r³ is a map reduce engine written in python using a redis backend. It's purpose is to be simple.
r³ has only three concepts to grasp: input streams, mappers and reducers.
The diagram below relates how they interact:
If the diagram above is a little too much to grasp right now, don't worry. Keep reading and use this diagram later for reference.
A fairly simple map-reduce example to solve is counting the number of occurrences of each word in an extensive document. We'll use this scenario as our example.
Installing r³ is as easy as:
pip install r3
After successful installation, you'll have three new commands: r3-app,
r3-map and r3-web.
In order to use r³ you must have a redis database running. Getting one up in your system is beyond the scope of this document.
We'll assume you have one running at 127.0.0.1, port 7778 and configured to require the password 'r3' using database 0.
The service that is at the heart of r³ is r3-app. It is the web-server that
will receive requests for map-reduce jobs and return the results.
To run r3-app, given the above redis back-end, type:
r3-app --redis-port=7778 --redis-pass=r3 -c config.py
We'll learn more about the configuration file below.
Given that you have a proper configuration file, your r3 service will be
available at http://localhost:9999.
As to how we actually perform a map-reduce operation, we'll see that after the
Running Mappers section.
In the above section we specified a file called config.py as configuration.
Now we'll see what that file contains.
The configuration file that we pass to the r3-app command is responsible for
specifying input stream processors and reducers that should be enabled.
Let's see a sample configuration file:
INPUT_STREAMS = [
'test.count_words_stream.CountWordsStream'
]
REDUCERS = [
'test.count_words_reducer.CountWordsReducer'
]
This configuration specifies that there should be a CountWordsStream input
stream processor and a CountWordsReducer reducer. Both will be used by the
stream service to perform a map-reduce operation.
We'll learn more about input streams and reducers in the sections below.
The input stream processor is the class responsible for creating the input streams upon which the mapping will occur.
In our counting words in a document sample, the input stream processor class
should open the document, read the lines in the document and then return each
line to r3-app.
Let's see a possible implementation:
from os.path import abspath, dirname, join
class CountWordsStream:
job_type = 'count-words'
group_size = 1000
def process(self, app, arguments):
with open(abspath(join(dirname(__file__), 'chekhov.txt'))) as f:
contents = f.readlines()
return [line.lower() for line in contents]
The job_type property is required and specifies the relationship that this
input stream has with mappers and with a specific reducer.
The group_size property specifies how big is an input stream. In the above
example, our input stream processor returns all the lines in the document, but
r³ will group the resulting lines in batches of 1000 lines to be processed by
each mapper. How big is your group size varies wildly depending on what your
mapping consists of.
Input stream processors and reducers are sequential and thus run in-process
in the r³ app. Mappers, on the other hand, are inherently parallel and are run
on their own as independent worker units.
Considering the above example of input stream and reducer, we'll use a
CountWordsMapper class to run our mapper.
We can easily start the mapper with:
r3-map --redis-port=7778 --redis-pass=r3 --mapper-key=mapper-1 --mapper-class="test.count_words_mapper.CountWordsMapper"
The redis-port and redis-pass arguments require no further explanation.
The mapper-key argument specifies a unique key for this mapper. This key
should be the same once this mapper restarts.
The mapper-class is the class r³ will use to map input streams.
Let's see what this map class looks like. If we are mapping lines (what we got out of the input stream steap), we should return each word and how many times it occurs.
from r3.worker.mapper import Mapper
class CountWordsMapper(Mapper):
job_type = 'count-words'
def map(self, lines):
return list(self.split_words(lines))
def split_words(self, lines):
for line in lines:
for word in line.split():
yield word, 1
The job_type property is required and specifies the relationship that this
mapper has with a specific input stream and with a specific reducer.
After all input streams have been mapped, it is time to reduce our data to one coherent value. This is what the reducer does.
In the case of counting word occurrences, a sample implementation is as follows:
from collections import defaultdict
class CountWordsReducer:
job_type = 'count-words'
def reduce(self, app, items):
word_freq = defaultdict(int)
for line in items:
for word, frequency in line:
word_freq[word] += frequency
return word_freq
The job_type property is required and specifies the relationship that this
reducer has with mappers and with a specific input stream.
This reducer will return a dictionary that contains all the words and the frequency with which they occur in the given file.
To test the above solution, just clone r³'s repository and run the commands from the directory you just cloned.
Given that we have the above working, we should have r3-app running at
http://localhost:9999. In order to access our count-words job we'll point
our browser to:
http://localhost:9999/count-words
This should return a JSON document with the resulting occurrences of words in the sample document.
As you have probably guessed, creating new jobs of mapping and reducing is as
simple as implementing your own input stream processor, mapper and
reducer.
After they are implemented, just include the processor and reducer in the config file and fire up as many mappers as you want.
We talked about three available commands: r3-app, r3-map and r3-web.
The last one fires up a monitoring interface that helps you in understanding how your r³ farm is working.
Some screenshots of the monitoring application:
Failed jobs monitoring:
Stats:
r³ is licensed under the MIT License:
The MIT License
Copyright (c) 2012 Bernardo Heynemann [email protected]
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.