一、MapReduce简介
MapReduce是Google的一项重要技术,它首先是一个编程模型,用对于大数据量的计算,通常采用的处理手法就是并行计算。以进行大数据量的计算。MapReduce通过简化编程模型,降低了开发并行应用的入门门槛。
(1)Mapper负责“分”,即把复杂的任务分解为若干个“简单的任务”来处理。“简单的任务”包含三层含义:一是数据或计算的规模相对原任务要大大缩小;二是就近计算原则,即任务会分配到存放着所需数据的节点上进行计算;三是这些小任务可以并行计算,彼此间几乎没有依赖关系。
(2)Reducer负责对map阶段的结果进行汇总。至于需要多少个Reducer,用户可以根据具体问题,通过在mapred-site.xml配置文件里设置参数mapred.reduce.tasks的值,缺省值为1。
一个比较形象的语言解释MapReduce:
We want to count all the books in the library. You count up shelf #1, I count up shelf #2. That’s map. The more people we get, the faster it goes.
我们要数图书馆中的所有书。你数1号书架,我数2号书架。这就是“Map”。我们人越多,数书就更快。
Now we get together and add our individual counts. That’s reduce.
现在我们到一起,把所有人的统计数加在一起。这就是“Reduce”。
二、windows intelij 跑 MR WordCount
1.创建maven工程
2.添加maven依赖
在pom.xml添加依赖,对于hadoop 2.7.3版本的hadoop,需要的jar包有以下几个:
-
hadoop-common
-
hadoop-hdfs
-
hadoop-mapreduce-client-core
-
hadoop-mapreduce-client-jobclient
-
log4j( 打印日志)
pom.xml中的依赖如下:
<dependencies>
<dependency>
<groupId>junit</groupId>
<artifactId>junit</artifactId>
<version>4.12</version>
<scope>test</scope>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-common</artifactId>
<version>2.7.3</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-hdfs</artifactId>
<version>2.7.3</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-mapreduce-client-core</artifactId>
<version>2.7.3</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-mapreduce-client-jobclient</artifactId>
<version>2.7.3</version>
</dependency>
<dependency>
<groupId>log4j</groupId>
<artifactId>log4j</artifactId>
<version>1.2.17</version>
</dependency>
</dependencies>
3.配置log4j
在src/main/resources
目录下新增log4j的配置文件log4j.properties
,内容如下:
log4j.rootLogger = debug,stdout
### 输出信息到控制抬 ###
log4j.appender.stdout = org.apache.log4j.ConsoleAppender
log4j.appender.stdout.Target = System.out
log4j.appender.stdout.layout = org.apache.log4j.PatternLayout
log4j.appender.stdout.layout.ConversionPattern = [%-5p] %d{yyyy-MM-dd HH:mm:ss,SSS} method:%l%n%m%n
4.启动Hadoop
5.运行WordCount(从本地读取文件)
在工程根目录下新建input文件夹,input文件夹下新增dream.txt,随便写入一些单词:
I have a dream
a dream
在src/main/java目录下新建包,新增FileUtil.java,创建一个删除output文件的函数,以后就不用手动删除了。内容如下:
package com.mrtest.hadoop;
import java.io.File;
/**
* Created by bee on 3/25/17.
*/
public class FileUtil {
public static boolean deleteDir(String path) {
File dir = new File(path);
if (dir.exists()) {
for (File f : dir.listFiles()) {
if (f.isDirectory()) {
deleteDir(f.getName());
} else {
f.delete();
}
}
dir.delete();
return true;
} else {
System.out.println("文件(夹)不存在!");
return false;
}
}
}
2、wordcount程序实现
1、编写map方法
package com.neusoft;
import java.io.IOException;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;
/**
* 该类做为一个 mapTask 使用。类声名中所使用的四个泛型意义为别为:
*
* KEYIN: 默认情况下,是mr框架所读到的一行文本的起始偏移量,Long,
* 但是在hadoop中有自己的更精简的序列化接口,所以不直接用Long,而用LongWritable
* VALUEIN: 默认情况下,是mr框架所读到的一行文本的内容,String,同上,用Text
* KEYOUT: 是用户自定义逻辑处理完成之后输出数据中的key,在此处是单词,String,同上,用Text
* VALUEOUT:是用户自定义逻辑处理完成之后输出数据中的value,在此处是单词次数,Integer,同上,用IntWritable
*/
public class WordcountMapper extends Mapper<LongWritable, Text, Text, IntWritable> {
/**
* map阶段的业务逻辑就写在自定义的map()方法中 maptask会对每一行输入数据调用一次我们自定义的map()方法
*/
@Override
protected void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException {
// 将maptask传给我们的文本内容先转换成String
String line = value.toString();
// 根据空格将这一行切分成单词
String[] words = line.split(" ");
// 将单词输出为<单词,1>
for (String word : words) {
// 将单词作为key,将次数1作为value,以便于后续的数据分发,可以根据单词分发,以便于相同单词会到相同的reduce task
context.write(new Text(word), new IntWritable(1));
}
}
}
2、编写reduce方法
package com.neusoft;
import java.io.IOException;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;
/**
* 与 Mapper 类似,继承的同事声名四个泛型。
* KEYIN, VALUEIN 对应 mapper输出的KEYOUT,VALUEOUT类型对应
* KEYOUT, VALUEOUT 是自定义reduce逻辑处理结果的输出数据类型。此处 keyOut 表示单个单词,valueOut 对应的是总次数
*/
public class WordcountReducer extends Reducer<Text, IntWritable, Text, IntWritable>{
/**
* <angelababy,1><angelababy,1><angelababy,1><angelababy,1><angelababy,1>
* <hello,1><hello,1><hello,1><hello,1><hello,1><hello,1>
* <banana,1><banana,1><banana,1><banana,1><banana,1><banana,1>
* 入参key,是一组相同单词kv对的key
*/
@Override
protected void reduce(Text key, Iterable<IntWritable> values, Context context) throws IOException, InterruptedException {
int count=0;
for(IntWritable value : values){
count += value.get();
}
context.write(key, new IntWritable(count)); //输出每一个单词出现的次数
}
}
解释:
map()方法中我们设置的时候单词做为key,1作为value写出,比如
package com.neusoft;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
/**
* 相当于一个yarn集群的客户端
* 需要在此封装我们的mr程序的相关运行参数,指定jar包
* 最后提交给yarn
*/
public class WordcountDriver {
public static void main(String[] args) throws Exception {
System.setProperty("hadoop.home.dir", "e:/hadoop-2.8.3");
if (args == null || args.length == 0) {
return;
}
//该对象会默认读取环境中的 hadoop 配置。当然,也可以通过 set 重新进行配置
Configuration conf = new Configuration();
//job 是 yarn 中任务的抽象。
Job job = Job.getInstance(conf);
/*job.setJar("/home/hadoop/wc.jar");*/
//指定本程序的jar包所在的本地路径
job.setJarByClass(WordcountDriver.class);
//指定本业务job要使用的mapper/Reducer业务类
job.setMapperClass(WordcountMapper.class);
job.setReducerClass(WordcountReducer.class);
//指定mapper输出数据的kv类型。需要和 Mapper 中泛型的类型保持一致
job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(IntWritable.class);
//指定最终输出的数据的kv类型。这里也是 Reduce 的 key,value类型。
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(IntWritable.class);
//指定job的输入原始文件所在目录
FileInputFormat.setInputPaths(job, new Path(args[0]));
//指定job的输出结果所在目录
FileOutputFormat.setOutputPath(job, new Path(args[1]));
//将job中配置的相关参数,以及job所用的java类所在的jar包,提交给yarn去运行
/*job.submit();*/
boolean res = job.waitForCompletion(true);
System.exit(res?0:1);
}
}
这里在main函数中新增了一个String类型的数组,如果想用main函数的args数组接受参数,在运行时指定输入和输出路径也是可以的。运行WordCount之前,配置Configuration并指定Program arguments即可。
6.运行WordCount(从HDFS读取文件)
在HDFS上新建文件夹:
hadoop fs -mkdir /worddir
如果出现Namenode安全模式导致的不能创建文件夹提示:
mkdir: Cannot create directory /worddir. Name node is in safe mode.
运行以下命令关闭safe mode:
hadoop dfsadmin -safemode leave
上传本地文件:
hadoop fs -put dream.txt /worddir
修改otherArgs参数,指定输入为文件在HDFS上的路径:
String[] otherArgs = new String[]{"hdfs://localhost:8020/worddir/dream.txt","output"};
7.解决org.apache.hadoop.io.nativeio.NativeIO$Windows.access0(Ljava/lang/String;I)Z
Windows7出现这个问题
需要在你的 Module下创建一个名为org.apache.hadoop.io.nativeio的包
在这个包下创建一个Class 名为 NativeIO
然后粘贴如下代码
/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.io.nativeio;
import java.io.File;
import java.io.FileDescriptor;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.lang.reflect.Field;
import java.nio.ByteBuffer;
import java.nio.MappedByteBuffer;
import java.nio.channels.FileChannel;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.classification.InterfaceStability;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.CommonConfigurationKeys;
import org.apache.hadoop.fs.HardLink;
import org.apache.hadoop.io.IOUtils;
import org.apache.hadoop.io.SecureIOUtils.AlreadyExistsException;
import org.apache.hadoop.util.NativeCodeLoader;
import org.apache.hadoop.util.Shell;
import org.apache.hadoop.util.PerformanceAdvisory;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import sun.misc.Unsafe;
import com.google.common.annotations.VisibleForTesting;
/**
* JNI wrappers for various native IO-related calls not available in Java.
* These functions should generally be used alongside a fallback to another
* more portable mechanism.
*/
@InterfaceAudience.Private
@InterfaceStability.Unstable
public class NativeIO {
public static class POSIX {
// Flags for open() call from bits/fcntl.h
public static final int O_RDONLY = 00;
public static final int O_WRONLY = 01;
public static final int O_RDWR = 02;
public static final int O_CREAT = 0100;
public static final int O_EXCL = 0200;
public static final int O_NOCTTY = 0400;
public static final int O_TRUNC = 01000;
public static final int O_APPEND = 02000;
public static final int O_NONBLOCK = 04000;
public static final int O_SYNC = 010000;
public static final int O_ASYNC = 020000;
public static final int O_FSYNC = O_SYNC;
public static final int O_NDELAY = O_NONBLOCK;
// Flags for posix_fadvise() from bits/fcntl.h
/* No further special treatment. */
public static final int POSIX_FADV_NORMAL = 0;
/* Expect random page references. */
public static final int POSIX_FADV_RANDOM = 1;
/* Expect sequential page references. */
public static final int POSIX_FADV_SEQUENTIAL = 2;
/* Will need these pages. */
public static final int POSIX_FADV_WILLNEED = 3;
/* Don't need these pages. */
public static final int POSIX_FADV_DONTNEED = 4;
/* Data will be accessed once. */
public static final int POSIX_FADV_NOREUSE = 5;
/* Wait upon writeout of all pages
in the range before performing the
write. */
public static final int SYNC_FILE_RANGE_WAIT_BEFORE = 1;
/* Initiate writeout of all those
dirty pages in the range which are
not presently under writeback. */
public static final int SYNC_FILE_RANGE_WRITE = 2;
/* Wait upon writeout of all pages in
the range after performing the
write. */
public static final int SYNC_FILE_RANGE_WAIT_AFTER = 4;
private static final Log LOG = LogFactory.getLog(NativeIO.class);
private static boolean nativeLoaded = false;
private static boolean fadvisePossible = true;
private static boolean syncFileRangePossible = true;
static final String WORKAROUND_NON_THREADSAFE_CALLS_KEY =
"hadoop.workaround.non.threadsafe.getpwuid";
static final boolean WORKAROUND_NON_THREADSAFE_CALLS_DEFAULT = true;
private static long cacheTimeout = -1;
private static CacheManipulator cacheManipulator = new CacheManipulator();
public static CacheManipulator getCacheManipulator() {
return cacheManipulator;
}
public static void setCacheManipulator(CacheManipulator cacheManipulator) {
POSIX.cacheManipulator = cacheManipulator;
}
/**
* Used to manipulate the operating system cache.
*/
@VisibleForTesting
public static class CacheManipulator {
public void mlock(String identifier, ByteBuffer buffer,
long len) throws IOException {
POSIX.mlock(buffer, len);
}
public long getMemlockLimit() {
return NativeIO.getMemlockLimit();
}
public long getOperatingSystemPageSize() {
return NativeIO.getOperatingSystemPageSize();
}
public void posixFadviseIfPossible(String identifier,
FileDescriptor fd, long offset, long len, int flags)
throws NativeIOException {
NativeIO.POSIX.posixFadviseIfPossible(identifier, fd, offset,
len, flags);
}
public boolean verifyCanMlock() {
return NativeIO.isAvailable();
}
}
/**
* A CacheManipulator used for testing which does not actually call mlock.
* This allows many tests to be run even when the operating system does not
* allow mlock, or only allows limited mlocking.
*/
@VisibleForTesting
public static class NoMlockCacheManipulator extends CacheManipulator {
public void mlock(String identifier, ByteBuffer buffer,
long len) throws IOException {
LOG.info("mlocking " + identifier);
}
public long getMemlockLimit() {
return 1125899906842624L;
}
public long getOperatingSystemPageSize() {
return 4096;
}
public boolean verifyCanMlock() {
return true;
}
}
static {
if (NativeCodeLoader.isNativeCodeLoaded()) {
try {
Configuration conf = new Configuration();
workaroundNonThreadSafePasswdCalls = conf.getBoolean(
WORKAROUND_NON_THREADSAFE_CALLS_KEY,
WORKAROUND_NON_THREADSAFE_CALLS_DEFAULT);
initNative();
nativeLoaded = true;
cacheTimeout = conf.getLong(
CommonConfigurationKeys.HADOOP_SECURITY_UID_NAME_CACHE_TIMEOUT_KEY,
CommonConfigurationKeys.HADOOP_SECURITY_UID_NAME_CACHE_TIMEOUT_DEFAULT) *
1000;
LOG.debug("Initialized cache for IDs to User/Group mapping with a " +
" cache timeout of " + cacheTimeout/1000 + " seconds.");
} catch (Throwable t) {
// This can happen if the user has an older version of libhadoop.so
// installed - in this case we can continue without native IO
// after warning
PerformanceAdvisory.LOG.debug("Unable to initialize NativeIO libraries", t);
}
}
}
/**
* Return true if the JNI-based native IO extensions are available.
*/
public static boolean isAvailable() {
return NativeCodeLoader.isNativeCodeLoaded() && nativeLoaded;
}
private static void assertCodeLoaded() throws IOException {
if (!isAvailable()) {
throw new IOException("NativeIO was not loaded");
}
}
/** Wrapper around open(2) */
public static native FileDescriptor open(String path, int flags, int mode) throws IOException;
/** Wrapper around fstat(2) */
private static native Stat fstat(FileDescriptor fd) throws IOException;
/** Native chmod implementation. On UNIX, it is a wrapper around chmod(2) */
private static native void chmodImpl(String path, int mode) throws IOException;
public static void chmod(String path, int mode) throws IOException {
if (!Shell.WINDOWS) {
chmodImpl(path, mode);
} else {
try {
chmodImpl(path, mode);
} catch (NativeIOException nioe) {
if (nioe.getErrorCode() == 3) {
throw new NativeIOException("No such file or directory",
Errno.ENOENT);
} else {
LOG.warn(String.format("NativeIO.chmod error (%d): %s",
nioe.getErrorCode(), nioe.getMessage()));
throw new NativeIOException("Unknown error", Errno.UNKNOWN);
}
}
}
}
/** Wrapper around posix_fadvise(2) */
static native void posix_fadvise(
FileDescriptor fd, long offset, long len, int flags) throws NativeIOException;
/** Wrapper around sync_file_range(2) */
static native void sync_file_range(
FileDescriptor fd, long offset, long nbytes, int flags) throws NativeIOException;
/**
* Call posix_fadvise on the given file descriptor. See the manpage
* for this syscall for more information. On systems where this
* call is not available, does nothing.
*
* @throws NativeIOException if there is an error with the syscall
*/
static void posixFadviseIfPossible(String identifier,
FileDescriptor fd, long offset, long len, int flags)
throws NativeIOException {
if (nativeLoaded && fadvisePossible) {
try {
posix_fadvise(fd, offset, len, flags);
} catch (UnsupportedOperationException uoe) {
fadvisePossible = false;
} catch (UnsatisfiedLinkError ule) {
fadvisePossible = false;
}
}
}
/**
* Call sync_file_range on the given file descriptor. See the manpage
* for this syscall for more information. On systems where this
* call is not available, does nothing.
*
* @throws NativeIOException if there is an error with the syscall
*/
public static void syncFileRangeIfPossible(
FileDescriptor fd, long offset, long nbytes, int flags)
throws NativeIOException {
if (nativeLoaded && syncFileRangePossible) {
try {
sync_file_range(fd, offset, nbytes, flags);
} catch (UnsupportedOperationException uoe) {
syncFileRangePossible = false;
} catch (UnsatisfiedLinkError ule) {
syncFileRangePossible = false;
}
}
}
static native void mlock_native(
ByteBuffer buffer, long len) throws NativeIOException;
/**
* Locks the provided direct ByteBuffer into memory, preventing it from
* swapping out. After a buffer is locked, future accesses will not incur
* a page fault.
*
* See the mlock(2) man page for more information.
*
* @throws NativeIOException
*/
static void mlock(ByteBuffer buffer, long len)
throws IOException {
assertCodeLoaded();
if (!buffer.isDirect()) {
throw new IOException("Cannot mlock a non-direct ByteBuffer");
}
mlock_native(buffer, len);
}
/**
* Unmaps the block from memory. See munmap(2).
*
* There isn't any portable way to unmap a memory region in Java.
* So we use the sun.nio method here.
* Note that unmapping a memory region could cause crashes if code
* continues to reference the unmapped code. However, if we don't
* manually unmap the memory, we are dependent on the finalizer to
* do it, and we have no idea when the finalizer will run.
*
* @param buffer The buffer to unmap.
*/
public static void munmap(MappedByteBuffer buffer) {
if (buffer instanceof sun.nio.ch.DirectBuffer) {
sun.misc.Cleaner cleaner =
((sun.nio.ch.DirectBuffer)buffer).cleaner();
cleaner.clean();
}
}
/** Linux only methods used for getOwner() implementation */
private static native long getUIDforFDOwnerforOwner(FileDescriptor fd) throws IOException;
private static native String getUserName(long uid) throws IOException;
/**
* Result type of the fstat call
*/
public static class Stat {
private int ownerId, groupId;
private String owner, group;
private int mode;
// Mode constants
public static final int S_IFMT = 0170000; /* type of file */
public static final int S_IFIFO = 0010000; /* named pipe (fifo) */
public static final int S_IFCHR = 0020000; /* character special */
public static final int S_IFDIR = 0040000; /* directory */
public static final int S_IFBLK = 0060000; /* block special */
public static final int S_IFREG = 0100000; /* regular */
public static final int S_IFLNK = 0120000; /* symbolic link */
public static final int S_IFSOCK = 0140000; /* socket */
public static final int S_IFWHT = 0160000; /* whiteout */
public static final int S_ISUID = 0004000; /* set user id on execution */
public static final int S_ISGID = 0002000; /* set group id on execution */
public static final int S_ISVTX = 0001000; /* save swapped text even after use */
public static final int S_IRUSR = 0000400; /* read permission, owner */
public static final int S_IWUSR = 0000200; /* write permission, owner */
public static final int S_IXUSR = 0000100; /* execute/search permission, owner */
Stat(int ownerId, int groupId, int mode) {
this.ownerId = ownerId;
this.groupId = groupId;
this.mode = mode;
}
Stat(String owner, String group, int mode) {
if (!Shell.WINDOWS) {
this.owner = owner;
} else {
this.owner = stripDomain(owner);
}
if (!Shell.WINDOWS) {
this.group = group;
} else {
this.group = stripDomain(group);
}
this.mode = mode;
}
@Override
public String toString() {
return "Stat(owner='" + owner + "', group='" + group + "'" +
", mode=" + mode + ")";
}
public String getOwner() {
return owner;
}
public String getGroup() {
return group;
}
public int getMode() {
return mode;
}
}
/**
* Returns the file stat for a file descriptor.
*
* @param fd file descriptor.
* @return the file descriptor file stat.
* @throws IOException thrown if there was an IO error while obtaining the file stat.
*/
public static Stat getFstat(FileDescriptor fd) throws IOException {
Stat stat = null;
if (!Shell.WINDOWS) {
stat = fstat(fd);
stat.owner = getName(IdCache.USER, stat.ownerId);
stat.group = getName(IdCache.GROUP, stat.groupId);
} else {
try {
stat = fstat(fd);
} catch (NativeIOException nioe) {
if (nioe.getErrorCode() == 6) {
throw new NativeIOException("The handle is invalid.",
Errno.EBADF);
} else {
LOG.warn(String.format("NativeIO.getFstat error (%d): %s",
nioe.getErrorCode(), nioe.getMessage()));
throw new NativeIOException("Unknown error", Errno.UNKNOWN);
}
}
}
return stat;
}
private static String getName(IdCache domain, int id) throws IOException {
Map<Integer, CachedName> idNameCache = (domain == IdCache.USER)
? USER_ID_NAME_CACHE : GROUP_ID_NAME_CACHE;
String name;
CachedName cachedName = idNameCache.get(id);
long now = System.currentTimeMillis();
if (cachedName != null && (cachedName.timestamp + cacheTimeout) > now) {
name = cachedName.name;
} else {
name = (domain == IdCache.USER) ? getUserName(id) : getGroupName(id);
if (LOG.isDebugEnabled()) {
String type = (domain == IdCache.USER) ? "UserName" : "GroupName";
LOG.debug("Got " + type + " " + name + " for ID " + id +
" from the native implementation");
}
cachedName = new CachedName(name, now);
idNameCache.put(id, cachedName);
}
return name;
}
static native String getUserName(int uid) throws IOException;
static native String getGroupName(int uid) throws IOException;
private static class CachedName {
final long timestamp;
final String name;
public CachedName(String name, long timestamp) {
this.name = name;
this.timestamp = timestamp;
}
}
private static final Map<Integer, CachedName> USER_ID_NAME_CACHE =
new ConcurrentHashMap<Integer, CachedName>();
private static final Map<Integer, CachedName> GROUP_ID_NAME_CACHE =
new ConcurrentHashMap<Integer, CachedName>();
private enum IdCache { USER, GROUP }
public final static int MMAP_PROT_READ = 0x1;
public final static int MMAP_PROT_WRITE = 0x2;
public final static int MMAP_PROT_EXEC = 0x4;
public static native long mmap(FileDescriptor fd, int prot,
boolean shared, long length) throws IOException;
public static native void munmap(long addr, long length)
throws IOException;
}
private static boolean workaroundNonThreadSafePasswdCalls = false;
public static class Windows {
// Flags for CreateFile() call on Windows
public static final long GENERIC_READ = 0x80000000L;
public static final long GENERIC_WRITE = 0x40000000L;
public static final long FILE_SHARE_READ = 0x00000001L;
public static final long FILE_SHARE_WRITE = 0x00000002L;
public static final long FILE_SHARE_DELETE = 0x00000004L;
public static final long CREATE_NEW = 1;
public static final long CREATE_ALWAYS = 2;
public static final long OPEN_EXISTING = 3;
public static final long OPEN_ALWAYS = 4;
public static final long TRUNCATE_EXISTING = 5;
public static final long FILE_BEGIN = 0;
public static final long FILE_CURRENT = 1;
public static final long FILE_END = 2;
public static final long FILE_ATTRIBUTE_NORMAL = 0x00000080L;
/**
* Create a directory with permissions set to the specified mode. By setting
* permissions at creation time, we avoid issues related to the user lacking
* WRITE_DAC rights on subsequent chmod calls. One example where this can
* occur is writing to an SMB share where the user does not have Full Control
* rights, and therefore WRITE_DAC is denied.
*
* @param path directory to create
* @param mode permissions of new directory
* @throws IOException if there is an I/O error
*/
public static void createDirectoryWithMode(File path, int mode)
throws IOException {
createDirectoryWithMode0(path.getAbsolutePath(), mode);
}
/** Wrapper around CreateDirectory() on Windows */
private static native void createDirectoryWithMode0(String path, int mode)
throws NativeIOException;
/** Wrapper around CreateFile() on Windows */
public static native FileDescriptor createFile(String path,
long desiredAccess, long shareMode, long creationDisposition)
throws IOException;
/**
* Create a file for write with permissions set to the specified mode. By
* setting permissions at creation time, we avoid issues related to the user
* lacking WRITE_DAC rights on subsequent chmod calls. One example where
* this can occur is writing to an SMB share where the user does not have
* Full Control rights, and therefore WRITE_DAC is denied.
*
* This method mimics the semantics implemented by the JDK in
* {@link java.io.FileOutputStream}. The file is opened for truncate or
* append, the sharing mode allows other readers and writers, and paths
* longer than MAX_PATH are supported. (See io_util_md.c in the JDK.)
*
* @param path file to create
* @param append if true, then open file for append
* @param mode permissions of new directory
* @return FileOutputStream of opened file
* @throws IOException if there is an I/O error
*/
public static FileOutputStream createFileOutputStreamWithMode(File path,
boolean append, int mode) throws IOException {
long desiredAccess = GENERIC_WRITE;
long shareMode = FILE_SHARE_READ | FILE_SHARE_WRITE;
long creationDisposition = append ? OPEN_ALWAYS : CREATE_ALWAYS;
return new FileOutputStream(createFileWithMode0(path.getAbsolutePath(),
desiredAccess, shareMode, creationDisposition, mode));
}
/** Wrapper around CreateFile() with security descriptor on Windows */
private static native FileDescriptor createFileWithMode0(String path,
long desiredAccess, long shareMode, long creationDisposition, int mode)
throws NativeIOException;
/** Wrapper around SetFilePointer() on Windows */
public static native long setFilePointer(FileDescriptor fd,
long distanceToMove, long moveMethod) throws IOException;
/** Windows only methods used for getOwner() implementation */
private static native String getOwner(FileDescriptor fd) throws IOException;
/** Supported list of Windows access right flags */
public static enum AccessRight {
ACCESS_READ (0x0001), // FILE_READ_DATA
ACCESS_WRITE (0x0002), // FILE_WRITE_DATA
ACCESS_EXECUTE (0x0020); // FILE_EXECUTE
private final int accessRight;
AccessRight(int access) {
accessRight = access;
}
public int accessRight() {
return accessRight;
}
};
/** Windows only method used to check if the current process has requested
* access rights on the given path. */
private static native boolean access0(String path, int requestedAccess);
/**
* Checks whether the current process has desired access rights on
* the given path.
*
* Longer term this native function can be substituted with JDK7
* function Files#isReadable, isWritable, isExecutable.
*
* @param path input path
* @param desiredAccess ACCESS_READ, ACCESS_WRITE or ACCESS_EXECUTE
* @return true if access is allowed
* @throws IOException I/O exception on error
*/
public static boolean access(String path, AccessRight desiredAccess)
throws IOException {
return true;
// return access0(path, desiredAccess.accessRight());
}
/**
* Extends both the minimum and maximum working set size of the current
* process. This method gets the current minimum and maximum working set
* size, adds the requested amount to each and then sets the minimum and
* maximum working set size to the new values. Controlling the working set
* size of the process also controls the amount of memory it can lock.
*
* @param delta amount to increment minimum and maximum working set size
* @throws IOException for any error
* @see POSIX#mlock(ByteBuffer, long)
*/
public static native void extendWorkingSetSize(long delta) throws IOException;
static {
if (NativeCodeLoader.isNativeCodeLoaded()) {
try {
initNative();
nativeLoaded = true;
} catch (Throwable t) {
// This can happen if the user has an older version of libhadoop.so
// installed - in this case we can continue without native IO
// after warning
PerformanceAdvisory.LOG.debug("Unable to initialize NativeIO libraries", t);
}
}
}
}
private static final Log LOG = LogFactory.getLog(NativeIO.class);
private static boolean nativeLoaded = false;
static {
if (NativeCodeLoader.isNativeCodeLoaded()) {
try {
initNative();
nativeLoaded = true;
} catch (Throwable t) {
// This can happen if the user has an older version of libhadoop.so
// installed - in this case we can continue without native IO
// after warning
PerformanceAdvisory.LOG.debug("Unable to initialize NativeIO libraries", t);
}
}
}
/**
* Return true if the JNI-based native IO extensions are available.
*/
public static boolean isAvailable() {
return NativeCodeLoader.isNativeCodeLoaded() && nativeLoaded;
}
/** Initialize the JNI method ID and class ID cache */
private static native void initNative();
/**
* Get the maximum number of bytes that can be locked into memory at any
* given point.
*
* @return 0 if no bytes can be locked into memory;
* Long.MAX_VALUE if there is no limit;
* The number of bytes that can be locked into memory otherwise.
*/
static long getMemlockLimit() {
return isAvailable() ? getMemlockLimit0() : 0;
}
private static native long getMemlockLimit0();
/**
* @return the operating system's page size.
*/
static long getOperatingSystemPageSize() {
try {
Field f = Unsafe.class.getDeclaredField("theUnsafe");
f.setAccessible(true);
Unsafe unsafe = (Unsafe)f.get(null);
return unsafe.pageSize();
} catch (Throwable e) {
LOG.warn("Unable to get operating system page size. Guessing 4096.", e);
return 4096;
}
}
private static class CachedUid {
final long timestamp;
final String username;
public CachedUid(String username, long timestamp) {
this.timestamp = timestamp;
this.username = username;
}
}
private static final Map<Long, CachedUid> uidCache =
new ConcurrentHashMap<Long, CachedUid>();
private static long cacheTimeout;
private static boolean initialized = false;
/**
* The Windows logon name has two part, NetBIOS domain name and
* user account name, of the format DOMAIN\UserName. This method
* will remove the domain part of the full logon name.
*
* @param Fthe full principal name containing the domain
* @return name with domain removed
*/
private static String stripDomain(String name) {
int i = name.indexOf('\\');
if (i != -1)
name = name.substring(i + 1);
return name;
}
public static String getOwner(FileDescriptor fd) throws IOException {
ensureInitialized();
if (Shell.WINDOWS) {
String owner = Windows.getOwner(fd);
owner = stripDomain(owner);
return owner;
} else {
long uid = POSIX.getUIDforFDOwnerforOwner(fd);
CachedUid cUid = uidCache.get(uid);
long now = System.currentTimeMillis();
if (cUid != null && (cUid.timestamp + cacheTimeout) > now) {
return cUid.username;
}
String user = POSIX.getUserName(uid);
LOG.info("Got UserName " + user + " for UID " + uid
+ " from the native implementation");
cUid = new CachedUid(user, now);
uidCache.put(uid, cUid);
return user;
}
}
/**
* Create a FileInputStream that shares delete permission on the
* file opened, i.e. other process can delete the file the
* FileInputStream is reading. Only Windows implementation uses
* the native interface.
*/
public static FileInputStream getShareDeleteFileInputStream(File f)
throws IOException {
if (!Shell.WINDOWS) {
// On Linux the default FileInputStream shares delete permission
// on the file opened.
//
return new FileInputStream(f);
} else {
// Use Windows native interface to create a FileInputStream that
// shares delete permission on the file opened.
//
FileDescriptor fd = Windows.createFile(
f.getAbsolutePath(),
Windows.GENERIC_READ,
Windows.FILE_SHARE_READ |
Windows.FILE_SHARE_WRITE |
Windows.FILE_SHARE_DELETE,
Windows.OPEN_EXISTING);
return new FileInputStream(fd);
}
}
/**
* Create a FileInputStream that shares delete permission on the
* file opened at a given offset, i.e. other process can delete
* the file the FileInputStream is reading. Only Windows implementation
* uses the native interface.
*/
public static FileInputStream getShareDeleteFileInputStream(File f, long seekOffset)
throws IOException {
if (!Shell.WINDOWS) {
RandomAccessFile rf = new RandomAccessFile(f, "r");
if (seekOffset > 0) {
rf.seek(seekOffset);
}
return new FileInputStream(rf.getFD());
} else {
// Use Windows native interface to create a FileInputStream that
// shares delete permission on the file opened, and set it to the
// given offset.
//
FileDescriptor fd = NativeIO.Windows.createFile(
f.getAbsolutePath(),
NativeIO.Windows.GENERIC_READ,
NativeIO.Windows.FILE_SHARE_READ |
NativeIO.Windows.FILE_SHARE_WRITE |
NativeIO.Windows.FILE_SHARE_DELETE,
NativeIO.Windows.OPEN_EXISTING);
if (seekOffset > 0)
NativeIO.Windows.setFilePointer(fd, seekOffset, NativeIO.Windows.FILE_BEGIN);
return new FileInputStream(fd);
}
}
/**
* Create the specified File for write access, ensuring that it does not exist.
* @param f the file that we want to create
* @param permissions we want to have on the file (if security is enabled)
*
* @throws AlreadyExistsException if the file already exists
* @throws IOException if any other error occurred
*/
public static FileOutputStream getCreateForWriteFileOutputStream(File f, int permissions)
throws IOException {
if (!Shell.WINDOWS) {
// Use the native wrapper around open(2)
try {
FileDescriptor fd = NativeIO.POSIX.open(f.getAbsolutePath(),
NativeIO.POSIX.O_WRONLY | NativeIO.POSIX.O_CREAT
| NativeIO.POSIX.O_EXCL, permissions);
return new FileOutputStream(fd);
} catch (NativeIOException nioe) {
if (nioe.getErrno() == Errno.EEXIST) {
throw new AlreadyExistsException(nioe);
}
throw nioe;
}
} else {
// Use the Windows native APIs to create equivalent FileOutputStream
try {
FileDescriptor fd = NativeIO.Windows.createFile(f.getCanonicalPath(),
NativeIO.Windows.GENERIC_WRITE,
NativeIO.Windows.FILE_SHARE_DELETE
| NativeIO.Windows.FILE_SHARE_READ
| NativeIO.Windows.FILE_SHARE_WRITE,
NativeIO.Windows.CREATE_NEW);
NativeIO.POSIX.chmod(f.getCanonicalPath(), permissions);
return new FileOutputStream(fd);
} catch (NativeIOException nioe) {
if (nioe.getErrorCode() == 80) {
// ERROR_FILE_EXISTS
// 80 (0x50)
// The file exists
throw new AlreadyExistsException(nioe);
}
throw nioe;
}
}
}
private synchronized static void ensureInitialized() {
if (!initialized) {
cacheTimeout =
new Configuration().getLong("hadoop.security.uid.cache.secs",
4*60*60) * 1000;
LOG.info("Initialized cache for UID to User mapping with a cache" +
" timeout of " + cacheTimeout/1000 + " seconds.");
initialized = true;
}
}
/**
* A version of renameTo that throws a descriptive exception when it fails.
*
* @param src The source path
* @param dst The destination path
*
* @throws NativeIOException On failure.
*/
public static void renameTo(File src, File dst)
throws IOException {
if (!nativeLoaded) {
if (!src.renameTo(dst)) {
throw new IOException("renameTo(src=" + src + ", dst=" +
dst + ") failed.");
}
} else {
renameTo0(src.getAbsolutePath(), dst.getAbsolutePath());
}
}
public static void link(File src, File dst) throws IOException {
if (!nativeLoaded) {
HardLink.createHardLink(src, dst);
} else {
link0(src.getAbsolutePath(), dst.getAbsolutePath());
}
}
/**
* A version of renameTo that throws a descriptive exception when it fails.
*
* @param src The source path
* @param dst The destination path
*
* @throws NativeIOException On failure.
*/
private static native void renameTo0(String src, String dst)
throws NativeIOException;
private static native void link0(String src, String dst)
throws NativeIOException;
/**
* Unbuffered file copy from src to dst without tainting OS buffer cache
*
* In POSIX platform:
* It uses FileChannel#transferTo() which internally attempts
* unbuffered IO on OS with native sendfile64() support and falls back to
* buffered IO otherwise.
*
* It minimizes the number of FileChannel#transferTo call by passing the the
* src file size directly instead of a smaller size as the 3rd parameter.
* This saves the number of sendfile64() system call when native sendfile64()
* is supported. In the two fall back cases where sendfile is not supported,
* FileChannle#transferTo already has its own batching of size 8 MB and 8 KB,
* respectively.
*
* In Windows Platform:
* It uses its own native wrapper of CopyFileEx with COPY_FILE_NO_BUFFERING
* flag, which is supported on Windows Server 2008 and above.
*
* Ideally, we should use FileChannel#transferTo() across both POSIX and Windows
* platform. Unfortunately, the wrapper(Java_sun_nio_ch_FileChannelImpl_transferTo0)
* used by FileChannel#transferTo for unbuffered IO is not implemented on Windows.
* Based on OpenJDK 6/7/8 source code, Java_sun_nio_ch_FileChannelImpl_transferTo0
* on Windows simply returns IOS_UNSUPPORTED.
*
* Note: This simple native wrapper does minimal parameter checking before copy and
* consistency check (e.g., size) after copy.
* It is recommended to use wrapper function like
* the Storage#nativeCopyFileUnbuffered() function in hadoop-hdfs with pre/post copy
* checks.
*
* @param src The source path
* @param dst The destination path
* @throws IOException
*/
public static void copyFileUnbuffered(File src, File dst) throws IOException {
if (nativeLoaded && Shell.WINDOWS) {
copyFileUnbuffered0(src.getAbsolutePath(), dst.getAbsolutePath());
} else {
FileInputStream fis = null;
FileOutputStream fos = null;
FileChannel input = null;
FileChannel output = null;
try {
fis = new FileInputStream(src);
fos = new FileOutputStream(dst);
input = fis.getChannel();
output = fos.getChannel();
long remaining = input.size();
long position = 0;
long transferred = 0;
while (remaining > 0) {
transferred = input.transferTo(position, remaining, output);
remaining -= transferred;
position += transferred;
}
} finally {
IOUtils.cleanup(LOG, output);
IOUtils.cleanup(LOG, fos);
IOUtils.cleanup(LOG, input);
IOUtils.cleanup(LOG, fis);
}
}
}
private static native void copyFileUnbuffered0(String src, String dst)
throws NativeIOException;
}
然后应该就可以了
将程序放到hadoop集群运行
1.打包插件
<plugins>
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-jar-plugin</artifactId>
<configuration>
<archive>
<manifest>
<!-- 指定 mainclass,该配置将最终体现在jar包manifest文件中 -->
<mainClass>com.roadom.WordcountDriver</mainClass>
</manifest>
</archive>
<classesDirectory>
</classesDirectory>
</configuration>
</plugin>
</plugins>
1.打jar包
2.将jar包(不包括hadoop lib)上传到hadoop集群
D:\hadoopmr\out\artifacts\hadoopmr_jar\hadoopmr.jar
3.执行命令运行mr作业
./hadoop jar hadoopmr.jar com.neusoft.WordCount /wc.input /out42
可能遇到的问题
hadoop.dll 和 winutils.exe 添加到hadoop解压目录
如果双击winutils.exe弹出错误,需要安装
DirectX_Repair_3.7_Enhanced_XiaZaiBa.zip
更新电脑的dll
三、wordcount详解shuffle机制
十步
1.map前+map+suffer6步+reduce+文件写入
一、map方法执行之前
我们知道,HDFS里的文件是分块存放在Datanode上面的,而我们写的mapper程序也是跑在各个节点上的。这里就涉及到一个问题,哪一个节点上的mapper读哪一些节点上的文件块呢?hadoop会自动将这个文件分片(split),得到好多split,这每一个split放到一个节点的一个mapper里面去读。然后在每一台有mapper任务的节点上都执行了这么一个操作,将分得到的split切割成一行一行的键值对,然后传给map方法。键是这每一行在split中的偏移量,值是每一行得到的字符串。
二、执行map方法
写过wordcount的朋友都知道,这个过程就是读到每一行,切割字符串,生成键值对写出去。
三、shuffle操作(一)
这个过程是在有map任务的节点上完成的
1. partition
将得到的键值对按照一定的规则分组,例如例子中将首字母为a的全部分到一组,将首字母为b的分到一组。这里只是为了讲明白这个方式,进行了过程简化,实际不一定是分为两组,也不一定是按照首字母分组。
2. sort
对每一个组中的键值对根据键的哈希码排序。
3. combine
将具有相同键的键值对合成一个新的键值对,这个新的键值对的键是原来的键,键值是所有键的键值之和。
四、shuffle操作(二)
这个过程是在有reduce任务的节点上完成的。
1. 拉取partition
hadoop决定有多少个reducer的时候会规定有多少个partition,每一个reducer拉取自己要处理的那个分组的全部成员。例如,某台节点要处理所有以a开头的键值对,它就会将所有mapper中的以a开头的那一组全部拉取过来。
2. merge
在每一个reducer上,将具有相同键的键值对生成另外一个新的键值对,键是以前的键,键值是一个以前键值的集合。
3. sort
在每一台reducer节点上,将新生成的键值对进行排序,根据 哈希码值。
五、reduce操作
写过wordcount的朋友都知道,在reduce方法中,hadoop回传过来一个一个的键值对,键是每一个单词,键值就是四中新生成的键值对的键值。执行reduce操作,就是将每一个键值对中的键值累加起来。然后以键值对的形式将结果写出去。
六、文件写入HDFS
在每一台reducer节点上将文件写入,实际上是写成一个一个的文件块,但对外的表现形式是一整个大的结果文件。
四、map和reduce
maptask
一个job的map阶段并行度由客户端在提交job时决定
客户端对map阶段并行度的规划基本逻辑为:
一、将待处理的文件进行逻辑切片(根据处理数据文件的大小,划分多个split),然后每一个split分配一个maptask并行处理实例
二、具体切片规划是由FileInputFormat实现类的getSplits()方法完成
切分规则如下:
1.简单地按照文件的内容长度进行切片
2.切片大小默认是datanode的切块大小128M
3.切片时不是考虑一个整体数据集,而是针对每一个文件单独切片
比如待处理数据有两个文件:
file1.txt 200M
file2.txt 50M
经过FileInputFormat的切片机制运算后,形成的切片信息如下:
file1.txt.split1– 0~128M —–maptask
file1.txt.split2– 128M~200M —–maptask
file2.txt.split1– 0~50M —–maptask
三、如何改变切片大小(参数设置)
源码是通过这个方法来规划切片大小的
protected long computeSplitSize(long blockSize, long minSize,
long maxSize) {
return Math.max(minSize, Math.min(maxSize, blockSize));
}
minsize:默认值:1;配置参数: mapreduce.input.fileinputformat.split.minsize
maxsize:默认值:Long.MAXValue; 配置参数:mapreduce.input.fileinputformat.split.maxsize
blocksize:hdfs切片大小
调整切片大小结论:
maxsize(切片最大值):
参数如果调得比blocksize小,则会让切片变小,而且就等于配置的这个参数的值
minsize (切片最小值):
参数调的比blockSize大,则可以让切片变得比blocksize还大
控制map个数的核心源码
long minSize = Math.max(getFormatMinSplitSize(), getMinSplitSize(job));
//getFormatMinSplitSize 默认返回1,getMinSplitSize 为用户设置的最小分片数, 如果用户设置的大于1,则为用户设置的最小分片数
long maxSize = getMaxSplitSize(job);
//getMaxSplitSize为用户设置的最大分片数,默认最大为long 9223372036854775807L
long splitSize = computeSplitSize(blockSize, minSize,
maxSize);
protected long computeSplitSize(long blockSize, long minSize, long maxSize) {
return Math.max(minSize, Math.min(maxSize, blockSize));
}
由上述代码可以看出在
maxSize默认等于long(长整形)
blockSize默认在hadoop2.0之后为128M
minSize默认等于1
因此默认的切片大小splitSize等于128M也就是说等于块大小
一个切片对应于一个map任务,因此在默认情况下一个块对应于一个map任务。
要想人为控制map的个数可以从minSize和MaxSize入手。
想要增加map的个数,可以将maxSize调整小于blockSize;想要减小map的个数,可以调整minSize>blockSize。
具体调整可以在job配置中增加如下配置
FileInputFormat.setMinInputSplitSize(job, 301349250);//设置minSize
FileInputFormat.setMaxInputSplitSize(job, 10000);//设置maxSize
在实验中,
测试 文件大小 297M(311349250)
块大小128M
测试代码
FileInputFormat.setMinInputSplitSize(job, 301349250);
FileInputFormat.setMaxInputSplitSize(job, 10000);
测试后Map个数为1,由上面分片公式算出分片大小为301349250, 比 311349250小, 理论应该为两个map, 这是为什么呢?在上源码
while (bytesRemaining / splitSize > 1.1D) {
int blkIndex = getBlockIndex(blkLocations, length
- bytesRemaining);
splits.add(makeSplit(path, length - bytesRemaining,
splitSize, blkLocations[blkIndex].getHosts()));
bytesRemaining -= splitSize;
}
可以看出只要剩余的文件大小不超过分片大小的1.1倍, 则会分到一个分片中,避免开两个MAP, 其中一个运行数据太小,浪费资源。
总结,分片过程大概为,先遍历目标文件,过滤部分不符合要求的文件, 然后添加到列表,然后按照文件名来切分分片 (大小为前面计算分片大小的公式, 最后有个文件尾可能合并,其实常写网络程序的都知道), 然后添加到分片列表,然后每个分片读取自身对应的部分给MAP处理
reducetask
1、我们知道map的数量和文件数、文件大小、块大小、以及split大小有关,而reduce的数量跟哪些因素有关呢?
设置mapred.tasktracker.reduce.tasks.maximum的大小可以决定单个tasktracker一次性启动reduce的数目,但是不能决定总的reduce数目。
conf.setNumReduceTasks(4);JobConf对象的这个方法可以用来设定总的reduce的数目,看下Job Counters的统计:
Job Counters
Data-local map tasks=2
Total time spent by all maps waiting after reserving slots (ms)=0
Total time spent by all reduces waiting after reserving slots (ms)=0
SLOTS_MILLIS_MAPS=10695
SLOTS_MILLIS_REDUCES=29502
Launched map tasks=2
Launched reduce tasks=4
确实启动了4个reduce:看下输出:
diegoball@diegoball:~/IdeaProjects/test/build/classes$ hadoop fs -ls /user/diegoball/join_ou1123
11/03/25 15:28:45 INFO security.Groups: Group mapping impl=org.apache.hadoop.security.ShellBasedUnixGroupsMapping; cacheTimeout=300000
11/03/25 15:28:45 WARN conf.Configuration: mapred.task.id is deprecated. Instead, use mapreduce.task.attempt.id
Found 5 items
-rw-r--r-- 1 diegoball supergroup 0 2011-03-25 15:28 /user/diegoball/join_ou1123/_SUCCESS
-rw-r--r-- 1 diegoball supergroup 124 2011-03-25 15:27 /user/diegoball/join_ou1123/part-00000
-rw-r--r-- 1 diegoball supergroup 0 2011-03-25 15:27 /user/diegoball/join_ou1123/part-00001
-rw-r--r-- 1 diegoball supergroup 214 2011-03-25 15:28 /user/diegoball/join_ou1123/part-00002
-rw-r--r-- 1 diegoball supergroup 0 2011-03-25 15:28 /user/diegoball/join_ou1123/part-00003
只有2个reduce在干活。为什么呢?
shuffle的过程,需要根据key的值决定将这条<K,V> (map的输出),送到哪一个reduce中去。送到哪一个reduce中去靠调用默认的org.apache.hadoop.mapred.lib.HashPartitioner的getPartition()方法来实现。
HashPartitioner类:
package org.apache.hadoop.mapred.lib;
import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.classification.InterfaceStability;
import org.apache.hadoop.mapred.Partitioner;
import org.apache.hadoop.mapred.JobConf;
/** Partition keys by their {@link Object#hashCode()}.
*/
@InterfaceAudience.Public
@InterfaceStability.Stable
public class HashPartitioner<K2, V2> implements Partitioner<K2, V2> {
public void configure(JobConf job) {}
/** Use {@link Object#hashCode()} to partition. */
public int getPartition(K2 key, V2 value,
int numReduceTasks) {
return (key.hashCode() & Integer.MAX_VALUE) % numReduceTasks;
}
}
numReduceTasks的值在JobConf中可以设置。默认的是1:显然太小。
这也是为什么默认的设置中总启动一个reduce的原因。
返回与运算的结果和numReduceTasks求余。
Mapreduce根据这个返回结果决定将这条<K,V>,送到哪一个reduce中去。
key传入的是LongWritable类型,看下这个LongWritable类的hashcode()方法:
public int hashCode() {
return (int)value;
}
简简单单的返回了原值的整型值。
因为getPartition(K2 key, V2 value,int numReduceTask)返回的结果只有2个不同的值,所以最终只有2个reduce在干活。
HashPartitioner是默认的partition类,我们也可以自定义partition类 :
package com.alipay.dw.test;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.Partitioner;
/**
* Created by IntelliJ IDEA.
* User: diegoball
* Date: 11-3-10
* Time: 下午5:26
* To change this template use File | Settings | File Templates.
*/
public class MyPartitioner implements Partitioner<IntWritable, IntWritable> {
public int getPartition(IntWritable key, IntWritable value, int numPartitions) {
/* Pretty ugly hard coded partitioning function. Don't do that in practice, it is just for the sake of understanding. */
int nbOccurences = key.get();
if (nbOccurences > 20051210)
return 0;
else
return 1;
}
public void configure(JobConf arg0) {
}
}
仅仅需要覆盖getPartition()方法就OK。通过:
conf.setPartitionerClass(MyPartitioner.class);
可以设置自定义的partition类。
同样由于之返回2个不同的值0,1,不管conf.setNumReduceTasks(4);设置多少个reduce,也同样只会有2个reduce在干活。
由于每个reduce的输出key都是经过排序的,上述自定义的Partitioner还可以达到排序结果集的目的:
11/03/25 15:24:49 WARN conf.Configuration: mapred.task.id is deprecated. Instead, use mapreduce.task.attempt.id
Found 5 items
-rw-r--r-- 1 diegoball supergroup 0 2011-03-25 15:23 /user/diegoball/opt.del/_SUCCESS
-rw-r--r-- 1 diegoball supergroup 24546 2011-03-25 15:23 /user/diegoball/opt.del/part-00000
-rw-r--r-- 1 diegoball supergroup 10241 2011-03-25 15:23 /user/diegoball/opt.del/part-00001
-rw-r--r-- 1 diegoball supergroup 0 2011-03-25 15:23 /user/diegoball/opt.del/part-00002
-rw-r--r-- 1 diegoball supergroup 0 2011-03-25 15:23 /user/diegoball/opt.del/part-00003
part-00000和part-00001是这2个reduce的输出,由于使用了自定义的MyPartitioner,所有key小于20051210的的<K,V>都会放到第一个reduce中处理,key大于20051210就会被放到第二个reduce中处理。
每个reduce的输出key又是经过key排序的,所以最终的结果集降序排列。
但是如果使用上面自定义的partition类,又conf.setNumReduceTasks(1)的话,会怎样? 看下Job Counters:
Job Counters
Data-local map tasks=2
Total time spent by all maps waiting after reserving slots (ms)=0
Total time spent by all reduces waiting after reserving slots (ms)=0
SLOTS_MILLIS_MAPS=16395
SLOTS_MILLIS_REDUCES=3512
Launched map tasks=2
Launched reduce tasks=1
只启动了一个reduce。
(1)、 当setNumReduceTasks( int a) a=1(即默认值),不管Partitioner返回不同值的个数b为多少,只启动1个reduce,这种情况下自定义的Partitioner类没有起到任何作用。
(2)、 若a!=1:
a、当setNumReduceTasks( int a)里 a设置小于Partitioner返回不同值的个数b的话:
public int getPartition(IntWritable key, IntWritable value, int numPartitions) {
/* Pretty ugly hard coded partitioning function. Don't do that in practice, it is just for the sake of understanding. */
int nbOccurences = key.get();
if (nbOccurences < 20051210)
return 0;
if (nbOccurences >= 20051210 && nbOccurences < 20061210)
return 1;
if (nbOccurences >= 20061210 && nbOccurences < 20081210)
return 2;
else
return 3;
}
同时设置setNumReduceTasks( 2)。
于是抛出异常:
11/03/25 17:03:41 INFO mapreduce.Job: Task Id : attempt_201103241018_0023_m_000000_1, Status : FAILED
java.io.IOException: Illegal partition for 20110116 (3)
at org.apache.hadoop.mapred.MapTask$MapOutputBuffer.collect(MapTask.java:900)
at org.apache.hadoop.mapred.MapTask$OldOutputCollector.collect(MapTask.java:508)
at com.alipay.dw.test.KpiMapper.map(Unknown Source)
at com.alipay.dw.test.KpiMapper.map(Unknown Source)
at org.apache.hadoop.mapred.MapRunner.run(MapRunner.java:54)
at org.apache.hadoop.mapred.MapTask.runOldMapper(MapTask.java:397)
at org.apache.hadoop.mapred.MapTask.run(MapTask.java:330)
at org.apache.hadoop.mapred.Child$4.run(Child.java:217)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:396)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:742)
at org.apache.hadoop.mapred.Child.main(Child.java:211)
某些key没有找到所对应的reduce去处。原因是只启动了a个reduce。
b、当setNumReduceTasks( int a)里 a设置大于Partitioner返回不同值的个数b的话,同样会启动a个reduce,但是只有b个redurce上会得到数据。启动的其他的a-b个reduce浪费了。
c、理想状况是a=b,这样可以合理利用资源,负载更均衡。
yarn基本架构和工作机制
一、yarn基本架构
二、yarn工作机制
(0)Mr 程序提交到客户端所在的节点。
(1)Yarnrunner 向 Resourcemanager 申请一个 Application。
(2)rm 将该应用程序的资源路径返回给 yarnrunner。
(3)该程序将运行所需资源提交到 HDFS 上。
(4)程序资源提交完毕后,申请运行 mrAppMaster。
(5)RM 将用户的请求初始化成一个 task。
(6)其中一个 NodeManager 领取到 task 任务。
(7)该 NodeManager 创建容器 Container,并产生 MRAppmaster。
(8)Container 从 HDFS 上拷贝资源到本地。
(9)MRAppmaster 向 RM 申请运行 maptask 资源。
(10)RM 将运行 maptask 任务分配给另外两个 NodeManager,另两个 NodeManager 分
别领取任务并创建容器。
(11)MR 向两个接收到任务的 NodeManager 发送程序启动脚本,这两个 NodeManager
分别启动 maptask,maptask 对数据分区排序。
(12)MrAppMaster 等待所有 maptask 运行完毕后,向 RM 申请容器,运行 reduce task。
(13)reduce task 向 maptask 获取相应分区的数据。
(14)程序运行完毕后,MR 会向 RM 申请注销自己。