MAPREDUCE

FUNDAMENTAL
CONCEPTS
Why MapReduce?

■ Distributes the processing of data on your cluster

■ Divides your data up into partitions that are MAPPED
(transformed) and REDUCED (aggregated) by mapper
and reducer functions you define
■ Resilient to failure – an application master monitors
your mappers and reducers on each partition
Let’s illustrate with an example

■ How many movies did each user rate in the MovieLens data set?
 How MapReduce Works: Mapping

■ The MAPPER converts raw source data into key/value pairs

INPUT DATA

Mapper

K1:V K2:V K3:V K1:V K1:V

Example: MovieLens Data (u.data file)
USER ID|MOVIE ID|RATING|TIMESTAMP

196 242 3 881250949

186 302 3 891717742
196 377 1 878887116
244 51 2 880606923
166 346 1 886397596
186 474 4 884182806
186 265 2 881171488
Map users to movies they watched

USER ID|MOVIE ID|RATING|TIMESTAMP

196 242 3 881250949
186 302 3 891717742
196 377 1 878887116
244 51 2 880606923
166 346 1 886397596
186 474 4 884182806
186 265 2 881171488

Mapper

196:242 186:302 196:377 244:51 166:346 186:274 186:265

Extract and Organize What We Care
About

196:242 186:302 196:377 244:51 166:346 186:474 186:265

 MapReduce Sorts and Groups the
Mapped Data (“Shuffle and Sort”)

196:242 186:302 196:377 244:51 166:346 186:474 186:265

166:346 186:302,474,265 196:242,377 244:51

The REDUCER Processes Each Key’s Values

166:346 186:302,474,265 196:242,377 244:51

len(movies)

166:1 186:3 196:2 244:1

Putting it All Together
USER ID|MOVIE ID|RATING|TIMESTAMP
196 242 3 881250949
186 302 3 891717742
196 377 1 878887116
244 51 2 880606923
166 346 1 886397596
186 474 4 884182806
186 265 2 881171488

MAPPER

196:242 186:302 196:377 244:51 166:346 186:474 186:265

SHUFFLE AND SORT

166:346 186:302,474,265 196:242,377 244:51

REDUCER

166:1 186:3 196:2 244:1

MAPREDUCE ON A
CLUSTER
How MapReduce Scales
Putting it All Together
USER ID|MOVIE ID|RATING|TIMESTAMP
196 242 3 881250949
186 302 3 891717742
196 377 1 878887116
244 51 2 880606923
166 346 1 886397596
186 474 4 884182806
186 265 2 881171488

MAPPER

196:242 186:302 196:377 244:51 166:346 186:274 186:265

SHUFFLE AND SORT

166:346 186:302,274,265 196:242,377 244:51

REDUCER

166:1 186:3 196:2 244:1

What’s Happening
NodeManager
YARN MapReduce
Client Node Resource Application
Manager Master

NodeManager Node NodeManager

MapTask / MapTask / MapTask /

HDFS ReduceTask ReduceTask ReduceTask
How are mappers and reducers written?

■ MapReduce is natively Java

■ STREAMING allows interfacing to other languages (ie Python)

MapTask /
ReduceTask

Key/values
stdin stdout

Streaming
Process
Handling Failure

■ Application master monitors worker tasks for errors or hanging

– Restarts as needed
– Preferably on a different node
■ What if the application master goes down?
– YARN can try to restart it
■ What if an entire Node goes down?
– This could be the application master
– The resource manager will try to restart it
■ What if the resource manager goes down?
– Can set up “high availability” (HA) using Zookeeper to
have a hot standby
MAPREDUCE: A REAL
EXAMPLE
How many of each rating type exist?
How many of each movie rating exist?
 Making it a MapReduce problem

■ MAP each input line to (rating, 1)

■ REDUCE each rating with the sum of all the 1’s

USER ID|MOVIE ID|RATING|TIMESTAMP

196 242 3 881250949 3,1 Shuffle
186 302 3 891717742 Map 3,1 & Sort 1 -> 1, 1 Reduce 1, 2
196 377 1 878887116 1,1 2 -> 1, 1 2, 2
244 51 2 880606923 2,1 3 -> 1, 1 3, 2
166 346 1 886397596 1,1 4 -> 1 4, 1
186 474 4 884182806 4,1
186 265 2 881171488 2,1
Writing the Mapper

USER ID|MOVIE ID|RATING|TIMESTAMP

196 242 3 881250949 3,1 Shuffle
186 302 3 891717742 Map 3,1 & Sort 1 -> 1, 1 Reduce 1, 2
196 377 1 878887116 1,1 2 -> 1, 1 2, 2
244 51 2 880606923 2,1 3 -> 1, 1 3, 2
166 346 1 886397596 1,1 4 -> 1 4, 1
186 474 4 884182806 4,1
186 265 2 881171488 2,1
Writing the Reducer

USER ID|MOVIE ID|RATING|TIMESTAMP

196 242 3 881250949 3,1 Shuffle
186 302 3 891717742 Map 3,1 & Sort 1 -> 1, 1 Reduce 1, 2
196 377 1 878887116 1,1 2 -> 1, 1 2, 2
244 51 2 880606923 2,1 3 -> 1, 1 3, 2
166 346 1 886397596 1,1 4 -> 1 4, 1
186 474 4 884182806 4,1
186 265 2 881171488 2,1
Putting it all together
 RUNNING
Run our MapReduce job in our Hadoop installation
MAPREDUCE WITH MRJOB
Installing what we need: HDP 2.5
■ PIP
– cd /etc/yum.repos.d
– cp sandbox.repo /tmp
– rm sandbox.repo
– cd ~
– yum install python-pip
■ MRJob
– pip install google-api-python-client==1.6.4
– pip install mrjob==0.5.11
■ Nano
– yum install nano
■ Data files and the script
– wget http://media.sundog-soft.com/hadoop/ml-100k/u.data
– wget http://media.sundog-soft.com/hadoop/RatingsBreakdown.py
Installing what we need: HDP 2.6.5

■ PIP
– Utility for installing Python packages
– yum install python-pip
■ MRJob
– pip install mrjob==0.5.11
■ Nano
– yum install nano
■ Data files and the script
– wget http://media.sundog-soft.com/hadoop/ml-
100k/u.data
– wget http://media.sundog-
soft.com/hadoop/RatingsBreakdown.py
Running with mrjob

■ Run locally
– python RatingsBreakdown.py u.data
■ Run with Hadoop
– python MostPopularMovie.py -r hadoop --hadoop-streaming-jar
/usr/hdp/current/hadoop-mapreduce-client/hadoop-streaming.jar
u.data
YOUR CHALLENGE
Sort movies by popularity with Hadoop
Challenge exercise

■ Count up ratings given for each movie

– All you need is to change one thing in the mapper – we don’t care about
ratings now, we care about movie ID’s!
– Start with this and make sure you can do it.
– You can use nano to just edit the existing RatingsBreakdown.py script
Stretch goal

■ Sort the movies by their numbers of ratings

■ Strategy:
– Map to (movieID, 1) key/value pairs
– Reduce with output of (rating count, movieID)
– Send this to a second reducer so we end up with things sorted by rating
count!
■ Gotchas:
– How do we set up more than one MapReduce step?
– How do we ensure the rating counts are sorted properly?
Multi-stage jobs

■ You can chain map/reduce stages together like this:

def steps(self):
return [
MRStep(mapper=self.mapper_get_ratings,
reducer=self.reducer_count_ratings),
MRStep(reducer=self.reducer_sorted_output)
]
Ensuring proper sorting

■ By default, streaming treats all input and output as strings. So things get
sorted as strings, not numerically.
■ There are different formats you can specify. But for now let’s just zero-pad
our numbers so they’ll sort properly.
■ The second reducer will look like this:
def reducer_count_ratings(self, key, values):
yield str(sum(values)).zfill(5), key
Iterating through the results

■ Spoiler alert!
def reducer_sorted_output(self, count, movies):
for movie in movies:
yield movie, count
CHECK YOUR RESULTS
Did it work?
My solution