airawat

This gist demonstrates how to do a map-side join, loading one small dataset from DistributedCache into a HashMap
in memory, and joining with a larger dataset.
 
Includes:
---------
1. Input data and script download
2. Dataset structure review
3. Expected results
4. Mapper code
5. Driver code

airawat

This gist demonstrates how to do a map-side join, joining a MapFile from distributedcache
with a larger dataset in HDFS.
 
Includes:
---------
1. Input data and script download
2. Dataset structure review
3. Expected results
4. Mapper code
5. Driver code

airawat

This gist is part of a series of gists related to Map-side joins in Java map-reduce.
In the gist - https://gist.github.com/airawat/6597557, we added the reference data available 
in HDFS to the distributed cache from the driver code.

This gist demonstrates adding a local file via command line to distributed cache.
Refer gist at https://gist.github.com/airawat/6597557 for-
1.  Data samples and structure
2.  Expected results
3.  Commands to load data to HDFS

airawat

Secondary sort in Mapreduce

With mapreduce framework, the keys are sorted but the values associated with each key 
are not.  In order for the values to be sorted, we need to write code to perform what is 
referred to a secondary sort.  The sample code in this gist demonstrates such a sort.

The input to the program is a bunch of employee attributes.
The output required is department number (deptNo) in ascending order, and the employee last name, 
first name and employee ID in descending order.

airawat

********************************
Gist
********************************

Motivation
-----------
The typical mapreduce job creates files with the prefix "part-"..and then the "m" or "r" depending
on whether it is a map or a reduce output, and then the part number.  There are scenarios where we
may want to create separate files based on criteria-data keys and/or values.  Enter the "MultipleOutputs" 
functionality.

airawat

**********************
Gist
**********************

A common interview question for a Hadoop developer position is whether we can control the number of 
mappers for a job.  We can - there are a few ways of controlling the number of mappers, as needed. 
Using NLineInputFormat is one way.

About NLineInputFormat
----------------------

airawat

*************************
Gist
*************************

One more gist related to controlling the number of mappers in a mapreduce task.

Background on Inputsplits
--------------------------
An inputsplit is a chunk of the input data allocated to a map task for processing.  FileInputFormat 
generates inputsplits (and divides the same into records) - one inputsplit for each file, unless the 

airawat

**********************
**Gist
**********************
 
This gist details how to inner join two large datasets on the map-side, leveraging the join capability 
in mapreduce.  Such a join makes sense if both input datasets are too large to qualify for distribution 
through distributedcache, and can be implemented if both input datasets can be joined by the join key 
and both input datasets are sorted in the same order, by the join key.
 
There are two critical pieces to engaging the join behavior: 

airawat

My blog has an introduction to reduce side join in Java map reduce-
http://hadooped.blogspot.com/2013/09/reduce-side-join-options-in-java-map.html

airawat

This gist covers the Oozie SSH action.
It includes components of a sample Oozie workflow application- scripts/code, 
sample data and commands;  Oozie actions covered: secure shell action, email 
action.

My blog has documentation, and highlights of a very basic sample program.
http://hadooped.blogspot.com/2013/10/apache-oozie-part-13-oozie-ssh-action_30.html
 
 
This gist includes: