Hadoop pipes编程

1. Hadoop pipes编程介绍

Hadoop pipes允许C++程序员编写mapreduce程序,它允许用户混用C++和Java的RecordReader, Mapper, Partitioner,Rducer和RecordWriter等五个组件。关于Hadoop pipes的设计思想,可参见我这篇文章:Hadoop Pipes设计原理

本文介绍了Hadoop pipes编程的基本方法,并给出了若干编程示例,最后介绍了Hadoop pipes高级编程方法,包括怎样在MapReduce中加载词典,怎么传递参数,怎样提高效率等。

2. Hadoop pipes编程初体验

Hadoop-0.20.2源代码中自带了三个pipes编程示例,它们位于目录src/examples/pipes/impl中,分别为wordcount-simple.cc,wordcount-part.cc和wordcount-nopipe.cc。下面简要介绍一下这三个程序。

(1) wordcount-simple.cc:Mapper和Reducer组件采用C++语言编写,RecordReader, Partitioner和RecordWriter采用Java语言编写,其中,RecordReader 为LineRecordReader(位于InputTextInputFormat中,按行读取数据,行所在的偏移量为key,行中的字符串为value),Partitioner为PipesPartitioner,RecordWriter为LineRecordWriter(位于InputTextOutputFormat中,输出格式为”key\tvalue\n”)

(2) wordcount-part.cc:Mapper,Partitioner和Reducer组件采用C++语言编写,其他采用Java编写

(3)wordcount-nopipe.cc:RecordReader,Mapper,Rducer和RecordWriter采用C++编写

接下来简单介绍一下wordcount-simple.cc的编译和运行方法。

在Hadoop的安装目录下,执行下面命令:

1
ant -Dcompile.c++=yes examples

则wordcount-simple.cc生成的可执行文件wordcount-simple被保存到了目录build/c++-examples/Linux-amd64-64/bin/中,然后将该可执行文件上传到HDFS的某一个目录下,如/user/XXX/ bin下:

1
bin/hadoop  -put  build/c++-examples/Linux-amd64-64/bin/wordcount-simple  /user/XXX/ bin/

上传一份数据到HDFS的/user/XXX /pipes_test_data目录下:

1
bin/hadoop  -put  data.txt  /user/XXX /pipes_test_data

直接使用下面命令提交作业:

1
2
3
4
5
6
7
8
9
10
11
12
13
bin/hadoop pipes \
 
-D hadoop.pipes.java.recordreader= true \
 
-D hadoop.pipes.java.recordwriter= true \
 
-D mapred.job.name= wordcount \
 
-input /user/XXX /pipes_test_data \
 
-output /user/XXX /pipes_test_output \
 
-program /user/XXX/ bin/wordcount-simple

3. Hadoop pipes编程方法

先从最基础的两个组件Mapper和Reducer说起。

(1) Mapper编写方法

用户若要实现Mapper组件,需继承HadoopPipes::Mapper虚基类,它的定义如下:

1
2
3
4
5
6
7
class Mapper: public Closable {
 
public :
 
virtual void map(MapContext& context) = 0;
 
};

用户必须实现map函数,它的参数是MapContext,该类的声明如下:

1
2
3
4
5
6
7
8
9
10
11
class MapContext: public TaskContext {
 
public :
 
virtual const std::string& getInputSplit() = 0;
 
virtual const std::string& getInputKeyClass() = 0;
 
virtual const std::string& getInputValueClass() = 0;
 
};

而TaskContext类地声明如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
class TaskContext {
 
public :
 
class Counter {
 
……
 
public :
 
Counter( int counterId) : id(counterId) {}
 
Counter( const Counter& counter) : id(counter.id) {}
 
……
 
};
 
virtual const JobConf* getJobConf() = 0;
 
virtual const std::string& getInputKey() = 0;
 
virtual const std::string& getInputValue() = 0;
 
virtual void emit( const std::string& key, const std::string& value) = 0;
 
virtual void progress() = 0;
 
…….
 
};

用户可以从context参数中获取当前的key,value,progress和inputsplit等数据信息,此外,还可以调用emit将结果回传给Java代码。

Mapper的构造函数带有一个HadoopPipes::TaskContext参数,用户可以通过它注册一些全局counter,对于程序调试和跟踪作业进度非常有用:

如果你想注册全局counter,在构造函数添加一些类似的代码:

1
2
3
4
5
6
7
WordCountMap(HadoopPipes::TaskContext& context) {
 
inputWords1 = context.getCounter(“group”, ”counter1”);
 
inputWords2 = context.getCounter(“group”, ”counter2”);
 
}

当需要增加counter值时,可以这样:

1
2
3
context.incrementCounter(inputWords1, 1);
 
context.incrementCounter(inputWords2, 1);

其中getCounter的两个参数分别为组名和组内计数器名,一个组中可以存在多个counter。

用户自定义的counter会在程序结束时,输出到屏幕上,当然,用户可以用通过web界面看到。

(2) Reducer编写方法

Reducer组件的编写方法跟Mapper组件类似,它需要继承虚基类public HadoopPipes::Reducer。

与Mapper组件唯一不同的地方时,map函数的参数类型为HadoopPipes::ReduceContext,它包含一个nextValue()方法,这允许用于遍历当前key对应的value列表,依次进行处理。

接下来介绍RecordReader, Partitioner和RecordWriter的编写方法:

(3) RecordReader编写方法

用户自定义的RecordReader类需要继承虚基类HadoopPipes::RecordReader,它的声明如下:

1
2
3
4
5
6
7
8
9
class RecordReader: public Closable {
 
public :
 
virtual bool next(std::string& key, std::string& value) = 0;
 
virtual float getProgress() = 0;
 
};

用户需要实现next和 getProgress两个方法。

用户自定义的RecordReader的构造函数可携带类型为HadoopPipes::MapContext的参数,通过该参数的getInputSplit()的方法,用户可以获取经过序列化的InpuSplit对象,Java端采用不同的InputFormat可导致InputSplit对象格式不同,但对于大多数InpuSplit对象,它们可以提供至少三个信息:当前要处理的InputSplit所在的文件名,所在文件中的偏移量,它的长度。用户获取这三个信息后,可使用libhdfs库读取文件,以实现next方法。

下面介绍一下反序列化InputSplit对象的方法:

【1】 如果Java端采用的InputFormat为WordCountInpuFormat,可以这样:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
class XXXReader: public HadoopPipes::RecordReader {
 
public :
 
XXXReader (HadoopPipes::MapContext& context) {
 
std::string filename;
 
HadoopUtils::StringInStream stream(context.getInputSplit());
 
HadoopUtils::deserializeString(filename, stream);
 
……
 
};

【2】 如果Java端采用的InputFormat为TextInpuFormat,可以这样:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
class XXXReader: public HadoopPipes::RecordReader {
 
public :
 
XXXReader (HadoopPipes::MapContext& context) {
 
std::string filename;
 
HadoopUtils::StringInStream stream(context.getInputSplit());
 
readString(filename, stream);
 
int start = ( int )readLong(stream);
 
int len = ( int )readLong(stream);
 
……
 
private :
 
void readString(std::string& t, HadoopUtils::StringInStream& stream)
 
{
 
int len = readShort(stream);
 
if (len > 0) {
 
// resize the string to the right length
 
t.resize(len);
 
// read into the string in 64k chunks
 
const int bufSize = 65536;
 
int offset = 0;
 
char buf[bufSize];
 
while (len > 0) {
 
int chunkLength = len > bufSize ? bufSize : len;
 
stream.read(buf, chunkLength);
 
t.replace(offset, chunkLength, buf, chunkLength);
 
offset += chunkLength;
 
len -= chunkLength;
 
}
 
} else {
 
t.clear();
 
}
 
}
 
long readLong(HadoopUtils::StringInStream& stream) {
 
long n;
 
char b;
 
stream.read(&b, 1);
 
n = ( long )(b & 0xff) << 56 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) << 48 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) << 40 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) << 32 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) << 24 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) << 16 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) << 8 ;
 
stream.read(&b, 1);
 
n |= ( long )(b & 0xff) ;
 
return n;
 
}
 
};

(4) Partitioner编写方法

用户自定义的Partitioner类需要继承虚基类HadoopPipes:: Partitioner,它的声明如下:

1
2
3
4
5
6
7
8
9
class Partitioner {
 
public :
 
virtual int partition( const std::string& key, int numOfReduces) = 0;
 
virtual ~Partitioner() {}
 
};

用户需要实现partition方法和 析构函数。

对于partition方法,框架会自动为它传入两个参数,分别为key值和reduce task的个数numOfReduces,用户只需返回一个0~ numOfReduces-1的值即可。

(5) RecordWriter编写方法

用户自定义的RecordWriter类需要继承虚基类HadoopPipes:: RecordWriter,它的声明如下:

1
2
3
4
5
6
7
8
9
class RecordWriter: public Closable {
 
public :
 
virtual void emit( const std::string& key,
 
const std::string& value) = 0;
 
};

用户自定的RecordWriter的构造函数可携带类型为HadoopPipes::MapContext的参数,通过该参数的getJobConf()可获取一个HadoopPipes::JobConf的对象,用户可从该对象中获取该reduce task的各种参数,如:该reduce task的编号(这对于确定输出文件名有用),reduce task的输出目录等。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
class MyWriter: public HadoopPipes::RecordWriter {
 
public :
 
MyWriter(HadoopPipes::ReduceContext& context) {
 
const HadoopPipes::JobConf* job = context.getJobConf();
 
int part = job->getInt( "mapred.task.partition" );
 
std::string outDir = job->get( "mapred.work.output.dir" );
 
……
 
}
 
}

用户需实现emit方法,将数据写入某个文件。

4. Hadoop pipes编程示例

网上有很多人怀疑Hadoop pipes自带的程序wordcount-nopipe.cc不能运行,各个论坛都有讨论,在此介绍该程序的设计原理和运行方法。

该运行需要具备以下前提:

(1) 采用的InputFormat为WordCountInputFormat,它位于src/test/下的org.apache.hadoop.mapred.pipes中

(2) 输入目录和输出目录需位于各个datanode的本地磁盘上,格式为:file:///home/xxx/pipes_test (注意,hdfs中的各种接口同时支持本地路径和HDFS路径,如果是HDFS上的路径,需要使用hdfs://host:9000/user/xxx,表示/user/xxx为namenode 为host的hdfs上的路径,而本地路径,需使用file:///home/xxx/pipes_test,表示/home/xxx/pipes_test为本地路径。例如,bin/hadoop fs –ls file:///home/xxx/pipes_test表示列出本地磁盘上/home/xxx/pipes_tes下的文件)

待确定好各个datanode的本地磁盘上有输入数据/home/xxx/pipes_test/data.txt后,用户首先上传可执行文件到HDFS中:

1
bin/hadoop  -put  build/c++-examples/Linux-amd64-64/bin/wordcount-nopipe  /user/XXX/bin/

然后使用下面命令提交该作业:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
bin/hadoop pipes \
 
-D hadoop.pipes.java.recordreader= false \
 
-D hadoop.pipes.java.recordwriter= false \
 
-D mapred.job.name=wordcount \
 
-D mapred.input.format. class =org.apache.hadoop.mapred.pipes.WordCountInputFormat \
 
-libjars hadoop-0.20.2-test.jar \
 
-input file: ///home/xxx/pipes_test/data.txt \
 
-output file: ///home/xxx/pipes_output \
 
-program /user/XXX/bin/wordcount-nopipe

5. Hadoop pipes高级编程

如果用户需要在mapreduce作业中加载词典或者传递参数,可这样做:

(1) 提交作业时,用-files选项,将词典(需要传递参数可以放到一个配置文件中)上传给各个datanode,如:

1
2
3
4
5
6
7
8
9
10
11
bin/hadoop pipes \
 
-D hadoop.pipes.java.recordreader= false \
 
-D hadoop.pipes.java.recordwriter= false \
 
-D mapred.job.name=wordcount \
 
-files dic.txt \
 
….

(2) 在Mapper或者Reducer的构造函数中,将字典文件以本地文件的形式打开,并把内容保存到一个map或者set中,然后再map()或者reduce()函数中使用即可,如:

1
2
3
4
5
6
7
WordCountMap(HadoopPipes::TaskContext& context) {
 
file = fopen (“dic.txt”, "r" ); //C库函数
 
…….
 
}

为了提高系能,RecordReader和RecordWriter最好采用Java代码实现(或者重用Hadoop中自带的),这是因为Hadoop自带的C++库libhdfs采用JNI实现,底层还是要调用Java相关接口,效率很低,此外,如果要处理的文件为二进制文件或者其他非文本文件,libhdfs可能不好处理。

6. 总结

Hadoop pipes使C++程序员编写MapReduce作业变得可能,它简单好用,提供了用户所需的大部分功能。

1.Hadoop pipes编程介绍

Hadoop pipes允许C++程序员编写mapreduce程序,它允许用户混用C++和Java的RecordReader,Mapper,Partitioner,Rducer和RecordWriter等五个组件。关于Hadoop pipes的设计思想,可参见我这篇文章:

本文介绍了Hadoop pipes编程的基本方法,并给出了若干编程示例,最后介绍了Hadoop pipes高级编程方法,包括怎样在MapReduce中加载词典,怎么传递参数,怎样提高效率等。

2.Hadoop pipes编程初体验

Hadoop-0.20.2源代码中自带了三个pipes编程示例,它们位于目录src/examples/pipes/impl中,分别为wordcount-simple.cc,wordcount-part.cc和wordcount-nopipe.cc。下面简要介绍一下这三个程序。

(1)wordcount-simple.cc:Mapper和Reducer组件采用C++语言编写,RecordReader, Partitioner和RecordWriter采用Java语言编写,其中,RecordReader为LineRecordReader(位于InputTextInputFormat中,按行读取数据,行所在的偏移量为key,行中的字符串为value),Partitioner为PipesPartitioner,RecordWriter为LineRecordWriter(位于InputTextOutputFormat中,输出格式为”key\tvalue\n”)

(2)wordcount-part.cc:Mapper,Partitioner和Reducer组件采用C++语言编写,其他采用Java编写

(3)wordcount-nopipe.cc:RecordReader,Mapper,Rducer和RecordWriter采用C++编写

接下来简单介绍一下wordcount-simple.cc的编译和运行方法。

在Hadoop的安装目录下,执行下面命令:

ant -Dcompile.c++=yes examples

则wordcount-simple.cc生成的可执行文件wordcount-simple被保存到了目录build/c++-examples/Linux-amd64-64/bin/中,然后将该可执行文件上传到HDFS的某一个目录下,如/user/XXX/ bin下:

bin/hadoop-putbuild/c++-examples/Linux-amd64-64/bin/wordcount-simple/user/XXX/ bin/

上传一份数据到HDFS的/user/XXX /pipes_test_data目录下:

bin/hadoop-putdata.txt/user/XXX /pipes_test_data

直接使用下面命令提交作业:

bin/hadoop pipes \

-D hadoop.pipes.java.recordreader=true \

-D hadoop.pipes.java.recordwriter=true \

-D mapred.job.name= wordcount \

-input /user/XXX /pipes_test_data \

-output /user/XXX /pipes_test_output \

-program /user/XXX/ bin/wordcount-simple

3.Hadoop pipes编程方法

先从最基础的两个组件Mapper和Reducer说起。

(1)Mapper编写方法

用户若要实现Mapper组件,需继承HadoopPipes::Mapper虚基类,它的定义如下:

class Mapper: public Closable {

public:

virtual void map(MapContext& context) = 0;

};

用户必须实现map函数,它的参数是MapContext,该类的声明如下:

class MapContext: public TaskContext {

public:

virtual const std::string& getInputSplit() = 0;

virtual const std::string& getInputKeyClass() = 0;

virtual const std::string& getInputValueClass() = 0;

};

而TaskContext类地声明如下:

class TaskContext {

public:

class Counter {

……

public:

Counter(int counterId) : id(counterId) {}

Counter(const Counter& counter) : id(counter.id) {}

……

};

virtual const JobConf* getJobConf() = 0;

virtual const std::string& getInputKey() = 0;

virtual const std::string& getInputValue() = 0;

virtual void emit(const std::string& key, const std::string& value) = 0;

virtual void progress() = 0;

…….

};

用户可以从context参数中获取当前的key,value,progress和inputsplit等数据信息,此外,还可以调用emit将结果回传给Java代码。

Mapper的构造函数带有一个HadoopPipes::TaskContext参数,用户可以通过它注册一些全局counter,对于程序调试和跟踪作业进度非常有用:

如果你想注册全局counter,在构造函数添加一些类似的代码:

WordCountMap(HadoopPipes::TaskContext& context) {

inputWords1 = context.getCounter(“group”, ”counter1”);

inputWords2 = context.getCounter(“group”, ”counter2”);

}

当需要增加counter值时,可以这样:

context.incrementCounter(inputWords1, 1);

context.incrementCounter(inputWords2, 1);

其中getCounter的两个参数分别为组名和组内计数器名,一个组中可以存在多个counter。

用户自定义的counter会在程序结束时,输出到屏幕上,当然,用户可以用通过web界面看到。

(2)Reducer编写方法

Reducer组件的编写方法跟Mapper组件类似,它需要继承虚基类public HadoopPipes::Reducer。

与Mapper组件唯一不同的地方时,map函数的参数类型为HadoopPipes::ReduceContext,它包含一个nextValue()方法,这允许用于遍历当前key对应的value列表,依次进行处理。

接下来介绍RecordReader,Partitioner和RecordWriter的编写方法:

(3)RecordReader编写方法

用户自定义的RecordReader类需要继承虚基类HadoopPipes::RecordReader,它的声明如下:

class RecordReader: public Closable {

public:

virtual bool next(std::string& key, std::string& value) = 0;

virtual float getProgress() = 0;

};

用户需要实现next和getProgress两个方法。

用户自定义的RecordReader的构造函数可携带类型为HadoopPipes::MapContext的参数,通过该参数的getInputSplit()的方法,用户可以获取经过序列化的InpuSplit对象,Java端采用不同的InputFormat可导致InputSplit对象格式不同,但对于大多数InpuSplit对象,它们可以提供至少三个信息:当前要处理的InputSplit所在的文件名,所在文件中的偏移量,它的长度。用户获取这三个信息后,可使用libhdfs库读取文件,以实现next方法。

(4)Partitioner编写方法

用户自定义的Partitioner类需要继承虚基类HadoopPipes:: Partitioner,它的声明如下:

class Partitioner {

public:

virtual int partition(const std::string& key, int numOfReduces) = 0;

virtual ~Partitioner() {}

};

用户需要实现partition方法和析构函数。

对于partition方法,框架会自动为它传入两个参数,分别为key值和reduce task的个数numOfReduces,用户只需返回一个0~ numOfReduces-1的值即可。

(5)RecordWriter编写方法

用户自定义的RecordWriter类需要继承虚基类HadoopPipes:: RecordWriter,它的声明如下:

class RecordWriter: public Closable {

public:

virtual void emit(const std::string& key,

const std::string& value) = 0;

};

用户自定的RecordWriter的构造函数可携带类型为HadoopPipes::MapContext的参数,通过该参数的getJobConf()可获取一个HadoopPipes::JobConf的对象,用户可从该对象中获取该reduce task的各种参数,如:该reduce task的编号(这对于确定输出文件名有用),reduce task的输出目录等。

class WordCountWriter: public HadoopPipes::RecordWriter {

public:

MyWriter(HadoopPipes::ReduceContext& context) {

const HadoopPipes::JobConf* job = context.getJobConf();

int part = job->getInt(“mapred.task.partition”);

std::string outDir = job->get(“mapred.work.output.dir”);

……

}

}

用户需实现emit方法,将数据写入某个文件。

4.Hadoop pipes编程示例

网上有很多人怀疑Hadoop pipes自带的程序wordcount-nopipe.cc不能运行,各个论坛都有讨论,在此介绍该程序的设计原理和运行方法。

该运行需要具备以下前提:

(1) 采用的InputFormat为WordCountInputFormat,它位于src/test/下的org.apache.hadoop.mapred.pipes中

(2) 输入目录和输出目录需位于各个datanode的本地磁盘上,格式为:file:///home/xxx/pipes_test(注意,hdfs中的各种接口同时支持本地路径和HDFS路径,如果是HDFS上的路径,需要使用hdfs://host:9000/user/xxx,表示/user/xxx为namenode为host的hdfs上的路径,而本地路径,需使用file:///home/xxx/pipes_test,表示/home/xxx/pipes_test为本地路径)

待确定好各个datanode的本地磁盘上有输入数据/home/xxx/pipes_test/data.txt后,用户首先上传可执行文件到HDFS中:

bin/hadoop-putbuild/c++-examples/Linux-amd64-64/bin/wordcount-simple/user/XXX/ bin/

然后使用下面命令运行该程序:

bin/hadoop pipes \

-D hadoop.pipes.java.recordreader=false \

-D hadoop.pipes.java.recordwriter=false \

-D mapred.job.name=wordcount \

-D mapred.input.format.class=org.apache.hadoop.mapred.pipes.WordCountInputFormat \

-libjars hadoop-0.20.2-test.jar \

-input file:/home/xxx/pipes_test/data.txt \

-output file:/home/xxx/pipes_output \

-program /user/XXX/ bin/wordcount-nopipe

5.Hadoop pipes高级编程

如果用户需要在mapreduce作业中加载词典或者传递参数,可这样做:

(1) 提交作业时,用-files选项,将词典(需要传递参数可以放到一个配置文件中)上传给各个datanode,如

bin/hadoop pipes \

-D hadoop.pipes.java.recordreader=false \

-D hadoop.pipes.java.recordwriter=false \

-D mapred.job.name=wordcount \

-files dic.txt \

….

(2)在Mapper或者Reducer的构造函数中,将字典文件以本地文件的形式打开,并把内容保存到一个map或者set中,然后再map()或者reduce()函数中使用即可,如

WordCountMap(HadoopPipes::TaskContext& context) {

file = fopen(“dic.txt”, “r”); //C库函数

…….

}

为了提高系能,RecordReader和RecordWriter最好采用Java代码实现(或者重用Hadoop中自带的),这是因为Hadoop自带的C++库libhdfs采用JNI实现,底层还是要调用Java相关接口,效率很低,此外,如果要处理的文件为二进制文件或者其他非文本文件,libhdfs可能不好处理。

6.总结

1. Hadoop pipes编程介绍

Hadoop pipes允许C++程序员编写mapreduce程序,它允许用户混用C++和Java的RecordReader,Mapper,Partitioner,Rducer和RecordWriter等五个组件。关于Hadoop pipes的设计思想,可参见我这篇文章:

本文介绍了Hadoop pipes编程的基本方法,并给出了若干编程示例,最后介绍了Hadoop pipes高级编程方法,包括怎样在MapReduce中加载词典,怎么传递参数,怎样提高效率等。

2. Hadoop pipes编程初体验

Hadoop-0.20.2源代码中自带了三个pipes编程示例,它们位于目录src/examples/pipes/impl中,分别为wordcount-simple.cc,wordcount-part.cc和wordcount-nopipe.cc。下面简要介绍一下这三个程序。

(1) wordcount-simple.cc:Mapper和Reducer组件采用C++语言编写,RecordReader, Partitioner和RecordWriter采用Java语言编写,其中,RecordReader 为LineRecordReader(位于InputTextInputFormat中,按行读取数据,行所在的偏移量为key,行中的字符串为value),Partitioner为PipesPartitioner,RecordWriter为LineRecordWriter(位于InputTextOutputFormat中,输出格式为”key\tvalue\n”)

(2) wordcount-part.cc:Mapper,Partitioner和Reducer组件采用C++语言编写,其他采用Java编写

(3)wordcount-nopipe.cc:RecordReader,Mapper,Rducer和RecordWriter采用C++编写

接下来简单介绍一下wordcount-simple.cc的编译和运行方法。

在Hadoop的安装目录下,执行下面命令:

ant -Dcompile.c++=yes examples

则wordcount-simple.cc生成的可执行文件wordcount-simple被保存到了目录build/c++-examples/Linux-amd64-64/bin/中,然后将该可执行文件上传到HDFS的某一个目录下,如/user/XXX/ bin下:

bin/hadoop -put build/c++-examples/Linux-amd64-64/bin/wordcount-simple /user/XXX/ bin/

上传一份数据到HDFS的/user/XXX /pipes_test_data目录下:

bin/hadoop -put data.txt /user/XXX /pipes_test_data

直接使用下面命令提交作业:

bin/hadoop pipes \

-D hadoop.pipes.java.recordreader=true \

-D hadoop.pipes.java.recordwriter=true \

-D mapred.job.name= wordcount \

-input /user/XXX /pipes_test_data \

-output /user/XXX /pipes_test_output \

-program /user/XXX/ bin/wordcount-simple

3. Hadoop pipes编程方法

先从最基础的两个组件Mapper和Reducer说起。

(1) Mapper编写方法

用户若要实现Mapper组件,需继承HadoopPipes::Mapper虚基类,它的定义如下:

class Mapper: public Closable {

public:

virtual void map(MapContext& context) = 0;

};

用户必须实现map函数,它的参数是MapContext,该类的声明如下:

class MapContext: public TaskContext {

public:

virtual const std::string& getInputSplit() = 0;

virtual const std::string& getInputKeyClass() = 0;

virtual const std::string& getInputValueClass() = 0;

};

而TaskContext类地声明如下:

class TaskContext {

public:

class Counter {

……

public:

Counter(int counterId) : id(counterId) {}

Counter(const Counter& counter) : id(counter.id) {}

……

};

virtual const JobConf* getJobConf() = 0;

virtual const std::string& getInputKey() = 0;

virtual const std::string& getInputValue() = 0;

virtual void emit(const std::string& key, const std::string& value) = 0;

virtual void progress() = 0;

…….

};

用户可以从context参数中获取当前的key,value,progress和inputsplit等数据信息,此外,还可以调用emit将结果回传给Java代码。

Mapper的构造函数带有一个HadoopPipes::TaskContext参数,用户可以通过它注册一些全局counter,对于程序调试和跟踪作业进度非常有用:

如果你想注册全局counter,在构造函数添加一些类似的代码:

WordCountMap(HadoopPipes::TaskContext& context) {

inputWords1 = context.getCounter(“group”, ”counter1”);

inputWords2 = context.getCounter(“group”, ”counter2”);

}

当需要增加counter值时,可以这样:

context.incrementCounter(inputWords1, 1);

context.incrementCounter(inputWords2, 1);

其中getCounter的两个参数分别为组名和组内计数器名,一个组中可以存在多个counter。

用户自定义的counter会在程序结束时,输出到屏幕上,当然,用户可以用通过web界面看到。

(2) Reducer编写方法

Reducer组件的编写方法跟Mapper组件类似,它需要继承虚基类public HadoopPipes::Reducer。

与Mapper组件唯一不同的地方时,map函数的参数类型为HadoopPipes::ReduceContext,它包含一个nextValue()方法,这允许用于遍历当前key对应的value列表,依次进行处理。

接下来介绍RecordReader, Partitioner和RecordWriter的编写方法:

(3) RecordReader编写方法

用户自定义的RecordReader类需要继承虚基类HadoopPipes::RecordReader,它的声明如下:

class RecordReader: public Closable {

public:

virtual bool next(std::string& key, std::string& value) = 0;

virtual float getProgress() = 0;

};

用户需要实现next和 getProgress两个方法。

用户自定义的RecordReader的构造函数可携带类型为HadoopPipes::MapContext的参数,通过该参数的getInputSplit()的方法,用户可以获取经过序列化的InpuSplit对象,Java端采用不同的InputFormat可导致InputSplit对象格式不同,但对于大多数InpuSplit对象,它们可以提供至少三个信息:当前要处理的InputSplit所在的文件名,所在文件中的偏移量,它的长度。用户获取这三个信息后,可使用libhdfs库读取文件,以实现next方法。

(4) Partitioner编写方法

用户自定义的Partitioner类需要继承虚基类HadoopPipes:: Partitioner,它的声明如下:

class Partitioner {

public:

virtual int partition(const std::string& key, int numOfReduces) = 0;

virtual ~Partitioner() {}

};

用户需要实现partition方法和 析构函数。

对于partition方法,框架会自动为它传入两个参数,分别为key值和reduce task的个数numOfReduces,用户只需返回一个0~ numOfReduces-1的值即可。

(5) RecordWriter编写方法

用户自定义的RecordWriter类需要继承虚基类HadoopPipes:: RecordWriter,它的声明如下:

class RecordWriter: public Closable {

public:

virtual void emit(const std::string& key,

const std::string& value) = 0;

};

用户自定的RecordWriter的构造函数可携带类型为HadoopPipes::MapContext的参数,通过该参数的getJobConf()可获取一个HadoopPipes::JobConf的对象,用户可从该对象中获取该reduce task的各种参数,如:该reduce task的编号(这对于确定输出文件名有用),reduce task的输出目录等。

class WordCountWriter: public HadoopPipes::RecordWriter {

public:

MyWriter(HadoopPipes::ReduceContext& context) {

const HadoopPipes::JobConf* job = context.getJobConf();

int part = job->getInt(“mapred.task.partition”);

std::string outDir = job->get(“mapred.work.output.dir”);

……

}

}

用户需实现emit方法,将数据写入某个文件。

4. Hadoop pipes编程示例

网上有很多人怀疑Hadoop pipes自带的程序wordcount-nopipe.cc不能运行,各个论坛都有讨论,在此介绍该程序的设计原理和运行方法。

该运行需要具备以下前提:

(1) 采用的InputFormat为WordCountInputFormat,它位于src/test/下的org.apache.hadoop.mapred.pipes中

(2) 输入目录和输出目录需位于各个datanode的本地磁盘上,格式为:file:///home/xxx/pipes_test (注意,hdfs中的各种接口同时支持本地路径和HDFS路径,如果是HDFS上的路径,需要使用hdfs://host:9000/user/xxx,表示/user/xxx为namenode 为host的hdfs上的路径,而本地路径,需使用file:///home/xxx/pipes_test,表示/home/xxx/pipes_test为本地路径)

待确定好各个datanode的本地磁盘上有输入数据/home/xxx/pipes_test/data.txt后,用户首先上传可执行文件到HDFS中:

bin/hadoop -put build/c++-examples/Linux-amd64-64/bin/wordcount-simple /user/XXX/ bin/

然后使用下面命令运行该程序:

bin/hadoop pipes \

-D hadoop.pipes.java.recordreader=false \

-D hadoop.pipes.java.recordwriter=false \

-D mapred.job.name=wordcount \

-D mapred.input.format.class=org.apache.hadoop.mapred.pipes.WordCountInputFormat \

-libjars hadoop-0.20.2-test.jar \

-input file:/home/xxx/pipes_test/data.txt \

-output file:/home/xxx/pipes_output \

-program /user/XXX/ bin/wordcount-nopipe

5. Hadoop pipes高级编程

如果用户需要在mapreduce作业中加载词典或者传递参数,可这样做:

(1) 提交作业时,用-files选项,将词典(需要传递参数可以放到一个配置文件中)上传给各个datanode,如

bin/hadoop pipes \

-D hadoop.pipes.java.recordreader=false \

-D hadoop.pipes.java.recordwriter=false \

-D mapred.job.name=wordcount \

-files dic.txt \

….

(2) 在Mapper或者Reducer的构造函数中,将字典文件以本地文件的形式打开,并把内容保存到一个map或者set中,然后再map()或者reduce()函数中使用即可,如

WordCountMap(HadoopPipes::TaskContext& context) {

file = fopen(“dic.txt”, “r”); //C库函数

…….

}

为了提高系能,RecordReader和RecordWriter最好采用Java代码实现(或者重用Hadoop中自带的),这是因为Hadoop自带的C++库libhdfs采用JNI实现,底层还是要调用Java相关接口,效率很低,此外,如果要处理的文件为二进制文件或者其他非文本文件,libhdfs可能不好处理。

6. 总结

Hadoop pipes使C++程序员编写MapReduce作业变得可能,它简单好用,提供了用户所需的大部分功能。

Hadoop pipes使C++程序员编写MapReduce作业变得可能,它简单好用,提供了用户所需的大部分功能。

原创文章,转载请注明: 转载自董的博客

本文链接地址: http://dongxicheng.org/mapreduce/hadoop-pipes-programming/

  • 0
    点赞
  • 1
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值