Initial Setup

All instructions pertain to CDH3u3 on CentOS 6.

Place hadoop-hbase-streaming.jar in /usr/local/hadoop-hbase-streaming.jar

Add to : /etc/hadoop-0.20/conf/hadoop-env.sh

export HADOOP_CLASSPATH="/usr/local/hadoop-hbase-streaming.jar:$HADOOP_CLASSPATH"
export HADOOP_CLASSPATH="/usr/lib/hbase/lib/guava-r06.jar:$HADOOP_CLASSPATH"
export HADOOP_CLASSPATH="/usr/lib/hbase/hbase-0.90.4-cdh3u3.jar:$HADOOP_CLASSPATH"
export HADOOP_CLASSPATH="/usr/lib/zookeeper/zookeeper-3.3.4-cdh3u3.jar:$HADOOP_CLASSPATH" 

Loading Data into HBase

Create the output table with appropriate column families:

create 'outputtable', {NAME=>'cf1'}, {NAME=>'cf2'}

Create a reducer that will output in the following format (tab-delimited):

put	<rowid>	<cf>:<qualifier>	<value>

Run your map reduce job with the OutputFormat set to: org.childtv.hadoop.hbase.mapred.ListTableOutputFormat

As a test, create a file called source_input/test.tab and include the expected reducer output.

An example of the reducer output might be (tab-delimited):

put	r1	cf1:test	Value1
put	r1	cf2:test	Value2
put	r2	cf1:test	Value3 

Then invoke the hadoop streaming API with the outputformat set to org.childtv.hadoop.hbase.mapred.ListTableOutputFormat and the job configuration parameter reduce.output.table=outputtable

hadoop jar  /usr/lib/hadoop-0.20/contrib/streaming/hadoop-streaming-0.20.2-cdh3u3.jar \
	-input source_input -output dummy_output \
	-mapper /bin/cat \
	-outputformat org.childtv.hadoop.hbase.mapred.ListTableOutputFormat \
	-jobconf reduce.output.table=outputtable 

This will write the provided fields to HBase.

Extracting Data from HBase

For reading from hbase, create a dummy input directory containing no files.

mkdir dummy_input

Select your desired InputFormat. Two exist : JSON: org.childtv.hadoop.hbase.mapred.JSONTableInputFormat Tabular values: org.childtv.hadoop.hbase.mapred.ListTableInputFormat

Select you desired input column families using the job configuration parameter map.input.columns

The JSON format has the advantage that the format is stricter and more expressive.

r1	{"cf2:test":{"timestamp":"1333428648468","value":"Value1"},"cf1:test":{"timestamp":"1333428678724","value":"Value2"}} 
r2	{"cf2:test":{"timestamp":"1333428656033","value":"Value3"},"cf1:test":{"timestamp":"1333428660721","value":"Value4"}} 

The ListTableInputFormat only includes rowid and value. It does not include column names in any way.

r1	Value1 Value2
r2	Value3 Value4

To run a test job on an HBase table called sourcetable with column families cf1 and cf2 and run:

hadoop jar /usr/lib/hadoop-0.20/contrib/streaming/hadoop-streaming-0.20.2-cdh3u3.jar \
	-input dummy_input -inputformat org.childtv.hadoop.hbase.mapred.JSONTableInputFormat \
	-mapper /bin/cat \
	-jobconf map.input.table=sourcetable -jobconf "map.input.columns=cf1 cf2" \
	-output myoutput 

This will produce a file in myoutput/part-00000 that contains the JSON output.