0% found this document useful (0 votes)

30 views

CH 8 and CH 9: Mapreduce Types, Formats and Features

The document provides an overview of MapReduce types, formats, and features. It discusses the general MapReduce model including the map, reduce, and optional combiner functions. It then covers input formats like file input, text input, binary input, and database input. It also discusses output formats, counters, sorting, joins, side data distribution, and common MapReduce library classes. The video demonstrates a MapReduce word count example program.

Uploaded by

noman vadsariya

Available Formats

Download as PPT, PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

30 views

CH 8 and CH 9: Mapreduce Types, Formats and Features

Uploaded by

noman vadsariya

Available Formats

Download as PPT, PDF, TXT or read online on Scribd

You are on page 1/ 15

Ch 8 and Ch 9:

MapReduce Types, Formats

and Features
MapReduce Form Review
General form of Map/Reduce functions:

map: (K1, V1) -> list(K2, V2)

reduce: (K2, list(V2)) -> list(K3, V3)

General form with Combiner function:

map: (K1, V1) -> list(K2, V2)

combiner: (K2, list(V2)) -> list(K2, V2)

reduce: (K2, list(V2)) -> list(K3, V3)

Partition function:

partition: (K2, V2) -> integer

Input Formats - Basics
Input split - a chunk of the input that is processed by a single map

Each map processes a single split, which is divided into records (key-value pair) that are individually processed by the map

Represented by Java class InputSplit

Set of storage locations (hostname strings)

Contains reference to the data not the actual data

InputFormat - responsible for creating input splits and dividing them into records so you will not directly deal with with the InputSplit class

Controlling split size

Usually the size of the HDFS block

Minimum size: 1 byte

Maximum size: Maximum value of Java long datatype

Split size formula: max(minimumSize, min(maximumSize, blockSize))

minimumSize < blockSize < maximumSize

Input Formats - Basics
Avoid small files - storing a large number of small files increases the numbers of seeks needed to run the job

A sequence file can be used to merge small files into larger files to avoid a large number of small files

Preventing splitting - you might want to prevent splitting if you want a single mapper to process each input file
as an entire file

1. Increase the minimum split size to be larger than the largest file in the system

2. Subclass the subclass of FileInputFormat to override the isSplitable() method to return false

Reading an entire file as a record:

RecordRecorder - deliver file contents as the value of the record, must implement createRecordReader() to create a custom
implementation of the class

WholeFileInputFormat
Input Formats - File Input
FileInputFormat - the base class for all implementations of InputFormat that use a file as the source for data

Provides a place to define what files are included as input to a job and an implementation for generating splits for the input files

Input is often specified as a collection of paths

Splits large files (larger than HDFS block)

CombineFileInputFormat - Java class designed to work well with small files in Hadoop
Each split will contain many of the small files so that each mapper has more to process

Takes node and rack locality into account when deciding what blocks to place into the same split

WholeFileInputFormat - defines a format where the keys are not used and the values are the file contents

Takes a FileSplit and converts it into a single record

Input Formats - Text Input
TextInputFormat - default InputFormat where each record is a line of input

Key - byte offset within the file of the beginning of the line; Value - the contents of the line, not including any line terminators, packaged as a
Text object

mapreduce.input.linerecordreader.line.maxlength - can be used to set a maximum expected line length

Safeguards against corrupted files (often appears as a very long line)

KeyValueTextInputFormat - Used to interpret TextOutputFormat (default output that contains key-value pairs
separated by a delimiter)

mapreduce.input.keyvaluelinerecordreader.key.value.separator - used to specify the delimiter/separator which is tab by default

NLineInputFormat - used when the mappers need to receive a fixed number of lines of input

mapreduce.input.line.inputformat.linespermap - controls the number of input lines (N)

StreamXmlRecordReader - used to break XML documents into records

Input Formats - Binary Input, Multiple Inputs, and Database I/O

Binary Input:
SequenceFileInputFormat - stores sequences of binary key-value pairs

SequenceFileAsTextInputFormat - converts sequence file’s keys and values to Text objects

SequenceFileAsBinaryInputFormat - retrieves the sequence file’s keys and values as binary objects

FixedLengthInputFormat - reading fixed-width binary records from a file where the records are not separated by delimiters

Multiple Inputs:
All input is interpreted by a single InputFormat and a single Mapper

MultipleInputs - allows programmer to specify which InputFormat and Mapper to use on a per-path basis

Database Input/Output:
DBInputFormat - input format for reading data from a relational database

DBOutputFormat - output format for outputting data from a relational database

Output Formats
Text Output: TextOutputFormat - default output format; writes records as lines of text (keys and values are turned into
strings)

KeyValueTextInputFormat - breaks lines into key-value pairs based on a configurable separator

Binary Output:

SequenceFileOutputFormat - writes sequence files as output

SequenceFileAsBinaryOutputFormat - writes keys and values in binary format into a sequence file container

MapFileOutputFormat - writes map files as output

Multiple Outputs:

MultipleOutputs - allows programmer to write data to files whose names are derived from output keys and values to create more than one file

Lazy Output: LazyOutputFormat - wrapper output format that ensures the output file is created only when the first record
is emitted for a given partition
Counters
Useful for gathering statistics about a job, quality-control, and problem diagnosis

Built-in Counter Types:

Task Counters - gather info about tasks as they are executed and results are aggregated over all job tasks

Maintained by each task attempt and are sent to the application manager on a regular basis to be globally aggregated

May go down if a task fails

Job Counters - measure job-level statistics and are maintained by the application master so they do not need to be sent across
the network

User-Defined Counters: User can define a set of counters to be incremented in a mapper/reducer

function

Dynamic counters (not defined by Java enum) can be created by the user
Sorting
Partial Sort - does not produce a globally- Total Sort - produces a globally-sorted output
sorted output file file

Produce a set of sorted files that can be

concatenated to form a globally-sorted file

To do this: use a partitioner that respects the

total order of the output and the partition
sizes must be fairly even

Secondary Sort - Sorts the values of the

keys

These are usually not sorted by MapReduce

Joins
MapReduce can perform joins between large datasets.

Ex:
Joins - Map-Side vs Reduce-Side
Map-Side Join Reduce-Side Join

the inputs must be divided into Input datasets do not have to be

the same number of structured in a particular way
partitions and sorted by the
same key (the join key) Results in records with the
same key being brought
All the records for a particular together in the reducer
key must reside in the same function
partition
Uses MultipleInputs and a
CompositeInputFormat can be secondary sort
used to run a map-side join
Side Data Distribution
Side Data - extra read-only data needed by a job to process the main dataset

The main challenge is to make side data available to all the map or reduce tasks (which are spread across the cluster) in way that is convenient and efficient