Google Git
Sign in
apache / hadoop / refs/heads/branch-0.21-old / . / mapreduce / src / docs / src / documentation / content / xdocs / mapred_tutorial.xml
blob: ffcdebc463bc4201cacc2a06492e89ca6facac9c [file] [log] [blame]
<?xml version="1.0"?>
<!--
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.
-->
<!DOCTYPE document PUBLIC "-//APACHE//DTD Documentation V2.0//EN" "http://forrest.apache.org/dtd/document-v20.dtd">
<document>
<header>
<title>MapReduce Tutorial</title>
</header>
<body>
<section>
<title>Purpose</title>
<p>This document comprehensively describes all user-facing facets of the
Hadoop MapReduce framework and serves as a tutorial.
</p>
</section>
<section>
<title>Prerequisites</title>
<p>Make sure Hadoop is installed, configured and running. See these guides:
</p>
<ul>
<li>
<a href="ext:single-node-setup">Single Node Setup</a> for first-time users.
</li>
<li>
<a href="cluster_setup.html">Cluster Setup</a> for large, distributed clusters.
</li>
</ul>
</section>
<section>
<title>Overview</title>
<p>Hadoop MapReduce is a software framework for easily writing
applications which process vast amounts of data (multi-terabyte data-sets)
in-parallel on large clusters (thousands of nodes) of commodity
hardware in a reliable, fault-tolerant manner.</p>
<p>A MapReduce <em>job</em> usually splits the input data-set into
independent chunks which are processed by the <em>map tasks</em> in a
completely parallel manner. The framework sorts the outputs of the maps,
which are then input to the <em>reduce tasks</em>. Typically both the
input and the output of the job are stored in a file-system. The framework
takes care of scheduling tasks, monitoring them and re-executes the failed
tasks.</p>
<p>Typically the compute nodes and the storage nodes are the same, that is,
the MapReduce framework and the
<a href="http://hadoop.apache.org/hdfs/docs/current/index.html">Hadoop Distributed File System</a> (HDFS)
are running on the same set of nodes. This configuration
allows the framework to effectively schedule tasks on the nodes where data
is already present, resulting in very high aggregate bandwidth across the
cluster.</p>
<p>The MapReduce framework consists of a single master
<code>JobTracker</code> and one slave <code>TaskTracker</code> per
cluster-node. The master is responsible for scheduling the jobs' component
tasks on the slaves, monitoring them and re-executing the failed tasks. The
slaves execute the tasks as directed by the master.</p>
<p>Minimally, applications specify the input/output locations and supply
<em>map</em> and <em>reduce</em> functions via implementations of
appropriate interfaces and/or abstract-classes. These, and other job
parameters, comprise the <em>job configuration</em>. The Hadoop
<em>job client</em> then submits the job (jar/executable etc.) and
configuration to the <code>JobTracker</code> which then assumes the
responsibility of distributing the software/configuration to the slaves,
scheduling tasks and monitoring them, providing status and diagnostic
information to the job-client.</p>
<p>Although the Hadoop framework is implemented in Java<sup>TM</sup>,
MapReduce applications need not be written in Java.</p>
<ul>
<li>
<a href="ext:api/org/apache/hadoop/streaming/package-summary">
Hadoop Streaming</a> is a utility which allows users to create and run
jobs with any executables (e.g. shell utilities) as the mapper and/or
the reducer.
</li>
<li>
<a href="ext:api/org/apache/hadoop/mapred/pipes/package-summary">
Hadoop Pipes</a> is a <a href="http://www.swig.org/">SWIG</a>-
compatible <em>C++ API</em> to implement MapReduce applications (non
JNI<sup>TM</sup> based).
</li>
</ul>
</section>
<section>
<title>Inputs and Outputs</title>
<p>The MapReduce framework operates exclusively on
<code>&lt;key, value&gt;</code> pairs, that is, the framework views the
input to the job as a set of <code>&lt;key, value&gt;</code> pairs and
produces a set of <code>&lt;key, value&gt;</code> pairs as the output of
the job, conceivably of different types.</p>
<p>The <code>key</code> and <code>value</code> classes have to be
serializable by the framework and hence need to implement the
<a href="ext:api/org/apache/hadoop/io/writable">Writable</a>
interface. Additionally, the <code>key</code> classes have to implement the
<a href="ext:api/org/apache/hadoop/io/writablecomparable">
WritableComparable</a> interface to facilitate sorting by the framework.
</p>
<p>Input and Output types of a MapReduce job:</p>
<p>
(input) <code>&lt;k1, v1&gt;</code>
-&gt;
<strong>map</strong>
-&gt;
<code>&lt;k2, v2&gt;</code>
-&gt;
<strong>combine</strong>
-&gt;
<code>&lt;k2, v2&gt;</code>
-&gt;
<strong>reduce</strong>
-&gt;
<code>&lt;k3, v3&gt;</code> (output)
</p>
</section>
<section>
<title>Example: WordCount v1.0</title>
<p>Before we jump into the details, lets walk through an example MapReduce
application to get a flavour for how they work.</p>
<p><code>WordCount</code> is a simple application that counts the number of
occurences of each word in a given input set.</p>
<p>This example works with a
pseudo-distributed (<a href="ext:single-node-setup">Single Node Setup</a>)
or fully-distributed (<a href="cluster_setup.html">Cluster Setup</a>)
Hadoop installation.</p>
<section>
<title>Source Code</title>
<table>
<tr>
<th></th>
<th>WordCount.java</th>
</tr>
<tr>
<td>1.</td>
<td>
<code>package org.myorg;</code>
</td>
</tr>
<tr>
<td>2.</td>
<td></td>
</tr>
<tr>
<td>3.</td>
<td>
<code>import java.io.IOException;</code>
</td>
</tr>
<tr>
<td>4.</td>
<td>
<code>import java.util.*;</code>
</td>
</tr>
<tr>
<td>5.</td>
<td></td>
</tr>
<tr>
<td>6.</td>
<td>
<code>import org.apache.hadoop.fs.Path;</code>
</td>
</tr>
<tr>
<td>7.</td>
<td>
<code>import org.apache.hadoop.conf.*;</code>
</td>
</tr>
<tr>
<td>8.</td>
<td>
<code>import org.apache.hadoop.io.*;</code>
</td>
</tr>
<tr>
<td>9.</td>
<td>
<code>import org.apache.hadoop.mapred.*;</code>
</td>
</tr>
<tr>
<td>10.</td>
<td>
<code>import org.apache.hadoop.util.*;</code>
</td>
</tr>
<tr>
<td>11.</td>
<td></td>
</tr>
<tr>
<td>12.</td>
<td>
<code>public class WordCount {</code>
</td>
</tr>
<tr>
<td>13.</td>
<td></td>
</tr>
<tr>
<td>14.</td>
<td>
&nbsp;&nbsp;
<code>
public static class Map extends MapReduceBase
implements Mapper&lt;LongWritable, Text, Text, IntWritable&gt; {
</code>
</td>
</tr>
<tr>
<td>15.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
private final static IntWritable one = new IntWritable(1);
</code>
</td>
</tr>
<tr>
<td>16.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private Text word = new Text();</code>
</td>
</tr>
<tr>
<td>17.</td>
<td></td>
</tr>
<tr>
<td>18.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
public void map(LongWritable key, Text value,
OutputCollector&lt;Text, IntWritable&gt; output,
Reporter reporter) throws IOException {
</code>
</td>
</tr>
<tr>
<td>19.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>String line = value.toString();</code>
</td>
</tr>
<tr>
<td>20.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>StringTokenizer tokenizer = new StringTokenizer(line);</code>
</td>
</tr>
<tr>
<td>21.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>while (tokenizer.hasMoreTokens()) {</code>
</td>
</tr>
<tr>
<td>22.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>word.set(tokenizer.nextToken());</code>
</td>
</tr>
<tr>
<td>23.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>output.collect(word, one);</code>
</td>
</tr>
<tr>
<td>24.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>25.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>26.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>27.</td>
<td></td>
</tr>
<tr>
<td>28.</td>
<td>
&nbsp;&nbsp;
<code>
public static class Reduce extends MapReduceBase implements
Reducer&lt;Text, IntWritable, Text, IntWritable&gt; {
</code>
</td>
</tr>
<tr>
<td>29.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
public void reduce(Text key, Iterator&lt;IntWritable&gt; values,
OutputCollector&lt;Text, IntWritable&gt; output,
Reporter reporter) throws IOException {
</code>
</td>
</tr>
<tr>
<td>30.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>int sum = 0;</code>
</td>
</tr>
<tr>
<td>31.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>while (values.hasNext()) {</code>
</td>
</tr>
<tr>
<td>32.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>sum += values.next().get();</code>
</td>
</tr>
<tr>
<td>33.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>34.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>output.collect(key, new IntWritable(sum));</code>
</td>
</tr>
<tr>
<td>35.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>36.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>37.</td>
<td></td>
</tr>
<tr>
<td>38.</td>
<td>
&nbsp;&nbsp;
<code>
public static void main(String[] args) throws Exception {
</code>
</td>
</tr>
<tr>
<td>39.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
JobConf conf = new JobConf(WordCount.class);
</code>
</td>
</tr>
<tr>
<td>40.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setJobName("wordcount");</code>
</td>
</tr>
<tr>
<td>41.</td>
<td></td>
</tr>
<tr>
<td>42.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setOutputKeyClass(Text.class);</code>
</td>
</tr>
<tr>
<td>43.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setOutputValueClass(IntWritable.class);</code>
</td>
</tr>
<tr>
<td>44.</td>
<td></td>
</tr>
<tr>
<td>45.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setMapperClass(Map.class);</code>
</td>
</tr>
<tr>
<td>46.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setCombinerClass(Reduce.class);</code>
</td>
</tr>
<tr>
<td>47.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setReducerClass(Reduce.class);</code>
</td>
</tr>
<tr>
<td>48.</td>
<td></td>
</tr>
<tr>
<td>49.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setInputFormat(TextInputFormat.class);</code>
</td>
</tr>
<tr>
<td>50.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setOutputFormat(TextOutputFormat.class);</code>
</td>
</tr>
<tr>
<td>51.</td>
<td></td>
</tr>
<tr>
<td>52.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>FileInputFormat.setInputPaths(conf, new Path(args[0]));</code>
</td>
</tr>
<tr>
<td>53.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>FileOutputFormat.setOutputPath(conf, new Path(args[1]));</code>
</td>
</tr>
<tr>
<td>54.</td>
<td></td>
</tr>
<tr>
<td>55.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>JobClient.runJob(conf);</code>
</td>
</tr>
<tr>
<td>57.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>58.</td>
<td>
<code>}</code>
</td>
</tr>
<tr>
<td>59.</td>
<td></td>
</tr>
</table>
</section>
<section>
<title>Usage</title>
<p>Assuming <code>HADOOP_HOME</code> is the root of the installation and
<code>HADOOP_VERSION</code> is the Hadoop version installed, compile
<code>WordCount.java</code> and create a jar:</p>
<p>
<code>$ mkdir wordcount_classes</code><br/>
<code>
$ javac -classpath ${HADOOP_HOME}/hadoop-${HADOOP_VERSION}-core.jar
-d wordcount_classes WordCount.java
</code><br/>
<code>$ jar -cvf /usr/joe/wordcount.jar -C wordcount_classes/ .</code>
</p>
<p>Assuming that:</p>
<ul>
<li>
<code>/usr/joe/wordcount/input</code> - input directory in HDFS
</li>
<li>
<code>/usr/joe/wordcount/output</code> - output directory in HDFS
</li>
</ul>
<p>Sample text-files as input:</p>
<p>
<code>$ bin/hadoop dfs -ls /usr/joe/wordcount/input/</code><br/>
<code>/usr/joe/wordcount/input/file01</code><br/>
<code>/usr/joe/wordcount/input/file02</code><br/>
<br/>
<code>$ bin/hadoop dfs -cat /usr/joe/wordcount/input/file01</code><br/>
<code>Hello World Bye World</code><br/>
<br/>
<code>$ bin/hadoop dfs -cat /usr/joe/wordcount/input/file02</code><br/>
<code>Hello Hadoop Goodbye Hadoop</code>
</p>
<p>Run the application:</p>
<p>
<code>
$ bin/hadoop jar /usr/joe/wordcount.jar org.myorg.WordCount
/usr/joe/wordcount/input /usr/joe/wordcount/output
</code>
</p>
<p>Output:</p>
<p>
<code>
$ bin/hadoop dfs -cat /usr/joe/wordcount/output/part-00000
</code>
<br/>
<code>Bye 1</code><br/>
<code>Goodbye 1</code><br/>
<code>Hadoop 2</code><br/>
<code>Hello 2</code><br/>
<code>World 2</code><br/>
</p>
<p> Applications can specify a comma separated list of paths which
would be present in the current working directory of the task
using the option <code>-files</code>. The <code>-libjars</code>
option allows applications to add jars to the classpaths of the maps
and reduces. The option <code>-archives</code> allows them to pass
comma separated list of archives as arguments. These archives are
unarchived and a link with name of the archive is created in
the current working directory of tasks. More
details about the command line options are available at
<a href="commands_manual.html"> Hadoop Commands Guide.</a></p>
<p>Running <code>wordcount</code> example with
<code>-libjars</code>, <code>-files</code> and <code>-archives</code>:
<br/>
<code> hadoop jar hadoop-examples.jar wordcount -files cachefile.txt
-libjars mylib.jar -archives myarchive.zip input output </code>
Here, myarchive.zip will be placed and unzipped into a directory
by the name "myarchive.zip"
</p>
<p> Users can specify a different symbolic name for
files and archives passed through -files and -archives option, using #.
</p>
<p> For example,
<code> hadoop jar hadoop-examples.jar wordcount
-files dir1/dict.txt#dict1,dir2/dict.txt#dict2
-archives mytar.tgz#tgzdir input output </code>
Here, the files dir1/dict.txt and dir2/dict.txt can be accessed by
tasks using the symbolic names dict1 and dict2 respectively.
And the archive mytar.tgz will be placed and unarchived into a
directory by the name tgzdir
</p>
</section>
<section>
<title>Walk-through</title>
<p>The <code>WordCount</code> application is quite straight-forward.</p>
<p>The <code>Mapper</code> implementation (lines 14-26), via the
<code>map</code> method (lines 18-25), processes one line at a time,
as provided by the specified <code>TextInputFormat</code> (line 49).
It then splits the line into tokens separated by whitespaces, via the
<code>StringTokenizer</code>, and emits a key-value pair of
<code>&lt; &lt;word&gt;, 1&gt;</code>.</p>
<p>
For the given sample input the first map emits:<br/>
<code>&lt; Hello, 1&gt;</code><br/>
<code>&lt; World, 1&gt;</code><br/>
<code>&lt; Bye, 1&gt;</code><br/>
<code>&lt; World, 1&gt;</code><br/>
</p>
<p>
The second map emits:<br/>
<code>&lt; Hello, 1&gt;</code><br/>
<code>&lt; Hadoop, 1&gt;</code><br/>
<code>&lt; Goodbye, 1&gt;</code><br/>
<code>&lt; Hadoop, 1&gt;</code><br/>
</p>
<p>We'll learn more about the number of maps spawned for a given job, and
how to control them in a fine-grained manner, a bit later in the
tutorial.</p>
<p><code>WordCount</code> also specifies a <code>combiner</code> (line
46). Hence, the output of each map is passed through the local combiner
(which is same as the <code>Reducer</code> as per the job
configuration) for local aggregation, after being sorted on the
<em>key</em>s.</p>
<p>
The output of the first map:<br/>
<code>&lt; Bye, 1&gt;</code><br/>
<code>&lt; Hello, 1&gt;</code><br/>
<code>&lt; World, 2&gt;</code><br/>
</p>
<p>
The output of the second map:<br/>
<code>&lt; Goodbye, 1&gt;</code><br/>
<code>&lt; Hadoop, 2&gt;</code><br/>
<code>&lt; Hello, 1&gt;</code><br/>
</p>
<p>The <code>Reducer</code> implementation (lines 28-36), via the
<code>reduce</code> method (lines 29-35) just sums up the values,
which are the occurence counts for each key (i.e. words in this example).
</p>
<p>
Thus the output of the job is:<br/>
<code>&lt; Bye, 1&gt;</code><br/>
<code>&lt; Goodbye, 1&gt;</code><br/>
<code>&lt; Hadoop, 2&gt;</code><br/>
<code>&lt; Hello, 2&gt;</code><br/>
<code>&lt; World, 2&gt;</code><br/>
</p>
<p>The <code>run</code> method specifies various facets of the job, such
as the input/output paths (passed via the command line), key/value
types, input/output formats etc., in the <code>JobConf</code>.
It then calls the <code>JobClient.runJob</code> (line 55) to submit the
and monitor its progress.</p>
<p>We'll learn more about <code>JobConf</code>, <code>JobClient</code>,
<code>Tool</code> and other interfaces and classes a bit later in the
tutorial.</p>
</section>
</section>
<section>
<title>MapReduce - User Interfaces</title>
<p>This section provides a reasonable amount of detail on every user-facing
aspect of the MapReduce framwork. This should help users implement,
configure and tune their jobs in a fine-grained manner. However, please
note that the javadoc for each class/interface remains the most
comprehensive documentation available; this is only meant to be a tutorial.
</p>
<p>Let us first take the <code>Mapper</code> and <code>Reducer</code>
interfaces. Applications typically implement them to provide the
<code>map</code> and <code>reduce</code> methods.</p>
<p>We will then discuss other core interfaces including
<code>JobConf</code>, <code>JobClient</code>, <code>Partitioner</code>,
<code>OutputCollector</code>, <code>Reporter</code>,
<code>InputFormat</code>, <code>OutputFormat</code>,
<code>OutputCommitter</code> and others.</p>
<p>Finally, we will wrap up by discussing some useful features of the
framework such as the <code>DistributedCache</code>,
<code>IsolationRunner</code> etc.</p>
<section>
<title>Payload</title>
<p>Applications typically implement the <code>Mapper</code> and
<code>Reducer</code> interfaces to provide the <code>map</code> and
<code>reduce</code> methods. These form the core of the job.</p>
<section>
<title>Mapper</title>
<p><a href="ext:api/org/apache/hadoop/mapred/mapper">
Mapper</a> maps input key/value pairs to a set of intermediate
key/value pairs.</p>
<p>Maps are the individual tasks that transform input records into
intermediate records. The transformed intermediate records do not need
to be of the same type as the input records. A given input pair may
map to zero or many output pairs.</p>
<p>The Hadoop MapReduce framework spawns one map task for each
<code>InputSplit</code> generated by the <code>InputFormat</code> for
the job.</p>
<p>Overall, <code>Mapper</code> implementations are passed the
<code>JobConf</code> for the job via the
<a href="ext:api/org/apache/hadoop/mapred/jobconfigurable/configure">
JobConfigurable.configure(JobConf)</a> method and override it to
initialize themselves. The framework then calls
<a href="ext:api/org/apache/hadoop/mapred/mapper/map">
map(WritableComparable, Writable, OutputCollector, Reporter)</a> for
each key/value pair in the <code>InputSplit</code> for that task.
Applications can then override the
<a href="ext:api/org/apache/hadoop/io/closeable/close">
Closeable.close()</a> method to perform any required cleanup.</p>
<p>Output pairs do not need to be of the same types as input pairs. A
given input pair may map to zero or many output pairs. Output pairs
are collected with calls to
<a href="ext:api/org/apache/hadoop/mapred/outputcollector/collect">
OutputCollector.collect(WritableComparable,Writable)</a>.</p>
<p>Applications can use the <code>Reporter</code> to report
progress, set application-level status messages and update
<code>Counters</code>, or just indicate that they are alive.</p>
<p>All intermediate values associated with a given output key are
subsequently grouped by the framework, and passed to the
<code>Reducer</code>(s) to determine the final output. Users can
control the grouping by specifying a <code>Comparator</code> via
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setoutputkeycomparatorclass">
JobConf.setOutputKeyComparatorClass(Class)</a>.</p>
<p>The <code>Mapper</code> outputs are sorted and then
partitioned per <code>Reducer</code>. The total number of partitions is
the same as the number of reduce tasks for the job. Users can control
which keys (and hence records) go to which <code>Reducer</code> by
implementing a custom <code>Partitioner</code>.</p>
<p>Users can optionally specify a <code>combiner</code>, via
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setcombinerclass">
JobConf.setCombinerClass(Class)</a>, to perform local aggregation of
the intermediate outputs, which helps to cut down the amount of data
transferred from the <code>Mapper</code> to the <code>Reducer</code>.
</p>
<p>The intermediate, sorted outputs are always stored in a simple
(key-len, key, value-len, value) format.
Applications can control if, and how, the
intermediate outputs are to be compressed and the
<a href="ext:api/org/apache/hadoop/io/compress/compressioncodec">
CompressionCodec</a> to be used via the <code>JobConf</code>.
</p>
<section>
<title>How Many Maps?</title>
<p>The number of maps is usually driven by the total size of the
inputs, that is, the total number of blocks of the input files.</p>
<p>The right level of parallelism for maps seems to be around 10-100
maps per-node, although it has been set up to 300 maps for very
cpu-light map tasks. Task setup takes awhile, so it is best if the
maps take at least a minute to execute.</p>
<p>Thus, if you expect 10TB of input data and have a blocksize of
<code>128MB</code>, you'll end up with 82,000 maps, unless
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setnummaptasks">
setNumMapTasks(int)</a> (which only provides a hint to the framework)
is used to set it even higher.</p>
</section>
</section>
<section>
<title>Reducer</title>
<p><a href="ext:api/org/apache/hadoop/mapred/reducer">
Reducer</a> reduces a set of intermediate values which share a key to
a smaller set of values.</p>
<p>The number of reduces for the job is set by the user
via <a href="ext:api/org/apache/hadoop/mapred/jobconf/setnumreducetasks">
JobConf.setNumReduceTasks(int)</a>.</p>
<p>Overall, <code>Reducer</code> implementations are passed the
<code>JobConf</code> for the job via the
<a href="ext:api/org/apache/hadoop/mapred/jobconfigurable/configure">
JobConfigurable.configure(JobConf)</a> method and can override it to
initialize themselves. The framework then calls
<a href="ext:api/org/apache/hadoop/mapred/reducer/reduce">
reduce(WritableComparable, Iterator, OutputCollector, Reporter)</a>
method for each <code>&lt;key, (list of values)&gt;</code>
pair in the grouped inputs. Applications can then override the
<a href="ext:api/org/apache/hadoop/io/closeable/close">
Closeable.close()</a> method to perform any required cleanup.</p>
<p><code>Reducer</code> has 3 primary phases: shuffle, sort and reduce.
</p>
<section>
<title>Shuffle</title>
<p>Input to the <code>Reducer</code> is the sorted output of the
mappers. In this phase the framework fetches the relevant partition
of the output of all the mappers, via HTTP.</p>
</section>
<section>
<title>Sort</title>
<p>The framework groups <code>Reducer</code> inputs by keys (since
different mappers may have output the same key) in this stage.</p>
<p>The shuffle and sort phases occur simultaneously; while
map-outputs are being fetched they are merged.</p>
<section>
<title>Secondary Sort</title>
<p>If equivalence rules for grouping the intermediate keys are
required to be different from those for grouping keys before
reduction, then one may specify a <code>Comparator</code> via
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setoutputvaluegroupingcomparator">
JobConf.setOutputValueGroupingComparator(Class)</a>. Since
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setoutputkeycomparatorclass">
JobConf.setOutputKeyComparatorClass(Class)</a> can be used to
control how intermediate keys are grouped, these can be used in
conjunction to simulate <em>secondary sort on values</em>.</p>
</section>
</section>
<section>
<title>Reduce</title>
<p>In this phase the
<a href="ext:api/org/apache/hadoop/mapred/reducer/reduce">
reduce(WritableComparable, Iterator, OutputCollector, Reporter)</a>
method is called for each <code>&lt;key, (list of values)&gt;</code>
pair in the grouped inputs.</p>
<p>The output of the reduce task is typically written to the
<a href="ext:api/org/apache/hadoop/fs/filesystem">
FileSystem</a> via
<a href="ext:api/org/apache/hadoop/mapred/outputcollector/collect">
OutputCollector.collect(WritableComparable, Writable)</a>.</p>
<p>Applications can use the <code>Reporter</code> to report
progress, set application-level status messages and update
<code>Counters</code>, or just indicate that they are alive.</p>
<p>The output of the <code>Reducer</code> is <em>not sorted</em>.</p>
</section>
<section>
<title>How Many Reduces?</title>
<p>The right number of reduces seems to be <code>0.95</code> or
<code>1.75</code> multiplied by (&lt;<em>no. of nodes</em>&gt; *
<code>mapreduce.tasktracker.reduce.tasks.maximum</code>).</p>
<p>With <code>0.95</code> all of the reduces can launch immediately
and start transfering map outputs as the maps finish. With
<code>1.75</code> the faster nodes will finish their first round of
reduces and launch a second wave of reduces doing a much better job
of load balancing.</p>
<p>Increasing the number of reduces increases the framework overhead,
but increases load balancing and lowers the cost of failures.</p>
<p>The scaling factors above are slightly less than whole numbers to
reserve a few reduce slots in the framework for speculative-tasks and
failed tasks.</p>
</section>
<section>
<title>Reducer NONE</title>
<p>It is legal to set the number of reduce-tasks to <em>zero</em> if
no reduction is desired.</p>
<p>In this case the outputs of the map-tasks go directly to the
<code>FileSystem</code>, into the output path set by
<a href="ext:api/org/apache/hadoop/mapred/fileoutputformat/setoutputpath">
setOutputPath(Path)</a>. The framework does not sort the
map-outputs before writing them out to the <code>FileSystem</code>.
</p>
</section>
<section>
<title>Mark-Reset</title>
<p>While applications iterate through the values for a given key, it is
possible to mark the current position and later reset the iterator to
this position and continue the iteration process. The corresponding
methods are <code>mark()</code> and <code>reset()</code>.
</p>
<p><code>mark()</code> and <code>reset()</code> can be called any
number of times during the iteration cycle. The <code>reset()</code>
method will reset the iterator to the last record before a call to
the previous <code>mark()</code>.
</p>
<p>This functionality is available only with the new context based
reduce iterator.
</p>
<p> The following code snippet demonstrates the use of this
functionality.
</p>
<section>
<title>Source Code</title>
<table>
<tr><td>
<code>
public void reduce(IntWritable key,
Iterable&lt;IntWritable&gt; values,
Context context) throws IOException, InterruptedException {
</code>
</td></tr>
<tr><td></td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
MarkableIterator&lt;IntWritable&gt; mitr =
new MarkableIterator&lt;IntWritable&gt;(values.iterator());
</code>
</td></tr>
<tr><td></td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
// Mark the position
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
values.mark();
</code>
</td></tr>
<tr><td></td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
while (values.hasNext()) {
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;&nbsp;&nbsp;
i = values.next();
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;&nbsp;&nbsp;
// Do the necessary processing
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
}
</code>
</td></tr>
<tr><td></td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
// Reset
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
values.reset();
</code>
</td></tr>
<tr><td></td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
// Iterate all over again. Since mark was called before the first
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
// call to values.next() in this example, we will iterate over all
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
// the values now
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
while (values.hasNext()) {
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;&nbsp;&nbsp;
i = values.next();
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;&nbsp;&nbsp;
// Do the necessary processing
</code>
</td></tr>
<tr><td>
<code>
&nbsp;&nbsp;
}
</code>
</td></tr>
<tr><td></td></tr>
<tr><td>
<code>
}
</code>
</td></tr>
</table>
</section>
</section>
</section>
<section>
<title>Partitioner</title>
<p><a href="ext:api/org/apache/hadoop/mapred/partitioner">
Partitioner</a> partitions the key space.</p>
<p>Partitioner controls the partitioning of the keys of the
intermediate map-outputs. The key (or a subset of the key) is used to
derive the partition, typically by a <em>hash function</em>. The total
number of partitions is the same as the number of reduce tasks for the
job. Hence this controls which of the <code>m</code> reduce tasks the
intermediate key (and hence the record) is sent to for reduction.</p>
<p><a href="ext:api/org/apache/hadoop/mapred/lib/hashpartitioner">
HashPartitioner</a> is the default <code>Partitioner</code>.</p>
</section>
<section>
<title>Reporter</title>
<p><a href="ext:api/org/apache/hadoop/mapred/reporter">
Reporter</a> is a facility for MapReduce applications to report
progress, set application-level status messages and update
<code>Counters</code>.</p>
<p><code>Mapper</code> and <code>Reducer</code> implementations can use
the <code>Reporter</code> to report progress or just indicate
that they are alive. In scenarios where the application takes a
significant amount of time to process individual key/value pairs,
this is crucial since the framework might assume that the task has
timed-out and kill that task. Another way to avoid this is to
set the configuration parameter <code>mapreduce.task.timeout</code> to a
high-enough value (or even set it to <em>zero</em> for no time-outs).
</p>
<p>Applications can also update <code>Counters</code> using the
<code>Reporter</code>.</p>
</section>
<section>
<title>OutputCollector</title>
<p><a href="ext:api/org/apache/hadoop/mapred/outputcollector">
OutputCollector</a> is a generalization of the facility provided by
the MapReduce framework to collect data output by the
<code>Mapper</code> or the <code>Reducer</code> (either the
intermediate outputs or the output of the job).</p>
</section>
<p>Hadoop MapReduce comes bundled with a
<a href="ext:api/org/apache/hadoop/mapred/lib/package-summary">
library</a> of generally useful mappers, reducers, and partitioners.</p>
</section>
<section>
<title>Job Configuration</title>
<p><a href="ext:api/org/apache/hadoop/mapred/jobconf">
JobConf</a> represents a MapReduce job configuration.</p>
<p><code>JobConf</code> is the primary interface for a user to describe
a MapReduce job to the Hadoop framework for execution. The framework
tries to faithfully execute the job as described by <code>JobConf</code>,
however:</p>
<ul>
<li>f
Some configuration parameters may have been marked as
<a href="ext:api/org/apache/hadoop/conf/configuration/final_parameters">
final</a> by administrators and hence cannot be altered.
</li>
<li>
While some job parameters are straight-forward to set (e.g.
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setnumreducetasks">
setNumReduceTasks(int)</a>), other parameters interact subtly with
the rest of the framework and/or job configuration and are
more complex to set (e.g.
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setnummaptasks">
setNumMapTasks(int)</a>).
</li>
</ul>
<p><code>JobConf</code> is typically used to specify the
<code>Mapper</code>, combiner (if any), <code>Partitioner</code>,
<code>Reducer</code>, <code>InputFormat</code>,
<code>OutputFormat</code> and <code>OutputCommitter</code>
implementations. <code>JobConf</code> also
indicates the set of input files
(<a href="ext:api/org/apache/hadoop/mapred/fileinputformat/setinputpaths">setInputPaths(JobConf, Path...)</a>
/<a href="ext:api/org/apache/hadoop/mapred/fileinputformat/addinputpath">addInputPath(JobConf, Path)</a>)
and (<a href="ext:api/org/apache/hadoop/mapred/fileinputformat/setinputpathstring">setInputPaths(JobConf, String)</a>
/<a href="ext:api/org/apache/hadoop/mapred/fileinputformat/addinputpathstring">addInputPaths(JobConf, String)</a>)
and where the output files should be written
(<a href="ext:api/org/apache/hadoop/mapred/fileoutputformat/setoutputpath">setOutputPath(Path)</a>).</p>
<p>Optionally, <code>JobConf</code> is used to specify other advanced
facets of the job such as the <code>Comparator</code> to be used, files
to be put in the <code>DistributedCache</code>, whether intermediate
and/or job outputs are to be compressed (and how), debugging via
user-provided scripts
(<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmapdebugscript">setMapDebugScript(String)</a>/<a href="ext:api/org/apache/hadoop/mapred/jobconf/setreducedebugscript">setReduceDebugScript(String)</a>)
, whether job tasks can be executed in a <em>speculative</em> manner
(<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmapspeculativeexecution">setMapSpeculativeExecution(boolean)</a>)/(<a href="ext:api/org/apache/hadoop/mapred/jobconf/setreducespeculativeexecution">setReduceSpeculativeExecution(boolean)</a>)
, maximum number of attempts per task
(<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmaxmapattempts">setMaxMapAttempts(int)</a>/<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmaxreduceattempts">setMaxReduceAttempts(int)</a>)
, percentage of tasks failure which can be tolerated by the job
(<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmaxmaptaskfailurespercent">setMaxMapTaskFailuresPercent(int)</a>/<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmaxreducetaskfailurespercent">setMaxReduceTaskFailuresPercent(int)</a>)
etc.</p>
<p>Of course, users can use
<a href="ext:api/org/apache/hadoop/conf/configuration/set">set(String, String)</a>/<a href="ext:api/org/apache/hadoop/conf/configuration/get">get(String, String)</a>
to set/get arbitrary parameters needed by applications. However, use the
<code>DistributedCache</code> for large amounts of (read-only) data.</p>
</section>
<section>
<title>Task Execution &amp; Environment</title>
<p>The <code>TaskTracker</code> executes the <code>Mapper</code>/
<code>Reducer</code> <em>task</em> as a child process in a separate jvm.
</p>
<p>The child-task inherits the environment of the parent
<code>TaskTracker</code>. The user can specify additional options to the
child-jvm via the <code>mapred.{map|reduce}.child.java.opts</code>
configuration parameter in the <code>JobConf</code> such as non-standard
paths for the run-time linker to search shared libraries via
<code>-Djava.library.path=&lt;&gt;</code> etc. If the
<code>mapred.{map|reduce}.child.java.opts</code> parameters contains the
symbol <em>@taskid@</em> it is interpolated with value of
<code>taskid</code> of the MapReduce task.</p>
<p>Here is an example with multiple arguments and substitutions,
showing jvm GC logging, and start of a passwordless JVM JMX agent so that
it can connect with jconsole and the likes to watch child memory,
threads and get thread dumps. It also sets the maximum heap-size of the
map and reduce child jvm to 512MB &amp; 1024MB respectively. It also
adds an additional path to the <code>java.library.path</code> of the
child-jvm.</p>
<p>
<code>&lt;property&gt;</code><br/>
&nbsp;&nbsp;<code>&lt;name&gt;mapreduce.map.java.opts&lt;/name&gt;</code><br/>
&nbsp;&nbsp;<code>&lt;value&gt;</code><br/>
&nbsp;&nbsp;&nbsp;&nbsp;<code>
-Xmx512M -Djava.library.path=/home/mycompany/lib
-verbose:gc -Xloggc:/tmp/@taskid@.gc</code><br/>
&nbsp;&nbsp;&nbsp;&nbsp;<code>
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false</code><br/>
&nbsp;&nbsp;<code>&lt;/value&gt;</code><br/>
<code>&lt;/property&gt;</code>
</p>
<p>
<code>&lt;property&gt;</code><br/>
&nbsp;&nbsp;<code>&lt;name&gt;mapreduce.reduce.java.opts&lt;/name&gt;</code><br/>
&nbsp;&nbsp;<code>&lt;value&gt;</code><br/>
&nbsp;&nbsp;&nbsp;&nbsp;<code>
-Xmx1024M -Djava.library.path=/home/mycompany/lib
-verbose:gc -Xloggc:/tmp/@taskid@.gc</code><br/>
&nbsp;&nbsp;&nbsp;&nbsp;<code>
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false</code><br/>
&nbsp;&nbsp;<code>&lt;/value&gt;</code><br/>
<code>&lt;/property&gt;</code>
</p>
<section>
<title> Memory Management</title>
<p>Users/admins can also specify the maximum virtual memory
of the launched child-task, and any sub-process it launches
recursively, using <code>mapred.{map|reduce}.child.ulimit</code>. Note
that the value set here is a per process limit.
The value for <code>mapred.{map|reduce}.child.ulimit</code> should be
specified in kilo bytes (KB). And also the value must be greater than
or equal to the -Xmx passed to JavaVM, else the VM might not start.
</p>
<p>Note: <code>mapred.{map|reduce}.child.java.opts</code> are used only
for configuring the launched child tasks from task tracker. Configuring
the memory options for daemons is documented under
<a href="cluster_setup.html#Configuring+the+Environment+of+the+Hadoop+Daemons">
Configuring the Environment of the Hadoop Daemons</a> (Cluster Setup).</p>
<p>The memory available to some parts of the framework is also
configurable. In map and reduce tasks, performance may be influenced
by adjusting parameters influencing the concurrency of operations and
the frequency with which data will hit disk. Monitoring the filesystem
counters for a job- particularly relative to byte counts from the map
and into the reduce- is invaluable to the tuning of these
parameters.</p>
<p>Users can choose to override default limits of Virtual Memory and RAM
enforced by the task tracker, if memory management is enabled.
Users can set the following parameter per job:</p>
<table>
<tr><th>Name</th><th>Type</th><th>Description</th></tr>
<tr><td><code>mapred.task.maxvmem</code></td><td>int</td>
<td>A number, in bytes, that represents the maximum Virtual Memory
task-limit for each task of the job. A task will be killed if
it consumes more Virtual Memory than this number.
</td></tr>
<tr><td>mapred.task.maxpmem</td><td>int</td>
<td>A number, in bytes, that represents the maximum RAM task-limit
for each task of the job. This number can be optionally used by
Schedulers to prevent over-scheduling of tasks on a node based
on RAM needs.
</td></tr>
</table>
</section>
<section>
<title>Map Parameters</title>
<p>A record emitted from a map will be serialized into a buffer and
metadata will be stored into accounting buffers. As described in the
following options, when either the serialization buffer or the
metadata exceed a threshold, the contents of the buffers will be
sorted and written to disk in the background while the map continues
to output records. If either buffer fills completely while the spill
is in progress, the map thread will block. When the map is finished,
any remaining records are written to disk and all on-disk segments
are merged into a single file. Minimizing the number of spills to
disk can decrease map time, but a larger buffer also decreases the
memory available to the mapper.</p>
<table>
<tr><th>Name</th><th>Type</th><th>Description</th></tr>
<tr><td>mapreduce.task.io.sort.mb</td><td>int</td>
<td>The cumulative size of the serialization and accounting
buffers storing records emitted from the map, in megabytes.
</td></tr>
<tr><td>mapreduce.map.sort.record.percent</td><td>float</td>
<td>The ratio of serialization to accounting space can be
adjusted. Each serialized record requires 16 bytes of
accounting information in addition to its serialized size to
effect the sort. This percentage of space allocated from
<code>mapreduce.task.io.sort.mb</code> affects the
probability of a spill to
disk being caused by either exhaustion of the serialization
buffer or the accounting space. Clearly, for a map outputting
small records, a higher value than the default will likely
decrease the number of spills to disk.</td></tr>
<tr><td>mapreduce.map.sort.spill.percent</td><td>float</td>
<td>This is the threshold for the accounting and serialization
buffers. When this percentage of either buffer has filled,
their contents will be spilled to disk in the background. Let
<code>mapreduce.map.sort.record.percent</code> be <em>r</em>,
<code>mapreduce.task.io.sort.mb</code> be <em>x</em>,
and this value be
<em>q</em>. The maximum number of records collected before the
collection thread will spill is <code>r * x * q * 2^16</code>.
Note that a higher value may decrease the number of- or even
eliminate- merges, but will also increase the probability of
the map task getting blocked. The lowest average map times are
usually obtained by accurately estimating the size of the map
output and preventing multiple spills.</td></tr>
</table>
<p>Other notes</p>
<ul>
<li>If either spill threshold is exceeded while a spill is in
progress, collection will continue until the spill is finished.
For example, if <code>io.sort.buffer.spill.percent</code> is set
to 0.33, and the remainder of the buffer is filled while the spill
runs, the next spill will include all the collected records, or
0.66 of the buffer, and will not generate additional spills. In
other words, the thresholds are defining triggers, not
blocking.</li>
<li>A record larger than the serialization buffer will first
trigger a spill, then be spilled to a separate file. It is
undefined whether or not this record will first pass through the
combiner.</li>
</ul>
</section>
<section>
<title>Shuffle/Reduce Parameters</title>
<p>As described previously, each reduce fetches the output assigned
to it by the Partitioner via HTTP into memory and periodically
merges these outputs to disk. If intermediate compression of map
outputs is turned on, each output is decompressed into memory. The
following options affect the frequency of these merges to disk prior
to the reduce and the memory allocated to map output during the
reduce.</p>
<table>
<tr><th>Name</th><th>Type</th><th>Description</th></tr>
<tr><td>mapreduce.task.io.sort.factor</td><td>int</td>
<td>Specifies the number of segments on disk to be merged at
the same time. It limits the number of open files and
compression codecs during the merge. If the number of files
exceeds this limit, the merge will proceed in several passes.
Though this limit also applies to the map, most jobs should be
configured so that hitting this limit is unlikely
there.</td></tr>
<tr><td>mapreduce.reduce.merge.inmem.threshold</td><td>int</td>
<td>The number of sorted map outputs fetched into memory
before being merged to disk. Like the spill thresholds in the
preceding note, this is not defining a unit of partition, but
a trigger. In practice, this is usually set very high (1000)
or disabled (0), since merging in-memory segments is often
less expensive than merging from disk (see notes following
this table). This threshold influences only the frequency of
in-memory merges during the shuffle.</td></tr>
<tr><td>mapreduce.reduce.shuffle.merge.percent</td><td>float</td>
<td>The memory threshold for fetched map outputs before an
in-memory merge is started, expressed as a percentage of
memory allocated to storing map outputs in memory. Since map
outputs that can't fit in memory can be stalled, setting this
high may decrease parallelism between the fetch and merge.
Conversely, values as high as 1.0 have been effective for
reduces whose input can fit entirely in memory. This parameter
influences only the frequency of in-memory merges during the
shuffle.</td></tr>
<tr><td>mapreduce.reduce.shuffle.input.buffer.percent</td><td>float</td>
<td>The percentage of memory- relative to the maximum heapsize
as typically specified in <code>mapreduce.reduce.java.opts</code>-
that can be allocated to storing map outputs during the
shuffle. Though some memory should be set aside for the
framework, in general it is advantageous to set this high
enough to store large and numerous map outputs.</td></tr>
<tr><td>mapreduce.reduce.input.buffer.percent</td><td>float</td>
<td>The percentage of memory relative to the maximum heapsize
in which map outputs may be retained during the reduce. When
the reduce begins, map outputs will be merged to disk until
those that remain are under the resource limit this defines.
By default, all map outputs are merged to disk before the
reduce begins to maximize the memory available to the reduce.
For less memory-intensive reduces, this should be increased to
avoid trips to disk.</td></tr>
</table>
<p>Other notes</p>
<ul>
<li>If a map output is larger than 25 percent of the memory
allocated to copying map outputs, it will be written directly to
disk without first staging through memory.</li>
<li>When running with a combiner, the reasoning about high merge
thresholds and large buffers may not hold. For merges started
before all map outputs have been fetched, the combiner is run
while spilling to disk. In some cases, one can obtain better
reduce times by spending resources combining map outputs- making
disk spills small and parallelizing spilling and fetching- rather
than aggressively increasing buffer sizes.</li>
<li>When merging in-memory map outputs to disk to begin the
reduce, if an intermediate merge is necessary because there are
segments to spill and at least
<code>mapreduce.task.io.sort.factor</code>
segments already on disk, the in-memory map outputs will be part
of the intermediate merge.</li>
</ul>
</section>
<section>
<title> Directory Structure </title>
<p>The task tracker has local directory,
<code> ${mapreduce.cluster.local.dir}/taskTracker/</code> to create localized
cache and localized job. It can define multiple local directories
(spanning multiple disks) and then each filename is assigned to a
semi-random local directory. When the job starts, task tracker
creates a localized job directory relative to the local directory
specified in the configuration. Thus the task tracker directory
structure looks the following: </p>
<ul>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/archive/</code> :
The distributed cache. This directory holds the localized distributed
cache. Thus localized distributed cache is shared among all
the tasks and jobs </li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/</code> :
The localized job directory
<ul>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/work/</code>
: The job-specific shared directory. The tasks can use this space as
scratch space and share files among them. This directory is exposed
to the users through the configuration property
<code>mapreduce.job.local.dir</code>. The directory can accessed through
api <a href="ext:api/org/apache/hadoop/mapred/jobconf/getjoblocaldir">
JobConf.getJobLocalDir()</a>. It is available as System property also.
So, users (streaming etc.) can call
<code>System.getProperty("mapreduce.job.local.dir")</code> to access the
directory.</li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/jars/</code>
: The jars directory, which has the job jar file and expanded jar.
The <code>job.jar</code> is the application's jar file that is
automatically distributed to each machine. It is expanded in jars
directory before the tasks for the job start. The job.jar location
is accessible to the application through the api
<a href="ext:api/org/apache/hadoop/mapred/jobconf/getjar">
JobConf.getJar() </a>. To access the unjarred directory,
JobConf.getJar().getParent() can be called.</li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/job.xml</code>
: The job.xml file, the generic job configuration, localized for
the job. </li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/$taskid</code>
: The task directory for each task attempt. Each task directory
again has the following structure :
<ul>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/$taskid/job.xml</code>
: A job.xml file, task localized job configuration, Task localization
means that properties have been set that are specific to
this particular task within the job. The properties localized for
each task are described below.</li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/$taskid/output</code>
: A directory for intermediate output files. This contains the
temporary map reduce data generated by the framework
such as map output files etc. </li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/$taskid/work</code>
: The curernt working directory of the task.
With <a href="#Task+JVM+Reuse">jvm reuse</a> enabled for tasks, this
directory will be the directory on which the jvm has started</li>
<li><code>${mapreduce.cluster.local.dir}/taskTracker/jobcache/$jobid/$taskid/work/tmp</code>
: The temporary directory for the task.
(User can specify the property <code>mapreduce.task.tmp.dir</code> to set
the value of temporary directory for map and reduce tasks. This
defaults to <code>./tmp</code>. If the value is not an absolute path,
it is prepended with task's working directory. Otherwise, it is
directly assigned. The directory will be created if it doesn't exist.
Then, the child java tasks are executed with option
<code>-Djava.io.tmpdir='the absolute path of the tmp dir'</code>.
Anp pipes and streaming are set with environment variable,
<code>TMPDIR='the absolute path of the tmp dir'</code>). This
directory is created, if <code>mapreduce.task.tmp.dir</code> has the value
<code>./tmp</code> </li>
</ul>
</li>
</ul>
</li>
</ul>
</section>
<section>
<title>Task JVM Reuse</title>
<p>Jobs can enable task JVMs to be reused by specifying the job
configuration <code>mapreduce.job.jvm.numtasks</code>. If the
value is 1 (the default), then JVMs are not reused
(i.e. 1 task per JVM). If it is -1, there is no limit to the number
of tasks a JVM can run (of the same job). One can also specify some
value greater than 1 using the api
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setnumtaskstoexecuteperjvm">
JobConf.setNumTasksToExecutePerJvm(int)</a></p>
</section>
<section>
<title>Configured Parameters</title>
<p>The following properties are localized in the job configuration
for each task's execution: </p>
<table>
<tr><th>Name</th><th>Type</th><th>Description</th></tr>
<tr><td>mapreduce.job.id</td><td>String</td><td>The job id</td></tr>
<tr><td>mapreduce.job.jar</td><td>String</td>
<td>job.jar location in job directory</td></tr>
<tr><td>mapreduce.job.local.dir</td><td> String</td>
<td> The job specific shared scratch space</td></tr>
<tr><td>mapreduce.task.id</td><td> String</td>
<td> The task id</td></tr>
<tr><td>mapreduce.task.attempt.id</td><td> String</td>
<td> The task attempt id</td></tr>
<tr><td>mapreduce.task.ismap</td><td> boolean </td>
<td>Is this a map task</td></tr>
<tr><td>mapreduce.task.partition</td><td> int </td>
<td>The id of the task within the job</td></tr>
<tr><td>mapreduce.map.input.file</td><td> String</td>
<td> The filename that the map is reading from</td></tr>
<tr><td>mapreduce.map.input.start</td><td> long</td>
<td> The offset of the start of the map input split</td></tr>
<tr><td>mapreduce.map.input.length </td><td>long </td>
<td>The number of bytes in the map input split</td></tr>
<tr><td>mapreduce.task.output.dir</td><td> String </td>
<td>The task's temporary output directory</td></tr>
</table>
<p>
<strong>Note:</strong>
During the execution of a streaming job, the names of the "mapred" parameters are transformed.
The dots ( . ) become underscores ( _ ).
For example, mapreduce.job.id becomes mapreduce.job.id and mapreduce.job.jar becomes mapreduce.job.jar.
To get the values in a streaming job's mapper/reducer use the parameter names with the underscores.
</p>
</section>
<section>
<title>Task Logs</title>
<p>The standard output (stdout) and error (stderr) streams of the task
are read by the TaskTracker and logged to
<code>${HADOOP_LOG_DIR}/userlogs</code></p>
</section>
<section>
<title>Distributing Libraries</title>
<p>The <a href="#DistributedCache">DistributedCache</a> can also be used
to distribute both jars and native libraries for use in the map
and/or reduce tasks. The child-jvm always has its
<em>current working directory</em> added to the
<code>java.library.path</code> and <code>LD_LIBRARY_PATH</code>.
And hence the cached libraries can be loaded via
<a href="http://java.sun.com/javase/6/docs/api/java/lang/System.html#loadLibrary(java.lang.String)">
System.loadLibrary</a> or
<a href="http://java.sun.com/javase/6/docs/api/java/lang/System.html#load(java.lang.String)">
System.load</a>. More details on how to load shared libraries through
distributed cache are documented under
<a href="http://hadoop.apache.org/common/docs/current/native_libraries.html#Loading+Native+Libraries+Through+DistributedCache">
Building Native Hadoop Libraries</a>.</p>
</section>
</section>
<section>
<title>Job Submission and Monitoring</title>
<p><a href="ext:api/org/apache/hadoop/mapred/jobclient">
JobClient</a> is the primary interface by which user-job interacts
with the <code>JobTracker</code>.</p>
<p><code>JobClient</code> provides facilities to submit jobs, track their
progress, access component-tasks' reports and logs, get the MapReduce
cluster's status information and so on.</p>
<p>The job submission process involves:</p>
<ol>
<li>Checking the input and output specifications of the job.</li>
<li>Computing the <code>InputSplit</code> values for the job.</li>
<li>
Setting up the requisite accounting information for the
<code>DistributedCache</code> of the job, if necessary.
</li>
<li>
Copying the job's jar and configuration to the MapReduce system
directory on the <code>FileSystem</code>.
</li>
<li>
Submitting the job to the <code>JobTracker</code> and optionally
monitoring it's status.
</li>
</ol>
<p> User can view the history log summary for a given history file
using the following command <br/>
<code>$ bin/hadoop job -history history-file</code><br/>
This command will print job details, failed and killed tip
details. <br/>
More details about the job such as successful tasks and
task attempts made for each task can be viewed using the
following command <br/>
<code>$ bin/hadoop job -history all history-file</code><br/></p>
<p> User can use
<a href="ext:api/org/apache/hadoop/mapred/outputlogfilter">OutputLogFilter</a>
to filter log files from the output directory listing. </p>
<p>Normally the user creates the application, describes various facets
of the job via <code>JobConf</code>, and then uses the
<code>JobClient</code> to submit the job and monitor its progress.</p>
<section>
<title>Job Control</title>
<p>Users may need to chain MapReduce jobs to accomplish complex
tasks which cannot be done via a single MapReduce job. This is fairly
easy since the output of the job typically goes to distributed
file-system, and the output, in turn, can be used as the input for the
next job.</p>
<p>However, this also means that the onus on ensuring jobs are
complete (success/failure) lies squarely on the clients. In such
cases, the various job-control options are:</p>
<ul>
<li>
<a href="ext:api/org/apache/hadoop/mapred/jobclient/runjob">
runJob(JobConf)</a> : Submits the job and returns only after the
job has completed.
</li>
<li>
<a href="ext:api/org/apache/hadoop/mapred/jobclient/submitjob">
submitJob(JobConf)</a> : Only submits the job, then poll the
returned handle to the
<a href="ext:api/org/apache/hadoop/mapred/runningjob">
RunningJob</a> to query status and make scheduling decisions.
</li>
<li>
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setjobendnotificationuri">
JobConf.setJobEndNotificationURI(String)</a> : Sets up a
notification upon job-completion, thus avoiding polling.
</li>
</ul>
</section>
</section>
<section>
<title>Job Input</title>
<p><a href="ext:api/org/apache/hadoop/mapred/inputformat">
InputFormat</a> describes the input-specification for a MapReduce job.
</p>
<p>The MapReduce framework relies on the <code>InputFormat</code> of
the job to:</p>
<ol>
<li>Validate the input-specification of the job.</li>
<li>
Split-up the input file(s) into logical <code>InputSplit</code>
instances, each of which is then assigned to an individual
<code>Mapper</code>.
</li>
<li>
Provide the <code>RecordReader</code> implementation used to
glean input records from the logical <code>InputSplit</code> for
processing by the <code>Mapper</code>.
</li>
</ol>
<p>The default behavior of file-based <code>InputFormat</code>
implementations, typically sub-classes of
<a href="ext:api/org/apache/hadoop/mapred/fileinputformat">
FileInputFormat</a>, is to split the input into <em>logical</em>
<code>InputSplit</code> instances based on the total size, in bytes, of
the input files. However, the <code>FileSystem</code> blocksize of the
input files is treated as an upper bound for input splits. A lower bound
on the split size can be set via <code>mapreduce.input.fileinputformat.split.minsize</code>.</p>
<p>Clearly, logical splits based on input-size is insufficient for many
applications since record boundaries must be respected. In such cases,
the application should implement a <code>RecordReader</code>, who is
responsible for respecting record-boundaries and presents a
record-oriented view of the logical <code>InputSplit</code> to the
individual task.</p>
<p><a href="ext:api/org/apache/hadoop/mapred/textinputformat">
TextInputFormat</a> is the default <code>InputFormat</code>.</p>
<p>If <code>TextInputFormat</code> is the <code>InputFormat</code> for a
given job, the framework detects input-files with the <em>.gz</em>
extensions and automatically decompresses them using the
appropriate <code>CompressionCodec</code>. However, it must be noted that
compressed files with the above extensions cannot be <em>split</em> and
each compressed file is processed in its entirety by a single mapper.</p>
<section>
<title>InputSplit</title>
<p><a href="ext:api/org/apache/hadoop/mapred/inputsplit">
InputSplit</a> represents the data to be processed by an individual
<code>Mapper</code>.</p>
<p>Typically <code>InputSplit</code> presents a byte-oriented view of
the input, and it is the responsibility of <code>RecordReader</code>
to process and present a record-oriented view.</p>
<p><a href="ext:api/org/apache/hadoop/mapred/filesplit">
FileSplit</a> is the default <code>InputSplit</code>. It sets
<code>mapreduce.map.input.file</code> to the path of the input file for the
logical split.</p>
</section>
<section>
<title>RecordReader</title>
<p><a href="ext:api/org/apache/hadoop/mapred/recordreader">
RecordReader</a> reads <code>&lt;key, value&gt;</code> pairs from an
<code>InputSplit</code>.</p>
<p>Typically the <code>RecordReader</code> converts the byte-oriented
view of the input, provided by the <code>InputSplit</code>, and
presents a record-oriented to the <code>Mapper</code> implementations
for processing. <code>RecordReader</code> thus assumes the
responsibility of processing record boundaries and presents the tasks
with keys and values.</p>
</section>
</section>
<section>
<title>Job Output</title>
<p><a href="ext:api/org/apache/hadoop/mapred/outputformat">
OutputFormat</a> describes the output-specification for a MapReduce
job.</p>
<p>The MapReduce framework relies on the <code>OutputFormat</code> of
the job to:</p>
<ol>
<li>
Validate the output-specification of the job; for example, check that
the output directory doesn't already exist.
</li>
<li>
Provide the <code>RecordWriter</code> implementation used to
write the output files of the job. Output files are stored in a
<code>FileSystem</code>.
</li>
</ol>
<p><code>TextOutputFormat</code> is the default
<code>OutputFormat</code>.</p>
<section>
<title>Lazy Output Creation</title>
<p>It is possible to delay creation of output until the first write attempt
by using <a href="ext:api/org/apache/hadoop/mapred/lib/lazyoutputformat">
LazyOutputFormat</a>. This is particularly useful in preventing the
creation of zero byte files when there is no call to output.collect
(or Context.write). This is achieved by calling the static method
<code>setOutputFormatClass</code> of <code>LazyOutputFormat</code>
with the intended <code>OutputFormat</code> as the argument. The following example
shows how to delay creation of files when using the <code>TextOutputFormat</code>
</p>
<p>
<code> import org.apache.hadoop.mapred.lib.LazyOutputFormat;</code> <br/>
<code> LazyOutputFormat.setOutputFormatClass(conf, TextOutputFormat.class);</code>
</p>
</section>
<section>
<title>OutputCommitter</title>
<p><a href="ext:api/org/apache/hadoop/mapred/outputcommitter">
OutputCommitter</a> describes the commit of task output for a
MapReduce job.</p>
<p>The MapReduce framework relies on the <code>OutputCommitter</code>
of the job to:</p>
<ol>
<li>
Setup the job during initialization. For example, create
the temporary output directory for the job during the
initialization of the job.
Job setup is done by a separate task when the job is
in PREP state and after initializing tasks. Once the setup task
completes, the job will be moved to RUNNING state.
</li>
<li>
Cleanup the job after the job completion. For example, remove the
temporary output directory after the job completion.
Job cleanup is done by a separate task at the end of the job.
Job is declared SUCCEDED/FAILED/KILLED after the cleanup
task completes.
</li>
<li>
Setup the task temporary output.
Task setup is done as part of the same task, during task initialization.
</li>
<li>
Check whether a task needs a commit. This is to avoid the commit
procedure if a task does not need commit.
</li>
<li>
Commit of the task output.
Once task is done, the task will commit it's output if required.
</li>
<li>
Discard the task commit.
If the task has been failed/killed, the output will be cleaned-up.
If task could not cleanup (in exception block), a separate task
will be launched with same attempt-id to do the cleanup.
</li>
</ol>
<p><code>FileOutputCommitter</code> is the default
<code>OutputCommitter</code>. Job setup/cleanup tasks occupy
map or reduce slots, whichever is free on the TaskTracker. And
JobCleanup task, TaskCleanup tasks and JobSetup task have the highest
priority, and in that order.</p>
</section>
<section>
<title>Task Side-Effect Files</title>
<p>In some applications, component tasks need to create and/or write to
side-files, which differ from the actual job-output files.</p>
<p>In such cases there could be issues with two instances of the same
<code>Mapper</code> or <code>Reducer</code> running simultaneously (for
example, speculative tasks) trying to open and/or write to the same
file (path) on the <code>FileSystem</code>. Hence the
application-writer will have to pick unique names per task-attempt
(using the attemptid, say <code>attempt_200709221812_0001_m_000000_0</code>),
not just per task.</p>
<p>To avoid these issues the MapReduce framework, when the
<code>OutputCommitter</code> is <code>FileOutputCommitter</code>,
maintains a special
<code>${mapreduce.output.fileoutputformat.outputdir}/_temporary/_${taskid}</code> sub-directory
accessible via <code>${mapreduce.task.output.dir}</code>
for each task-attempt on the <code>FileSystem</code> where the output
of the task-attempt is stored. On successful completion of the
task-attempt, the files in the
<code>${mapreduce.output.fileoutputformat.outputdir}/_temporary/_${taskid}</code> (only)
are <em>promoted</em> to <code>${mapreduce.output.fileoutputformat.outputdir}</code>. Of course,
the framework discards the sub-directory of unsuccessful task-attempts.
This process is completely transparent to the application.</p>
<p>The application-writer can take advantage of this feature by
creating any side-files required in <code>${mapreduce.task.output.dir}</code>
during execution of a task via
<a href="ext:api/org/apache/hadoop/mapred/fileoutputformat/getworkoutputpath">
FileOutputFormat.getWorkOutputPath()</a>, and the framework will promote them
similarly for succesful task-attempts, thus eliminating the need to
pick unique paths per task-attempt.</p>
<p>Note: The value of <code>${mapreduce.task.output.dir}</code> during
execution of a particular task-attempt is actually
<code>${mapreduce.output.fileoutputformat.outputdir}/_temporary/_{$taskid}</code>, and this value is
set by the MapReduce framework. So, just create any side-files in the
path returned by
<a href="ext:api/org/apache/hadoop/mapred/fileoutputformat/getworkoutputpath">
FileOutputFormat.getWorkOutputPath() </a>from MapReduce
task to take advantage of this feature.</p>
<p>The entire discussion holds true for maps of jobs with
reducer=NONE (i.e. 0 reduces) since output of the map, in that case,
goes directly to HDFS.</p>
</section>
<section>
<title>RecordWriter</title>
<p><a href="ext:api/org/apache/hadoop/mapred/recordwriter">
RecordWriter</a> writes the output <code>&lt;key, value&gt;</code>
pairs to an output file.</p>
<p>RecordWriter implementations write the job outputs to the
<code>FileSystem</code>.</p>
</section>
</section>
<section>
<title>Other Useful Features</title>
<section>
<title>Submitting Jobs to Queues</title>
<p>Users submit jobs to Queues. Queues, as collection of jobs,
allow the system to provide specific functionality. For example,
queues use ACLs to control which users
who can submit jobs to them. Queues are expected to be primarily
used by Hadoop Schedulers. </p>
<p>Hadoop comes configured with a single mandatory queue, called
'default'. Queue names are defined in the
<code>mapred.queue.names</code> property of the Hadoop site
configuration. Some job schedulers, such as the
<a href="capacity_scheduler.html">Capacity Scheduler</a>,
support multiple queues.</p>
<p>A job defines the queue it needs to be submitted to through the
<code>mapreduce.job.queuename</code> property, or through the
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setqueuename">setQueueName(String)</a>
API. Setting the queue name is optional. If a job is submitted
without an associated queue name, it is submitted to the 'default'
queue.</p>
</section>
<section>
<title>Counters</title>
<p><code>Counters</code> represent global counters, defined either by
the MapReduce framework or applications. Each <code>Counter</code> can
be of any <code>Enum</code> type. Counters of a particular
<code>Enum</code> are bunched into groups of type
<code>Counters.Group</code>.</p>
<p>Applications can define arbitrary <code>Counters</code> (of type
<code>Enum</code>) and update them via
<a href="ext:api/org/apache/hadoop/mapred/reporter/incrcounterEnum">
Reporter.incrCounter(Enum, long)</a> or
<a href="ext:api/org/apache/hadoop/mapred/reporter/incrcounterString">
Reporter.incrCounter(String, String, long)</a>
in the <code>map</code> and/or
<code>reduce</code> methods. These counters are then globally
aggregated by the framework.</p>
</section>
<section>
<title>DistributedCache</title>
<p><a href="ext:api/org/apache/hadoop/filecache/distributedcache">
DistributedCache</a> distributes application-specific, large, read-only
files efficiently.</p>
<p><code>DistributedCache</code> is a facility provided by the
MapReduce framework to cache files (text, archives, jars and so on)
needed by applications.</p>
<p>Applications specify the files to be cached via urls (hdfs://)
in the <code>JobConf</code>. The <code>DistributedCache</code>
assumes that the files specified via hdfs:// urls are already present
on the <code>FileSystem</code>.</p>
<p>The framework will copy the necessary files to the slave node
before any tasks for the job are executed on that node. Its
efficiency stems from the fact that the files are only copied once
per job and the ability to cache archives which are un-archived on
the slaves.</p>
<p><code>DistributedCache</code> tracks the modification timestamps of
the cached files. Clearly the cache files should not be modified by
the application or externally while the job is executing.</p>
<p><code>DistributedCache</code> can be used to distribute simple,
read-only data/text files and more complex types such as archives and
jars. Archives (zip, tar, tgz and tar.gz files) are
<em>un-archived</em> at the slave nodes. Files
have <em>execution permissions</em> set. </p>
<p>The files/archives can be distributed by setting the property
<code>mapred.cache.{files|archives}</code>. If more than one
file/archive has to be distributed, they can be added as comma
separated paths. The properties can also be set by APIs
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/addcachefile">
DistributedCache.addCacheFile(URI,conf)</a>/
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/addcachearchive">
DistributedCache.addCacheArchive(URI,conf)</a> and
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/setcachefiles">
DistributedCache.setCacheFiles(URIs,conf)</a>/
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/setcachearchives">
DistributedCache.setCacheArchives(URIs,conf)</a>
where URI is of the form
<code>hdfs://host:port/absolute-path#link-name</code>.
In Streaming, the files can be distributed through command line
option <code>-cacheFile/-cacheArchive</code>.</p>
<p>Optionally users can also direct the <code>DistributedCache</code>
to <em>symlink</em> the cached file(s) into the <code>current working
directory</code> of the task via the
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/createsymlink">
DistributedCache.createSymlink(Configuration)</a> api. Or by setting
the configuration property <code>mapreduce.job.cache.symlink.create</code>
as <code>yes</code>. The DistributedCache will use the
<code>fragment</code> of the URI as the name of the symlink.
For example, the URI
<code>hdfs://namenode:port/lib.so.1#lib.so</code>
will have the symlink name as <code>lib.so</code> in task's cwd
for the file <code>lib.so.1</code> in distributed cache.</p>
<p>The <code>DistributedCache</code> can also be used as a
rudimentary software distribution mechanism for use in the
map and/or reduce tasks. It can be used to distribute both
jars and native libraries. The
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/addarchivetoclasspath">
DistributedCache.addArchiveToClassPath(Path, Configuration)</a> or
<a href="ext:api/org/apache/hadoop/filecache/distributedcache/addfiletoclasspath">
DistributedCache.addFileToClassPath(Path, Configuration)</a> api
can be used to cache files/jars and also add them to the
<em>classpath</em> of child-jvm. The same can be done by setting
the configuration properties
<code>mapreduce.job.classpath.{files|archives}</code>. Similarly the
cached files that are symlinked into the working directory of the
task can be used to distribute native libraries and load them.</p>
<p>The <code>DistributedCache</code> tracks modification timestamps
of the cache files/archives. Clearly the cache files/archives should
not be modified by the application or externally
while the job is executing.</p>
<p>Here is an illustrative example on how to use the
<code>DistributedCache</code>:<br/>
// Setting up the cache for the application
1. Copy the requisite files to the <code>FileSystem</code>:<br/>
<code>$ bin/hadoop fs -copyFromLocal lookup.dat /myapp/lookup.dat</code><br/>
<code>$ bin/hadoop fs -copyFromLocal map.zip /myapp/map.zip </code><br/>
<code>$ bin/hadoop fs -copyFromLocal mylib.jar /myapp/mylib.jar</code><br/>
<code>$ bin/hadoop fs -copyFromLocal mytar.tar /myapp/mytar.tar</code><br/>
<code>$ bin/hadoop fs -copyFromLocal mytgz.tgz /myapp/mytgz.tgz</code><br/>
<code>$ bin/hadoop fs -copyFromLocal mytargz.tar.gz /myapp/mytargz.tar.gz</code><br/>
2. Setup the job<br/>
<code>Job job = new Job(conf);</code><br/>
<code>job.addCacheFile(new URI("/myapp/lookup.dat#lookup.dat"));</code><br/>
<code>job.addCacheArchive(new URI("/myapp/map.zip"));</code><br/>
<code>job.addFileToClassPath(new Path("/myapp/mylib.jar"));</code><br/>
<code>job.addCacheArchive(new URI("/myapp/mytar.tar"));</code><br/>
<code>job.addCacheArchive(new URI("/myapp/mytgz.tgz"));</code><br/>
<code>job.addCacheArchive(new URI("/myapp/mytargz.tar.gz"));</code><br/>
3. Use the cached files in the
<code>{@link org.apache.hadoop.mapreduce.Mapper}
or {@link org.apache.hadoop.mapreduce.Reducer}:</code><br/>
<code>public static class MapClass extends Mapper&lt;K, V, K, V&gt; {</code><br/>
<code>private Path[] localArchives;</code><br/>
<code>private Path[] localFiles;</code><br/>
<code>public void setup(Context context) {</code><br/>
<code>// Get the cached archives/files</code><br/>
<code>localArchives = context.getLocalCacheArchives();</code><br/>
<code>localFiles = context.getLocalCacheFiles();</code><br/>
<code>}</code><br/>
<code>public void map(K key, V value,
Context context) throws IOException {</code><br/>
<code>// Use data from the cached archives/files here</code><br/>
<code>// ...</code><br/>
<code>// ...</code><br/>
<code>context.write(k, v);</code><br/>
<code>}</code><br/>
<code>}</code></p>
</section>
<section>
<title>Tool</title>
<p>The <a href="ext:api/org/apache/hadoop/util/tool">Tool</a>
interface supports the handling of generic Hadoop command-line options.
</p>
<p><code>Tool</code> is the standard for any MapReduce tool or
application. The application should delegate the handling of
standard command-line options to
<a href="ext:api/org/apache/hadoop/util/genericoptionsparser">
GenericOptionsParser</a> via
<a href="ext:api/org/apache/hadoop/util/toolrunner/run">
ToolRunner.run(Tool, String[])</a> and only handle its custom
arguments.</p>
<p>
The generic Hadoop command-line options are:<br/>
<code>
-conf &lt;configuration file&gt;
</code>
<br/>
<code>
-D &lt;property=value&gt;
</code>
<br/>
<code>
-fs &lt;local|namenode:port&gt;
</code>
<br/>
<code>
-jt &lt;local|jobtracker:port&gt;
</code>
</p>
</section>
<section>
<title>IsolationRunner</title>
<p><a href="ext:api/org/apache/hadoop/mapred/isolationrunner">
IsolationRunner</a> is a utility to help debug MapReduce programs.</p>
<p>To use the <code>IsolationRunner</code>, first set
<code>keep.failed.tasks.files</code> to <code>true</code>
(also see <code>keep.tasks.files.pattern</code>).</p>
<p>
Next, go to the node on which the failed task ran and go to the
<code>TaskTracker</code>'s local directory and run the
<code>IsolationRunner</code>:<br/>
<code>$ cd &lt;local path&gt;/taskTracker/${taskid}/work</code><br/>
<code>
$ bin/hadoop org.apache.hadoop.mapred.IsolationRunner ../job.xml
</code>
</p>
<p><code>IsolationRunner</code> will run the failed task in a single
jvm, which can be in the debugger, over precisely the same input.</p>
</section>
<section>
<title>Profiling</title>
<p>Profiling is a utility to get a representative (2 or 3) sample
of built-in java profiler for a sample of maps and reduces. </p>
<p>User can specify whether the system should collect profiler
information for some of the tasks in the job by setting the
configuration property <code>mapreduce.task.profile</code>. The
value can be set using the api
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setprofileenabled">
JobConf.setProfileEnabled(boolean)</a>. If the value is set
<code>true</code>, the task profiling is enabled. The profiler
information is stored in the user log directory. By default,
profiling is not enabled for the job. </p>
<p>Once user configures that profiling is needed, she/he can use
the configuration property
<code>mapreduce.task.profile.{maps|reduces}</code> to set the ranges
of MapReduce tasks to profile. The value can be set using the api
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setprofiletaskrange">
JobConf.setProfileTaskRange(boolean,String)</a>.
By default, the specified range is <code>0-2</code>.</p>
<p>User can also specify the profiler configuration arguments by
setting the configuration property
<code>mapreduce.task.profile.params</code>. The value can be specified
using the api
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setprofileparams">
JobConf.setProfileParams(String)</a>. If the string contains a
<code>%s</code>, it will be replaced with the name of the profiling
output file when the task runs. These parameters are passed to the
task child JVM on the command line. The default value for
the profiling parameters is
<code>-agentlib:hprof=cpu=samples,heap=sites,force=n,thread=y,verbose=n,file=%s</code>
</p>
</section>
<section>
<title>Debugging</title>
<p>The MapReduce framework provides a facility to run user-provided
scripts for debugging. When a MapReduce task fails, a user can run
a debug script, to process task logs for example. The script is
given access to the task's stdout and stderr outputs, syslog and
jobconf. The output from the debug script's stdout and stderr is
displayed on the console diagnostics and also as part of the
job UI. </p>
<p> In the following sections we discuss how to submit a debug script
with a job. The script file needs to be distributed and submitted to
the framework.</p>
<section>
<title> How to distribute the script file: </title>
<p>
The user needs to use
<a href="mapred_tutorial.html#DistributedCache">DistributedCache</a>
to <em>distribute</em> and <em>symlink</em> the script file.</p>
</section>
<section>
<title> How to submit the script: </title>
<p> A quick way to submit the debug script is to set values for the
properties <code>mapreduce.map.debug.script</code> and
<code>mapreduce.reduce.debug.script</code>, for debugging map and
reduce tasks respectively. These properties can also be set by using APIs
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmapdebugscript">
JobConf.setMapDebugScript(String) </a> and
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setreducedebugscript">
JobConf.setReduceDebugScript(String) </a>. In streaming mode, a debug
script can be submitted with the command-line options
<code>-mapdebug</code> and <code>-reducedebug</code>, for debugging
map and reduce tasks respectively.</p>
<p>The arguments to the script are the task's stdout, stderr,
syslog and jobconf files. The debug command, run on the node where
the MapReduce task failed, is: <br/>
<code> $script $stdout $stderr $syslog $jobconf </code> </p>
<p> Pipes programs have the c++ program name as a fifth argument
for the command. Thus for the pipes programs the command is <br/>
<code>$script $stdout $stderr $syslog $jobconf $program </code>
</p>
</section>
<section>
<title> Default Behavior: </title>
<p> For pipes, a default script is run to process core dumps under
gdb, prints stack trace and gives info about running threads. </p>
</section>
</section>
<section>
<title>JobControl</title>
<p><a href="ext:api/org/apache/hadoop/mapred/jobcontrol/package-summary">
JobControl</a> is a utility which encapsulates a set of MapReduce jobs
and their dependencies.</p>
</section>
<section>
<title>Data Compression</title>
<p>Hadoop MapReduce provides facilities for the application-writer to
specify compression for both intermediate map-outputs and the
job-outputs i.e. output of the reduces. It also comes bundled with
<a href="ext:api/org/apache/hadoop/io/compress/compressioncodec">
CompressionCodec</a> implementation for the
<a href="ext:zlib">zlib</a> compression
algorithm. The <a href="ext:gzip">gzip</a> file format is also
supported.</p>
<p>Hadoop also provides native implementations of the above compression
codecs for reasons of both performance (zlib) and non-availability of
Java libraries. For more information see the
<a href="http://hadoop.apache.org/common/docs/current/native_libraries.html">Native Libraries Guide</a>.</p>
<section>
<title>Intermediate Outputs</title>
<p>Applications can control compression of intermediate map-outputs
via the
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setcompressmapoutput">
JobConf.setCompressMapOutput(boolean)</a> api and the
<code>CompressionCodec</code> to be used via the
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmapoutputcompressorclass">
JobConf.setMapOutputCompressorClass(Class)</a> api.</p>
</section>
<section>
<title>Job Outputs</title>
<p>Applications can control compression of job-outputs via the
<a href="ext:api/org/apache/hadoop/mapred/fileoutputformat/setcompressoutput">
FileOutputFormat.setCompressOutput(JobConf, boolean)</a> api and the
<code>CompressionCodec</code> to be used can be specified via the
<a href="ext:api/org/apache/hadoop/mapred/fileoutputformat/setoutputcompressorclass">
FileOutputFormat.setOutputCompressorClass(JobConf, Class)</a> api.</p>
<p>If the job outputs are to be stored in the
<a href="ext:api/org/apache/hadoop/mapred/sequencefileoutputformat">
SequenceFileOutputFormat</a>, the required
<code>SequenceFile.CompressionType</code> (i.e. <code>RECORD</code> /
<code>BLOCK</code> - defaults to <code>RECORD</code>) can be
specified via the
<a href="ext:api/org/apache/hadoop/mapred/sequencefileoutputformat/setoutputcompressiontype">
SequenceFileOutputFormat.setOutputCompressionType(JobConf,
SequenceFile.CompressionType)</a> api.</p>
</section>
</section>
<section>
<title>Skipping Bad Records</title>
<p>Hadoop provides an option where a certain set of bad input
records can be skipped when processing map inputs. Applications
can control this feature through the
<a href="ext:api/org/apache/hadoop/mapred/skipbadrecords">
SkipBadRecords</a> class.</p>
<p>This feature can be used when map tasks crash deterministically
on certain input. This usually happens due to bugs in the
map function. Usually, the user would have to fix these bugs.
This is, however, not possible sometimes. The bug may be in third
party libraries, for example, for which the source code is not
available. In such cases, the task never completes successfully even
after multiple attempts, and the job fails. With this feature, only
a small portion of data surrounding the
bad records is lost, which may be acceptable for some applications
(those performing statistical analysis on very large data, for
example). </p>
<p>By default this feature is disabled. For enabling it,
refer to <a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/setmappermaxskiprecords">
SkipBadRecords.setMapperMaxSkipRecords(Configuration, long)</a> and
<a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/setreducermaxskipgroups">
SkipBadRecords.setReducerMaxSkipGroups(Configuration, long)</a>.
</p>
<p>With this feature enabled, the framework gets into 'skipping
mode' after a certain number of map failures. For more details,
see <a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/setattemptsTostartskipping">
SkipBadRecords.setAttemptsToStartSkipping(Configuration, int)</a>.
In 'skipping mode', map tasks maintain the range of records being
processed. To do this, the framework relies on the processed record
counter. See <a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/counter_map_processed_records">
SkipBadRecords.COUNTER_MAP_PROCESSED_RECORDS</a> and
<a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/counter_reduce_processed_groups">
SkipBadRecords.COUNTER_REDUCE_PROCESSED_GROUPS</a>.
This counter enables the framework to know how many records have
been processed successfully, and hence, what record range caused
a task to crash. On further attempts, this range of records is
skipped.</p>
<p>The number of records skipped depends on how frequently the
processed record counter is incremented by the application.
It is recommended that this counter be incremented after every
record is processed. This may not be possible in some applications
that typically batch their processing. In such cases, the framework
may skip additional records surrounding the bad record. Users can
control the number of skipped records through
<a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/setmappermaxskiprecords">
SkipBadRecords.setMapperMaxSkipRecords(Configuration, long)</a> and
<a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/setreducermaxskipgroups">
SkipBadRecords.setReducerMaxSkipGroups(Configuration, long)</a>.
The framework tries to narrow the range of skipped records using a
binary search-like approach. The skipped range is divided into two
halves and only one half gets executed. On subsequent
failures, the framework figures out which half contains
bad records. A task will be re-executed till the
acceptable skipped value is met or all task attempts are exhausted.
To increase the number of task attempts, use
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmaxmapattempts">
JobConf.setMaxMapAttempts(int)</a> and
<a href="ext:api/org/apache/hadoop/mapred/jobconf/setmaxreduceattempts">
JobConf.setMaxReduceAttempts(int)</a>.
</p>
<p>Skipped records are written to HDFS in the sequence file
format, for later analysis. The location can be changed through
<a href="ext:api/org/apache/hadoop/mapred/skipbadrecords/setskipoutputpath">
SkipBadRecords.setSkipOutputPath(JobConf, Path)</a>.
</p>
</section>
</section>
</section>
<section>
<title>Example: WordCount v2.0</title>
<p>Here is a more complete <code>WordCount</code> which uses many of the
features provided by the MapReduce framework we discussed so far.</p>
<p>This example needs the HDFS to be up and running, especially for the
<code>DistributedCache</code>-related features. Hence it only works with a
pseudo-distributed (<a href="ext:single-node-setup">Single Node Setup</a>)
or fully-distributed (<a href="cluster_setup.html#Fully-Distributed+Operation">Cluster Setup</a>)
Hadoop installation.</p>
<section>
<title>Source Code</title>
<table>
<tr>
<th></th>
<th>WordCount.java</th>
</tr>
<tr>
<td>1.</td>
<td>
<code>package org.myorg;</code>
</td>
</tr>
<tr>
<td>2.</td>
<td></td>
</tr>
<tr>
<td>3.</td>
<td>
<code>import java.io.*;</code>
</td>
</tr>
<tr>
<td>4.</td>
<td>
<code>import java.util.*;</code>
</td>
</tr>
<tr>
<td>5.</td>
<td></td>
</tr>
<tr>
<td>6.</td>
<td>
<code>import org.apache.hadoop.fs.Path;</code>
</td>
</tr>
<tr>
<td>7.</td>
<td>
<code>import org.apache.hadoop.filecache.DistributedCache;</code>
</td>
</tr>
<tr>
<td>8.</td>
<td>
<code>import org.apache.hadoop.conf.*;</code>
</td>
</tr>
<tr>
<td>9.</td>
<td>
<code>import org.apache.hadoop.io.*;</code>
</td>
</tr>
<tr>
<td>10.</td>
<td>
<code>import org.apache.hadoop.mapred.*;</code>
</td>
</tr>
<tr>
<td>11.</td>
<td>
<code>import org.apache.hadoop.util.*;</code>
</td>
</tr>
<tr>
<td>12.</td>
<td></td>
</tr>
<tr>
<td>13.</td>
<td>
<code>public class WordCount extends Configured implements Tool {</code>
</td>
</tr>
<tr>
<td>14.</td>
<td></td>
</tr>
<tr>
<td>15.</td>
<td>
&nbsp;&nbsp;
<code>
public static class Map extends MapReduceBase
implements Mapper&lt;LongWritable, Text, Text, IntWritable&gt; {
</code>
</td>
</tr>
<tr>
<td>16.</td>
<td></td>
</tr>
<tr>
<td>17.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
static enum Counters { INPUT_WORDS }
</code>
</td>
</tr>
<tr>
<td>18.</td>
<td></td>
</tr>
<tr>
<td>19.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
private final static IntWritable one = new IntWritable(1);
</code>
</td>
</tr>
<tr>
<td>20.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private Text word = new Text();</code>
</td>
</tr>
<tr>
<td>21.</td>
<td></td>
</tr>
<tr>
<td>22.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private boolean caseSensitive = true;</code>
</td>
</tr>
<tr>
<td>23.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private Set&lt;String&gt; patternsToSkip = new HashSet&lt;String&gt;();</code>
</td>
</tr>
<tr>
<td>24.</td>
<td></td>
</tr>
<tr>
<td>25.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private long numRecords = 0;</code>
</td>
</tr>
<tr>
<td>26.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private String inputFile;</code>
</td>
</tr>
<tr>
<td>27.</td>
<td></td>
</tr>
<tr>
<td>28.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>public void configure(JobConf job) {</code>
</td>
</tr>
<tr>
<td>29.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
caseSensitive = job.getBoolean("wordcount.case.sensitive", true);
</code>
</td>
</tr>
<tr>
<td>30.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>inputFile = job.get("mapreduce.map.input.file");</code>
</td>
</tr>
<tr>
<td>31.</td>
<td></td>
</tr>
<tr>
<td>32.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>if (job.getBoolean("wordcount.skip.patterns", false)) {</code>
</td>
</tr>
<tr>
<td>33.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>Path[] patternsFiles = new Path[0];</code>
</td>
</tr>
<tr>
<td>34.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>try {</code>
</td>
</tr>
<tr>
<td>35.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
patternsFiles = DistributedCache.getLocalCacheFiles(job);
</code>
</td>
</tr>
<tr>
<td>36.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>} catch (IOException ioe) {</code>
</td>
</tr>
<tr>
<td>37.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
System.err.println("Caught exception while getting cached files: "
+ StringUtils.stringifyException(ioe));
</code>
</td>
</tr>
<tr>
<td>38.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>39.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>for (Path patternsFile : patternsFiles) {</code>
</td>
</tr>
<tr>
<td>40.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>parseSkipFile(patternsFile);</code>
</td>
</tr>
<tr>
<td>41.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>42.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>43.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>44.</td>
<td></td>
</tr>
<tr>
<td>45.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>private void parseSkipFile(Path patternsFile) {</code>
</td>
</tr>
<tr>
<td>46.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>try {</code>
</td>
</tr>
<tr>
<td>47.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
BufferedReader fis =
new BufferedReader(new FileReader(patternsFile.toString()));
</code>
</td>
</tr>
<tr>
<td>48.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>String pattern = null;</code>
</td>
</tr>
<tr>
<td>49.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>while ((pattern = fis.readLine()) != null) {</code>
</td>
</tr>
<tr>
<td>50.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>patternsToSkip.add(pattern);</code>
</td>
</tr>
<tr>
<td>51.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>52.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>} catch (IOException ioe) {</code>
</td>
</tr>
<tr>
<td>53.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
System.err.println("Caught exception while parsing the cached file '" +
patternsFile + "' : " +
StringUtils.stringifyException(ioe));
</code>
</td>
</tr>
<tr>
<td>54.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>55.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>56.</td>
<td></td>
</tr>
<tr>
<td>57.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
public void map(LongWritable key, Text value,
OutputCollector&lt;Text, IntWritable&gt; output,
Reporter reporter) throws IOException {
</code>
</td>
</tr>
<tr>
<td>58.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
String line =
(caseSensitive) ? value.toString() :
value.toString().toLowerCase();
</code>
</td>
</tr>
<tr>
<td>59.</td>
<td></td>
</tr>
<tr>
<td>60.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>for (String pattern : patternsToSkip) {</code>
</td>
</tr>
<tr>
<td>61.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>line = line.replaceAll(pattern, "");</code>
</td>
</tr>
<tr>
<td>62.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>63.</td>
<td></td>
</tr>
<tr>
<td>64.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>StringTokenizer tokenizer = new StringTokenizer(line);</code>
</td>
</tr>
<tr>
<td>65.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>while (tokenizer.hasMoreTokens()) {</code>
</td>
</tr>
<tr>
<td>66.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>word.set(tokenizer.nextToken());</code>
</td>
</tr>
<tr>
<td>67.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>output.collect(word, one);</code>
</td>
</tr>
<tr>
<td>68.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>reporter.incrCounter(Counters.INPUT_WORDS, 1);</code>
</td>
</tr>
<tr>
<td>69.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>70.</td>
<td></td>
</tr>
<tr>
<td>71.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>if ((++numRecords % 100) == 0) {</code>
</td>
</tr>
<tr>
<td>72.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
reporter.setStatus("Finished processing " + numRecords +
" records " + "from the input file: " +
inputFile);
</code>
</td>
</tr>
<tr>
<td>73.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>74.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>75.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>76.</td>
<td></td>
</tr>
<tr>
<td>77.</td>
<td>
&nbsp;&nbsp;
<code>
public static class Reduce extends MapReduceBase implements
Reducer&lt;Text, IntWritable, Text, IntWritable&gt; {
</code>
</td>
</tr>
<tr>
<td>78.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
public void reduce(Text key, Iterator&lt;IntWritable&gt; values,
OutputCollector&lt;Text, IntWritable&gt; output,
Reporter reporter) throws IOException {
</code>
</td>
</tr>
<tr>
<td>79.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>int sum = 0;</code>
</td>
</tr>
<tr>
<td>80.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>while (values.hasNext()) {</code>
</td>
</tr>
<tr>
<td>81.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>sum += values.next().get();</code>
</td>
</tr>
<tr>
<td>82.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>83.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>output.collect(key, new IntWritable(sum));</code>
</td>
</tr>
<tr>
<td>84.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>85.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>86.</td>
<td></td>
</tr>
<tr>
<td>87.</td>
<td>
&nbsp;&nbsp;
<code>public int run(String[] args) throws Exception {</code>
</td>
</tr>
<tr>
<td>88.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
JobConf conf = new JobConf(getConf(), WordCount.class);
</code>
</td>
</tr>
<tr>
<td>89.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setJobName("wordcount");</code>
</td>
</tr>
<tr>
<td>90.</td>
<td></td>
</tr>
<tr>
<td>91.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setOutputKeyClass(Text.class);</code>
</td>
</tr>
<tr>
<td>92.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setOutputValueClass(IntWritable.class);</code>
</td>
</tr>
<tr>
<td>93.</td>
<td></td>
</tr>
<tr>
<td>94.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setMapperClass(Map.class);</code>
</td>
</tr>
<tr>
<td>95.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setCombinerClass(Reduce.class);</code>
</td>
</tr>
<tr>
<td>96.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setReducerClass(Reduce.class);</code>
</td>
</tr>
<tr>
<td>97.</td>
<td></td>
</tr>
<tr>
<td>98.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setInputFormat(TextInputFormat.class);</code>
</td>
</tr>
<tr>
<td>99.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>conf.setOutputFormat(TextOutputFormat.class);</code>
</td>
</tr>
<tr>
<td>100.</td>
<td></td>
</tr>
<tr>
<td>101.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
List&lt;String&gt; other_args = new ArrayList&lt;String&gt;();
</code>
</td>
</tr>
<tr>
<td>102.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>for (int i=0; i &lt; args.length; ++i) {</code>
</td>
</tr>
<tr>
<td>103.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>if ("-skip".equals(args[i])) {</code>
</td>
</tr>
<tr>
<td>104.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
DistributedCache.addCacheFile(new Path(args[++i]).toUri(), conf);
</code>
</td>
</tr>
<tr>
<td>105.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>
conf.setBoolean("wordcount.skip.patterns", true);
</code>
</td>
</tr>
<tr>
<td>106.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>} else {</code>
</td>
</tr>
<tr>
<td>107.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>other_args.add(args[i]);</code>
</td>
</tr>
<tr>
<td>108.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>109.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>110.</td>
<td></td>
</tr>
<tr>
<td>111.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>FileInputFormat.setInputPaths(conf, new Path(other_args.get(0)));</code>
</td>
</tr>
<tr>
<td>112.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>FileOutputFormat.setOutputPath(conf, new Path(other_args.get(1)));</code>
</td>
</tr>
<tr>
<td>113.</td>
<td></td>
</tr>
<tr>
<td>114.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>JobClient.runJob(conf);</code>
</td>
</tr>
<tr>
<td>115.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>return 0;</code>
</td>
</tr>
<tr>
<td>116.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>117.</td>
<td></td>
</tr>
<tr>
<td>118.</td>
<td>
&nbsp;&nbsp;
<code>
public static void main(String[] args) throws Exception {
</code>
</td>
</tr>
<tr>
<td>119.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>
int res = ToolRunner.run(new Configuration(), new WordCount(),
args);
</code>
</td>
</tr>
<tr>
<td>120.</td>
<td>
&nbsp;&nbsp;&nbsp;&nbsp;
<code>System.exit(res);</code>
</td>
</tr>
<tr>
<td>121.</td>
<td>
&nbsp;&nbsp;
<code>}</code>
</td>
</tr>
<tr>
<td>122.</td>
<td>
<code>}</code>
</td>
</tr>
<tr>
<td>123.</td>
<td></td>
</tr>
</table>
</section>
<section>
<title>Sample Runs</title>
<p>Sample text-files as input:</p>
<p>
<code>$ bin/hadoop dfs -ls /usr/joe/wordcount/input/</code><br/>
<code>/usr/joe/wordcount/input/file01</code><br/>
<code>/usr/joe/wordcount/input/file02</code><br/>
<br/>
<code>$ bin/hadoop dfs -cat /usr/joe/wordcount/input/file01</code><br/>
<code>Hello World, Bye World!</code><br/>
<br/>
<code>$ bin/hadoop dfs -cat /usr/joe/wordcount/input/file02</code><br/>
<code>Hello Hadoop, Goodbye to hadoop.</code>
</p>
<p>Run the application:</p>
<p>
<code>
$ bin/hadoop jar /usr/joe/wordcount.jar org.myorg.WordCount
/usr/joe/wordcount/input /usr/joe/wordcount/output
</code>
</p>
<p>Output:</p>
<p>
<code>
$ bin/hadoop dfs -cat /usr/joe/wordcount/output/part-00000
</code>
<br/>
<code>Bye 1</code><br/>
<code>Goodbye 1</code><br/>
<code>Hadoop, 1</code><br/>
<code>Hello 2</code><br/>
<code>World! 1</code><br/>
<code>World, 1</code><br/>
<code>hadoop. 1</code><br/>
<code>to 1</code><br/>
</p>
<p>Notice that the inputs differ from the first version we looked at,
and how they affect the outputs.</p>
<p>Now, lets plug-in a pattern-file which lists the word-patterns to be
ignored, via the <code>DistributedCache</code>.</p>
<p>
<code>$ hadoop dfs -cat /user/joe/wordcount/patterns.txt</code><br/>
<code>\.</code><br/>
<code>\,</code><br/>
<code>\!</code><br/>
<code>to</code><br/>
</p>
<p>Run it again, this time with more options:</p>
<p>
<code>
$ bin/hadoop jar /usr/joe/wordcount.jar org.myorg.WordCount
-Dwordcount.case.sensitive=true /usr/joe/wordcount/input
/usr/joe/wordcount/output -skip /user/joe/wordcount/patterns.txt
</code>
</p>
<p>As expected, the output:</p>
<p>
<code>
$ bin/hadoop dfs -cat /usr/joe/wordcount/output/part-00000
</code>
<br/>
<code>Bye 1</code><br/>
<code>Goodbye 1</code><br/>
<code>Hadoop 1</code><br/>
<code>Hello 2</code><br/>
<code>World 2</code><br/>
<code>hadoop 1</code><br/>
</p>
<p>Run it once more, this time switch-off case-sensitivity:</p>
<p>
<code>
$ bin/hadoop jar /usr/joe/wordcount.jar org.myorg.WordCount
-Dwordcount.case.sensitive=false /usr/joe/wordcount/input
/usr/joe/wordcount/output -skip /user/joe/wordcount/patterns.txt
</code>
</p>
<p>Sure enough, the output:</p>
<p>
<code>
$ bin/hadoop dfs -cat /usr/joe/wordcount/output/part-00000
</code>
<br/>
<code>bye 1</code><br/>
<code>goodbye 1</code><br/>
<code>hadoop 2</code><br/>
<code>hello 2</code><br/>
<code>world 2</code><br/>
</p>
</section>
<section>
<title>Highlights</title>
<p>The second version of <code>WordCount</code> improves upon the
previous one by using some features offered by the MapReduce framework:
</p>
<ul>
<li>
Demonstrates how applications can access configuration parameters
in the <code>configure</code> method of the <code>Mapper</code> (and
<code>Reducer</code>) implementations (lines 28-43).
</li>
<li>
Demonstrates how the <code>DistributedCache</code> can be used to
distribute read-only data needed by the jobs. Here it allows the user
to specify word-patterns to skip while counting (line 104).
</li>
<li>
Demonstrates the utility of the <code>Tool</code> interface and the
<code>GenericOptionsParser</code> to handle generic Hadoop
command-line options (lines 87-116, 119).
</li>
<li>
Demonstrates how applications can use <code>Counters</code> (line 68)
and how they can set application-specific status information via
the <code>Reporter</code> instance passed to the <code>map</code> (and
<code>reduce</code>) method (line 72).
</li>
</ul>
</section>
</section>
<p>
<em>Java and JNI are trademarks or registered trademarks of
Sun Microsystems, Inc. in the United States and other countries.</em>
</p>
</body>
</document>