java 读写Parquet格式的数据的示例代码

本文介绍了java 读写Parquet格式的数据，分享给大家，具体如下：

import java.io.BufferedReader;
import java.io.File;
import java.io.FileReader;
import java.io.IOException;
import java.util.Random;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.log4j.Logger;
import org.apache.parquet.example.data.Group;
import org.apache.parquet.example.data.GroupFactory;
import org.apache.parquet.example.data.simple.SimpleGroupFactory;
import org.apache.parquet.hadoop.ParquetReader;
import org.apache.parquet.hadoop.ParquetReader.Builder;
import org.apache.parquet.hadoop.ParquetWriter;
import org.apache.parquet.hadoop.example.GroupReadSupport;
import org.apache.parquet.hadoop.example.GroupWriteSupport;
import org.apache.parquet.schema.MessageType;
import org.apache.parquet.schema.MessageTypeParser;

public class ReadParquet {
 static Logger logger=Logger.getLogger(ReadParquet.class);
 public static void main(String[] args) throws Exception {

//    parquetWriter("test\\parquet-out2","input.txt");
   parquetReaderV2("test\\parquet-out2");
 }

static void parquetReaderV2(String inPath) throws Exception{
   GroupReadSupport readSupport = new GroupReadSupport();
   Builder<Group> reader= ParquetReader.builder(readSupport, new Path(inPath));
   ParquetReader<Group> build=reader.build();
   Group line=null;
   while((line=build.read())!=null){
Group time= line.getGroup("time", 0);
//通过下标和字段名称都可以获取
/*System.out.println(line.getString(0, 0)+"\t"+
line.getString(1, 0)+"\t"+
time.getInteger(0, 0)+"\t"+
time.getString(1, 0)+"\t");*/
System.out.println(line.getString("city", 0)+"\t"+
line.getString("ip", 0)+"\t"+
time.getInteger("ttl", 0)+"\t"+
time.getString("ttl2", 0)+"\t");
//System.out.println(line.toString());
   }
   System.out.println("读取结束");
 }
 //新版本中new ParquetReader()所有构造方法好像都弃用了,用上面的builder去构造对象
 static void parquetReader(String inPath) throws Exception{
   GroupReadSupport readSupport = new GroupReadSupport();
   ParquetReader<Group> reader = new ParquetReader<Group>(new Path(inPath),readSupport);
   Group line=null;
   while((line=reader.read())!=null){
    System.out.println(line.toString());
   }
   System.out.println("读取结束");

}
 /**
  *
  * @param outPath输出Parquet格式
  * @param inPath 输入普通文本文件
  * @throws IOException
  */
 static void parquetWriter(String outPath,String inPath) throws IOException{
   MessageType schema = MessageTypeParser.parseMessageType("message Pair {\n" +
       " required binary city (UTF8);\n" +
       " required binary ip (UTF8);\n" +
       " repeated group time {\n"+
       " required int32 ttl;\n"+
        " required binary ttl2;\n"+
       "}\n"+
      "}");
   GroupFactory factory = new SimpleGroupFactory(schema);
   Path path = new Path(outPath);
   Configuration configuration = new Configuration();
   GroupWriteSupport writeSupport = new GroupWriteSupport();
   writeSupport.setSchema(schema,configuration);
   ParquetWriter<Group> writer = new ParquetWriter<Group>(path,configuration,writeSupport);
//把本地文件读取进去，用来生成parquet格式文件
   BufferedReader br =new BufferedReader(new FileReader(new File(inPath)));
   String line="";
   Random r=new Random();
   while((line=br.readLine())!=null){
     String[] strs=line.split("\\s+");
     if(strs.length==2) {
       Group group = factory.newGroup()
           .append("city",strs[0])
           .append("ip",strs[1]);
       Group tmpG =group.addGroup("time");
       tmpG.append("ttl", r.nextInt(9)+1);
       tmpG.append("ttl2", r.nextInt(9)+"_a");
       writer.write(group);
     }
   }
   System.out.println("write end");
   writer.close();
 }
}

说下schema（写Parquet格式数据需要schema，读取的话"自动识别"了schema）

/*
* 每一个字段有三个属性：重复数、数据类型和字段名，重复数可以是以下三种：
*     required(出现1次)
*     repeated(出现0次或多次)
*     optional(出现0次或1次)
* 每一个字段的数据类型可以分成两种：
*     group(复杂类型)
*     primitive(基本类型)
* 数据类型有
* INT64, INT32, BOOLEAN, BINARY, FLOAT, DOUBLE, INT96, FIXED_LEN_BYTE_ARRAY
*/

这个repeated和required 不光是次数上的区别，序列化后生成的数据类型也不同，比如repeqted修饰 ttl2 打印出来为 WrappedArray([7,7_a]) 而 required修饰 ttl2 打印出来为 [7,7_a]除了用MessageTypeParser.parseMessageType类生成MessageType 还可以用下面方法

（注意这里有个坑--spark里会有这个问题--ttl2这里 as(OriginalType.UTF8) 和 required binary city (UTF8)作用一样，加上UTF8，在读取的时候可以转为StringType，不加的话会报错 [B cannot be cast to java.lang.String  ）

/*MessageType schema = MessageTypeParser.parseMessageType("message Pair {\n" +
       " required binary city (UTF8);\n" +
       " required binary ip (UTF8);\n" +
       "repeated group time {\n"+
       "required int32 ttl;\n"+
       "required binary ttl2;\n"+
       "}\n"+
       "}");*/

//import org.apache.parquet.schema.Types;
MessageType schema = Types.buildMessage()
     .required(PrimitiveTypeName.BINARY).as(OriginalType.UTF8).named("city")
     .required(PrimitiveTypeName.BINARY).as(OriginalType.UTF8).named("ip")
     .repeatedGroup().required(PrimitiveTypeName.INT32).named("ttl")
             .required(PrimitiveTypeName.BINARY).as(OriginalType.UTF8).named("ttl2")
             .named("time")
    .named("Pair");

解决 [B cannot be cast to java.lang.String 异常：

1.要么生成parquet文件的时候加个UTF8
2.要么读取的时候再提供一个同样的schema类指定该字段类型，比如下面:

maven依赖（我用的1.7）

<dependency>
 <groupId>org.apache.parquet</groupId>
 <artifactId>parquet-hadoop</artifactId>
 <version>1.7.0</version>
</dependency>

来源：http://www.cnblogs.com/yanghaolie/p/7156372.html