这篇文章主要讲解了“Hadoop序列化怎么实现”,文中的讲解内容简单清晰,易于学习与理解,下面请大家跟着小编的思路慢慢深入,一起来研究和学习“Hadoop序列化怎么实现”吧!
Hadoop I/O
Data Integrity
Hdfs: % hadoop fs -cat hdfs://namenode/data/a.txt
LocalFS: % hadoop fs -cat file:///tmp/a.txt
generate crc check sum file
%hadoop fs -copyToLocal -crc /data/a.txt file:///data/a.txt
check sum file: .a.txt.crc is a hidden file.
Ref: CRC-32,循环冗余校验算法,error-detecting.
io.bytes.per.checksum is deprecated, it's dfs.bytes-per-checksum, default is 512, Must not be larger than dfs.stream-buffer-size,which is the size of buffer to stream files. The size of this buffer should probably be a multiple of hardware page size (4096 on Intel x86), and it determines how much data is buffered during read and write operations.
Data Compression
常用算法
读书时,hadoop支持四种压缩算法,如果调解空间和效率的话,-1 ~ -9,代表从最优速度到最优空间. 压缩算法支持在org.apache.hadoop.io.compress.*.
deflate (.deflate), 就是常用的gzip, package ..DefaultCodec
Gzip (.gz),在deflate格式加了文件头和尾. 压缩速度(适中),解压速度(适中),压缩效率(适中),package ..GzipCodec, both of java and native
bzip2 (.bz2), 压缩速度(最差),< 解压速度(最差),压缩效率 (最好),特点是支持可切分(splitable),对map-red非常友好。,package ..BZip2Codec,java only
LZO (.lzo), 压缩速度(最快),解压速度(最快),压缩效率(最差),,package com.hadoop.compressiojn.lzo.lzopCodec, native only
如果禁用原生库,使用hadoop.native.lib.
如果使用原生库,可能对象创建的成本较高,所以可以使用CodecPool,重复使用这些对象。
对于一个非常大的数据文件,存储如下方案:
使用支持切分的bzip2
手动切分,并使压缩后的part接近于block size.
使用Sequence File, 它支持压缩和切分
使用Avro数据文件,它也支持压缩和切分,而且增加了很多编程语言的可读写性。
如果Map-Red的output自动压缩:
conf.setBoolean ("mared.output.compress",true);
conf.setClass("mapred.output.compression.codec",GzipCodec.class,CompressionCodec.class);
如果Map-Red的中间结果的自动压缩:
//or conf.setCompressMapOutput(true);
conf.setBoolean ("mared.compress.map.output",true);
//or conf.setMapOutputComressorClass(GzipCodec.class)
conf.setClass("mapred.map.output.compression.codec",GzipCodec.class,CompressionCodec.class);
序列化(Serialization/Deserialization)
Writable and WritableComparable
// core class for hadoop
public interface Writable{
void write(DataOutput out) throw IOException;
void readFields(DataInput in) throw IOException;
}
public interface Comparable<T>{
int compareTo(T o);
}
//core class for map-reduce shuffle
public interface WritableComparable<T> extends Writable, Comparable<T> {
}
// Sample
public class MyWritableComparable implements WritableComparable {
// Some data
private int counter;
private long timestamp;
public void write(DataOutput out) throws IOException {
out.writeInt(counter);
out.writeLong(timestamp);
}
public void readFields(DataInput in) throws IOException {
counter = in.readInt();
timestamp = in.readLong();
}
public int compareTo(MyWritableComparable o) {
int thisValue = this.value;
int thatValue = o.value;
return (thisValue < thatValue ? -1 : (thisValue==thatValue ? 0 : 1));
}
public int hashCode() {
final int prime = 31;
int result = 1;
result = prime * result + counter;
result = prime * result + (int) (timestamp ^ (timestamp >>> 32));
return result
}
}
//optimize for stream comparasion
public interface RawComparator<T> extends Comparator<T>{
// s1 start position, l1, length of bytes
public int compare(byte[] b1, int s1,int l1,byte[] b2,int s2,int l2);
}
public class WritableComparator implements RawComparator{
}
Comparator RawComparator WritableComparator
WritableComparator 提供了原始compator的compare反序列化对象的实现,性能较差。不过它作为RawComparator实例的工厂:
RawComparator<IntWritable> comparator = WritableComparator.get(IntWritable.class);
// 注册一个经过优化的比较算子。Register an optimized comparator for a WritableComparable implementation.
static void define(Class c, WritableComparator comparator);
// 获得一个WritableComparable的比较算子. Get a comparator for a WritableComparable implementation.
static WritableComparator get(Class<? extends WritableComparable> c);
public MyWritableComparator extends WritableComparator{
static{
define(MyWritableComparable.class, new MyWritableComparator());
}
public MyWritableComparator {
super(MyWritableComparable.class);
}
@Override
public int compare(byte[] b1, int s1,int l1,byte[] b2,int s2,int l2){
}
}
注: 要使static initializer被调用,除非有该类的实例被创建,或某静态方法或成员被访问。或者直接强制,代码如:
Class.forName("package.yourclass"); 它会强制初始化静态initializer.
Java Primitive Data Type wrapped by Writable
Extends from WritableComparable
Extends from Writable only
ArrayWritable
TwoDArrayWritable
AbstractMapWritable
MapWritable
SortedMapWritable
[Text]
值得一提的是Text的序列化方式是Zero-compressed encoding,这个看过一些资料,其实是一种编码方式,意图是省略掉高位0所占用的空间,对于小数,它能节省空间,对于大数会额外占用空间。相比压缩,它能比较快速。其实类似于VIntWritable, VLongWritable的编码方式。
- 如何选择变长和定长数值呢?
1. 定长适合分布非常均匀的数值(如hash),变长适合分布非常不均匀的数值。
2. 变长可以节省空间,而且可以在VIntWritable 和VLongWritable之间转换。
- Text和String的区别
1。String是char序列,Text是UTF-8的byte序列.
UTF-8类不能对字符串大于32767的进行utf-8编码。
(Indexing)索引:对于ASCII来说, Text和String是一样的, 对于Unicode就不同了。String类的长度是其所含char编码单元的长度,然而Text是UTF-8的字节码的长度。CodePointAt表示一个真正的Unicode字符,它可以是2char,4bytes的unicode。
Iteration(迭代): 将Text转换ByteBuffer,然后反复调用bytesToCodePoint()静态方法,可以取到整型的Unicode.
Mutable(易变性): 可以set,类似writable 和StringBuffer,getLength()返回有效字串长度,getbytes().length,返回空间大小。
[BytesWritable]
这是二进制数组的封装,类似于windows下的BSTR,都是前面一个整型表示字节长度,后面是字节的二进制流。
它也是mutable,getLength() != getBytes().length
[NullWritable]
NullWritable是Writable的一个特殊类型。它的序列化长度为0,其实只是一个占位符,既不读入,也不写出。只是存在于程序体中。
Immutable,是一个singleton。
[ObjectWritable]
ObjectWritable是Java的Array, String, 以及Primitive类型的通用封装 (注:不包含Integer)。它的序列化则使用java的类型序列化,写入类型信息等,比较占用空间。
通过两个特殊的构造:
public ObjectWritable(Object instance);
public ObjectWritable(Class declaredClass,Object instance);
举例子:
ObjectWritable objectw = new ObjectWritable(int.class,5);
[GenericWritable]
首先这是一个抽象类,需要被具象化才能使用。
观察下面这个实列,它以一种Union方式,显示的代理一个Writable实例,解决了Reduce函数的参数声明问题。
public class MyGenericWritable extends GenericWritable {
private static Class<? extends Writable>[] CLASSES = null;
static {
CLASSES = (Class<? extends Writable>[]) new Class[] {
IntWritable.class,
Text.class
//add as many different Writable class as you want
};
}
@Override
protected Class<? extends Writable>[] getTypes() {
return CLASSES;
}
@Override
public String toString() {
return "MyGenericWritable [getTypes()=" + Arrays.toString(getTypes()) + "]";
}
// override hashcode();
}
public class Reduce extends Reducer<Text, MyGenericWritable, Text, Text> {
public void reduce(Text key, Iterable<MyGenericWritable> values, Context context) throws IOException, InterruptedException {
}
[ArrayWritable /TwoDArrayWritable]
ArrayWritable aw = new ArrayWriable(Text.class);
[MapWritable / SortedMapWritable]
实现了java.util.Map<Writable,Writable> 和SortedMap...
它的serialize, 使用先写map<classname,id>,然后后边每个类的类型,以id来替代,节省空间。这些都在父类AbstractMapWritable中实现。
集合小结:
1. 如果是单类型的列表,使用ArrayWritable就足够了
2。如果是把不同类型的Writable存储在一个列表中:
-- 可以使用GenerickWritable,把元素封装在一个ArrayWritable,这个貌似只能同一类型。
public class MyGenericWritable extends GenericWritable {
private static Class<? extends Writable>[] CLASSES = null;
static {
CLASSES = (Class<? extends Writable>[]) new Class[] {
ArrayWritable.class,
//add as many different Writable class as you want
};
}
@Override
protected Class<? extends Writable>[] getTypes() {
return CLASSES;
}
-- 可以使用写一个仿照MapWritable的ListWritable
//注意实现hashcode,equals,toString, comparTo (if possible)
//hashcode尤其重要,HashPartitioner通常用hashcode来选择reduce分区,所以为你的类写一个比较好的hashcode非常必要。
public class ListWritable extends ArrayList<Writable> implements Writable {
}
/**
* @author cloudera
*
*/
public class ListWritable extends ArrayList<Writable> implements Writable {
private List<Writable> list = new ArrayList<Writable>();
public void set(Writable writable){
list.add(writable);
}
@Override
public void readFields(DataInput in) throws IOException {
int nsize = in.readInt();
Configuration conf = new Configuration();
Text className = new Text();
while(nsize-->0){
Class theClass = null;
try {
className.readFields(in);
theClass = Class.forName(className.toString());
} catch (ClassNotFoundException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
Writable w = (Writable)ReflectionUtils.newInstance(theClass,conf);
w.readFields(in);
add(w);
}
}
@Override
public void write(DataOutput out) throws IOException {
Writable w = null;
out.writeInt(size());
for(int i = 0;i<size();i++){
w = get(i);
new Text(w.getClass().getName()).write(out);
w.write(out);
}
}
}
感谢各位的阅读,以上就是“Hadoop序列化怎么实现”的内容了,经过本文的学习后,相信大家对Hadoop序列化怎么实现这一问题有了更深刻的体会,具体使用情况还需要大家实践验证。这里是天达云,小编将为大家推送更多相关知识点的文章,欢迎关注!