With the current Hadoop, it is a bit hard for the user to implement data joining apps.
HADOOP-475/485 attempt to provide some support for data joining jobs, but it seems to be had to implement.
This Jira rather calls for a application level support.
The idea is to provide a generic map/reduce classes implementing data join jobs,
and allows the user to extend those classes to add their special logic.
In particular, the user needs to define a mapper class
that extends DataJoinMapperBase class to implement methods for the
1. Compute the source tag of input values
2. Compute the map output value object
3. Compute the map output key object
The source tag will be used by the reducer to determine from which source
(which table in SQL terminology) a value comes. Computing the map output
value object amounts to performing projecting/filtering work in a SQL
statement (through the select/where clauses). Computing the map output key
amounts to choosing the join key. This class provides the appropriate plugin
points for the user defined subclasses to implement the appropriate logic.
The the user needs to define a reducer class
that extends DataJoinReduceBase class to implement the following:
protected abstract TaggedMapOutput combine(Object tags, Object values);
The above method is expected to produce one output value from an array of
records of different sources. The user code can also perform filtering here.
It can return null if it decides to the records do not meet certain conditions.
That is pretty much the user need to do in order to create a map/reduce job to join data
from different sources.