<CountProtos.CountResponse>();

                                         counter.count(controller, request, rpcCallback);

                                         CountProtos.CountResponse response = rpcCallback.get();

                                          if (controller.failedOnException()) {

                                                 throw controller.getFailedOn();

                                         }

                                          return (response != null && response.getCount() != 0) ? 

                                                                 response.getCount() : 0;

                                  }

                           });

 

              System. out.println(results.values());

 

               return results.size();

       }