Writing An Hadoop MapReduce Program In Python

In this tutorial, Michael will describe how to write a simple MapReduce program for Hadoop in the Python programming language.

Motivation

Even though the Hadoop framework is written in Java, programs for Hadoop need not to be coded in Java but can also be developed in other languages like Python or C++ (the latter since version 0.14.1). However, the documentation and the most prominent Python example on the Hadoop home page could make you think that youmust translate your Python code using Jython into a Java jar file. Obviously, this is not very convenient and can even be problematic if you depend on Python features not provided by Jython. Another issue of the Jython approach is the overhead of writing your Python program in such a way that it can interact with Hadoop – just have a look at the example in<HADOOP_INSTALL>/src/examples/python/WordCount.py and you see what I mean. I still recommend to have at least a look at the Jython approach and maybe even at the new C++ MapReduce API called Pipes, it’s really interesting.

Having that said, the ground is prepared for the purpose of this tutorial: writing a Hadoop MapReduce program in a more Pythonic way, i.e. in a way you should be familiar with.

What we want to do

We will write a simple MapReduce program (see also Wikipedia) for Hadoop in Python but without using Jython to translate our code to Java jar files.

Our program will mimick the WordCount example, i.e. it reads text files and counts how often words occur. The input is text files and the output is text files, each line of which contains a word and the count of how often it occured, separated by a tab.

Note: You can also use programming languages other than Python such as Perl or Ruby with the “technique” described in this tutorial. I wrote some words about what happens behind the scenes. Feel free to correct me if I’m wrong.

Prerequisites

You should have an Hadoop cluster up and running because we will get our hands dirty. If you don’t have a cluster yet, my following tutorials might help you to build one. The tutorials are tailored to Ubuntu Linux but the information does also apply to other Linux/Unix variants.

Python MapReduce Code

The “trick” behind the following Python code is that we will use HadoopStreaming (see also the wiki entry) for helping us passing data between our Map and Reduce code via STDIN (standard input) and STDOUT (standard output). We will simply use Python’s sys.stdin to read input data and print our own output to sys.stdout. That’s all we need to do because HadoopStreaming will take care of everything else! Amazing, isn’t it? Well, at least I had a “wow” experience…

Map: mapper.py

Save the following code in the file /home/hadoop/mapper.py. It will read data from STDIN (standard input), split it into words and output a list of lines mapping words to their (intermediate) counts to STDOUT (standard output). The Map script will not compute an (intermediate) sum of a word’s occurences. Instead, it will output “<word> 1″ immediately – even though the <word> might occur multiple times in the input – and just let the subsequent Reduce step do the final sum count. Of course, you can change this behavior in your own scripts as you please, but we will keep it like that in this tutorial because of didactic reasons :-)

Make sure the file has execution permission (chmod +x /home/hadoop/mapper.py should do the trick) or you will run into problems.

#!/usr/bin/env python

import sys

# input comes from STDIN (standard input)
for line in sys.stdin:
    # remove leading and trailing whitespace
    line = line.strip()
    # split the line into words
    words = line.split()
    # increase counters
    for word in words:
        # write the results to STDOUT (standard output);
        # what we output here will be the input for the
        # Reduce step, i.e. the input for reducer.py
        #
        # tab-delimited; the trivial word count is 1
        print '%s\t%s' % (word, 1)

Reduce: reducer.py

Save the following code in the file /home/hadoop/reducer.py. It will read the results of mapper.py from STDIN (standard input), and sum the occurences of each word to a final count, and output its results to STDOUT (standard output).

Make sure the file has execution permission (chmod +x /home/hadoop/reducer.py should do the trick) or you will run into problems.

#!/usr/bin/env python

from operator import itemgetter
import sys

# maps words to their counts
word2count = {}

# input comes from STDIN
for line in sys.stdin:
    # remove leading and trailing whitespace
    line = line.strip()

    # parse the input we got from mapper.py
    word, count = line.split('\t', 1)
    # convert count (currently a string) to int
    try:
        count = int(count)
        word2count[word] = word2count.get(word, 0) + count
    except ValueError:
        # count was not a number, so silently
        # ignore/discard this line
        pass

# sort the words lexigraphically;
#
# this step is NOT required, we just do it so that our
# final output will look more like the official Hadoop
# word count examples
sorted_word2count = sorted(word2count.items(), key=itemgetter(0))

# write the results to STDOUT (standard output)
for word, count in sorted_word2count:
    print '%s\t%s'% (word, count)

Test your code (cat data | map | sort | reduce)

I recommend to test your mapper.py and reducer.py scripts locally before using them in a MapReduce job. Otherwise your jobs might successfully complete but there will be no job result data at all or not the results you would have expected. If that happens, most likely it was you (or me) who screwed up.

Here are some ideas on how to test the functionality of the Map and Reduce scripts.

 # very basic test
 hadoop@ubuntu:~$ echo "foo foo quux labs foo bar quux" | /home/hadoop/mapper.py
 foo     1
 foo     1
 quux    1
 labs    1
 foo     1
 bar     1
 quux    1
hadoop@ubuntu:~$ echo "foo foo quux labs foo bar quux" | /home/hadoop/mapper.py | sort | /home/hadoop/reducer.py
 bar     1
 foo     3
 labs    1
 quux    2
 # using one of the ebooks as example input
 # (see below on where to get the ebooks)
 hadoop@ubuntu:~$ cat /tmp/gutenberg/20417-8.txt | /home/hadoop/mapper.py
 The     1
 Project 1
 Gutenberg       1
 EBook   1
 of      1
 [...]
 (you get the idea)

Running the Python Code on Hadoop

Download example input data

We will use three ebooks from Project Gutenberg for this example:

Download each ebook as plain text files in us-ascii encoding and store the uncompressed files in a temporary directory of choice, for example /tmp/gutenberg.

hadoop@ubuntu:~$ ls -l /tmp/gutenberg/
 total 3592
 -rw-r--r-- 1 hadoop hadoop  674425 2007-01-22 12:56 20417-8.txt
 -rw-r--r-- 1 hadoop hadoop 1423808 2006-08-03 16:36 7ldvc10.txt
 -rw-r--r-- 1 hadoop hadoop 1561677 2004-11-26 09:48 ulyss12.txt
 hadoop@ubuntu:~$

Copy local example data to HDFS

Before we run the actual MapReduce job, we first have to copy the files from our local file system to Hadoop’s HDFS.

hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop dfs -copyFromLocal /tmp/gutenberg gutenberg
 hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop dfs -ls
 Found 1 items
 /user/hadoop/gutenberg  <dir>
 hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop dfs -ls gutenberg
 Found 3 items
 /user/hadoop/gutenberg/20417-8.txt      <r 1>   674425
 /user/hadoop/gutenberg/7ldvc10.txt      <r 1>   1423808
 /user/hadoop/gutenberg/ulyss12.txt      <r 1>   1561677

Run the MapReduce job

Now that everything is prepared, we can finally run our Python MapReduce job on the Hadoop cluster. As I said above, we useHadoopStreaming for helping us passing data between our Map and Reduce code via STDIN (standard input) and STDOUT (standard output).

hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop jar contrib/streaming/hadoop-0.20.0-streaming.jar -file /home/hadoop/mapper.py -mapper /home/hadoop/mapper.py -file /home/hadoop/reducer.py -reducer /home/hadoop/reducer.py -input gutenberg/* -output gutenberg-output

If you want to modify some Hadoop settings on the fly like increasing the number of Reduce tasks, you can use the "-jobconf"option:

 hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop jar contrib/streaming/hadoop-0.20.0-streaming.jar -jobconf mapred.reduce.tasks=16 ...

An important note about mapred.map.tasks: Hadoop does not honor mapred.map.tasks beyond considering it a hint. But it accepts the user specified mapred.reduce.tasks and doesn’t manipulate that. You cannot force mapred.map.tasks but can specify mapred.reduce.tasks.

The job will read all the files in the HDFS directory gutenberg, process it, and store the results in a single result file in the HDFS directory gutenberg-output.

Exemplary output of the previous command in the console:

hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop jar contrib/streaming/hadoop-0.19.1-streaming.jar -mapper /home/hadoop/mapper.py -reducer /home/hadoop/reducer.py -input gutenberg/* -output gutenberg-output
 additionalConfSpec_:null
 null=@@@userJobConfProps_.get(stream.shipped.hadoopstreaming
 packageJobJar: [/usr/local/hadoop-datastore/hadoop-hadoop/hadoop-unjar54543/]
 [] /tmp/streamjob54544.jar tmpDir=null
 [...] INFO mapred.FileInputFormat: Total input paths to process : 7
 [...] INFO streaming.StreamJob: getLocalDirs(): [/usr/local/hadoop-datastore/hadoop-hadoop/mapred/local]
 [...] INFO streaming.StreamJob: Running job: job_200803031615_0021
 [...]
 [...] INFO streaming.StreamJob:  map 0%  reduce 0%
 [...] INFO streaming.StreamJob:  map 43%  reduce 0%
 [...] INFO streaming.StreamJob:  map 86%  reduce 0%
 [...] INFO streaming.StreamJob:  map 100%  reduce 0%
 [...] INFO streaming.StreamJob:  map 100%  reduce 33%
 [...] INFO streaming.StreamJob:  map 100%  reduce 70%
 [...] INFO streaming.StreamJob:  map 100%  reduce 77%
 [...] INFO streaming.StreamJob:  map 100%  reduce 100%
 [...] INFO streaming.StreamJob: Job complete: job_200803031615_0021
 [...] INFO streaming.StreamJob: Output: gutenberg-output
hadoop@ubuntu:/usr/local/hadoop$

As you can see in the output above, Hadoop also provides a basic web interface for statistics and information. When the Hadoop cluster is running, go to http://localhost:50030/ and browse around. Here’s a screenshot of the Hadoop web interface for the job we just ran.

A screenshot of Hadoop's web interface, showing the details of the MapReduce job we just ran.

Check if the result is successfully stored in HDFS directory gutenberg-output:

hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop dfs -ls gutenberg-output
 Found 1 items
 /user/hadoop/gutenberg-output/part-00000     <r 1>   903193  2007-09-21 13:00
 hadoop@ubuntu:/usr/local/hadoop$

You can then inspect the contents of the file with the dfs -cat command:

hadoop@ubuntu:/usr/local/hadoop$ bin/hadoop dfs -cat gutenberg-output/part-00000
 "(Lo)cra"       1
 "1490   1
 "1498," 1
 "35"    1
 "40,"   1
 "A      2
 "AS-IS".        2
 "A_     1
 "Absoluti       1
 [...]
 hadoop@ubuntu:/usr/local/hadoop$

Note that in this specific output above the quote signs (“) enclosing the words have not been inserted by Hadoop. They are the result of how our Python code splits words, and in this case it matched the beginning of a quote in the ebook texts. Just inspect the part-00000 file further to see it for yourself.

Improved Mapper and Reducer code: using Python iterators and generators

The Mapper and Reducer examples above should have given you an idea of how to create your first MapReduce application. The focus was code simplicity and ease of understanding, particularly for beginners of the Python programming language. In a real-world application however, you might want to optimize your code by using Python iterators and generators (an even better introduction in PDF) as some readers have pointed out.

Generally speaking, iterators and generators (functions that create iterators, for example with Python’s yield statement) have the advantage that an element of a sequence is not produced until you actually need it. This can help a lot in terms of computational expensiveness or memory consumption depending on the task at hand.

Note: The following Map and Reduce scripts will only work “correctly” when being run in the Hadoop context, i.e. as Mapper and Reducer in a MapReduce job. This means that running the naive test “cat DATA | ./mapper.py | sort | ./reducer.py” will not work correctly anymore because some functionality is intentionally outsourced to Hadoop.

Precisely, we compute the sum of a word’s occurences, e.g. (“foo”, 4), only if by chance the same word (“foo”) appears multiple times in succession. In the majority of cases, however, we let the Hadoop group the (key, value) pairs between the Map and the Reduce step because Hadoop is more efficient in this regard than our simple Python scripts.

mapper.py

#!/usr/bin/env python
"""A more advanced Mapper, using Python iterators and generators."""

import sys

def read_input(file):
    for line in file:
        # split the line into words
        yield line.split()

def main(separator='\t'):
    # input comes from STDIN (standard input)
    data = read_input(sys.stdin)
    for words in data:
        # write the results to STDOUT (standard output);
        # what we output here will be the input for the
        # Reduce step, i.e. the input for reducer.py
        #
        # tab-delimited; the trivial word count is 1
        for word in words:
            print '%s%s%d' % (word, separator, 1)

if __name__ == "__main__":
    main()

reducer.py

#!/usr/bin/env python
"""A more advanced Reducer, using Python iterators and generators."""

from itertools import groupby
from operator import itemgetter
import sys

def read_mapper_output(file, separator='\t'):
    for line in file:
        yield line.rstrip().split(separator, 1)

def main(separator='\t'):
    # input comes from STDIN (standard input)
    data = read_mapper_output(sys.stdin, separator=separator)
    # groupby groups multiple word-count pairs by word,
    # and creates an iterator that returns consecutive keys and their group:
    #   current_word - string containing a word (the key)
    #   group - iterator yielding all ["<current_word>", "<count>"] items
    for current_word, group in groupby(data, itemgetter(0)):
        try:
            total_count = sum(int(count) for current_word, count in group)
            print "%s%s%d" % (current_word, separator, total_count)
        except ValueError:
            # count was not a number, so silently discard this item
            pass

if __name__ == "__main__":
    main()

Related Links

From quuxlabs:

Leave a Reply

Your email address will not be published. Required fields are marked *

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>