Util library to be able to use Avro files as input and output of Hadoop Map/Reduce Jobs or use Avro files as input of Hadoop Streaming.

Use com.tomslabs.grid.avro.AvroFileInputFormat to use Avro files as the input of Map/Reduce jobs.

'map()' will be called with a key of type Avro's GenericRecord and a NullWritable value.

Use com.tomslabs.grid.avro.AvroFileOuputFormat to use Avro files as the output of Map/Reduce jobs.

In reduce() method, the context must emit with a key of type Avro's GenericRecord and a NullWritable value. Please note that the Avro Schema for the output data MUST be specified when setting up the Job.

src/test/java/com/tomslabs/grid/avro/AvroWordCount.java is an example showing how to use Avro files for both input and output of a Map/Reduce jobs.

This example can be run as unit test from AvroWordCountTest.java.

To use Avro files as input for Hadoop Streaming, use the jar generated by the project and specify the correct input format:

$ $HADOOP_HOME/bin/hadoop  jar $HADOOP_HOME/hadoop-streaming.jar \
    -libjars ./avro-utils-<VERSION>.jar,avro-1.5.1.jar,avro-mapred-1.5.1.jar  \
    -inputformat com.tomslabs.grid.avro.AvroTextFileInputFormat \
    -input <Avro file or dir> \
    -output <output dir> \
    -mapper <map command> \
    -reducer <reducer command>

For example, to count the number of expression, something like:

$ $HADOOP_HOME/bin/hadoop  jar $HADOOP_HOME/hadoop-streaming.jar \
    -libjars ./avro-utils-<VERSION>.jar,avro-1.5.1.jar,avro-mapred-1.5.1.jar  \
    -inputformat com.tomslabs.grid.avro.AvroTextFileInputFormat \
    -input /tmp/word-count.avro \
    -output /tmp/out \
    -mapper /bin/cat \
    -reducer /usr/bin/wc

The format of each line streamed through Avro looks like:

<JSON representation of a Avro record>\t

(i.e. there is a trailing tabulation at the end of each line)

To use Avro files as input for Dumbo, use the jar generated by the project and set correct input format to com.tomslabs.grid.avro.AvroAsTextTypedBytesInputFormat:

$ dumbo start <PYTHON_SCRIPT> \
     -input /tmp/word-count.avro \
     -output /tmp/out \
     -libjar avro-1.5.1.jar \
     -libjars avro-mapred-1.5.1.jar  \
     -libjar avro-utils-<VERSION> \
     -inputformat com.tomslabs.grid.avro.AvroAsTextTypedBytesInputFormat
     -hadoop <HADOOP_HOME>
     -python <PYTHON_HOME>
     -outputformat text

The Python script's mapper will get the Avro record as a JSON string in its value parameter (the key parameter is not used).

You can use Avro files as the output for Dumbo. This expects that the reducer will emit as the value a string containing the JSON representation of an Avro record. To store in Avro (binary) files instead of text file, when you start Dumbo, you must specifiy the properties:

-libjar avro-1.5.1.jar \
-libjars avro-mapred-1.5.1.jar  \
-libjar avro-utils-<VERSION> \
-outputformat com.tomslabs.grid.avro.TextTypedBytesToAvroOutputFormat \
-hadoopconf avro.output.schema=$SCHEMA

where SCHEMA is a String containing the JSON representation of the Avro schema to use to create the Avro records.

You can use Avro files as the output for Hadoop Streaming. This expects to receive in the reducer a Text key containing the JSON representation of a Avro record. To store in Avro (binary) files instead of text file, you must specifiy the properties:

-D avro.output.schema=$SCHEMA \
-libjars ./avro-utils-<VERSION>.jar,avro-1.5.1.jar,avro-mapred-1.5.1.jar  \
-reducer com.tomslabs.grid.avro.JSONTextToAvroRecordReducer \
-outputformat org.apache.avro.mapred.AvroOutputFormat 

where SCHEMA is a String containing the JSON representation of the Avro schema to use to create the Avro records.

• http://avro.apache.org/
• http://wiki.apache.org/hadoop/HadoopStreaming
• https://github.com/klbostee/dumbo/wiki/