===============================
Introduction
===============================

cascading.avro is a Cascading Scheme for the Apache Avro serialization format,
which has been publicly released by Bixo Labs under the Apache license.

This means you can use Avro as the source of tuples for a Cascading
flow, and as a sink for saving results. This is particularly useful when
you need to exchange data with other programming languages, as Avro is both
efficient and cross-language.

Information about Avro is available from http://hadoop.apache.org/avro/
and also http://wiki.apache.org/hadoop/Avro

===============================
Design
===============================

When you create an AvroSchema, you specify the fields and types (classes) of
each field. This lets the schema auto-generate a corresponding Avro scheme,
for both reading and writing data.

The set of supported types, with the corresponding Avro type, is:

	Integer.class, Schema.Type.INT
 	Long.class, Schema.Type.LONG
	Boolean.class, Schema.Type.BOOLEAN
	Double.class, Schema.Type.DOUBLE
	Float.class, Schema.Type.FLOAT
	String.class, Schema.Type.STRING
	BytesWritable.class, Schema.Type.BYTES
	List.class, Schema.Type.ARRAY
	Map.class, Schema.Type.MAP
	Enum.class, Schema.Type.ENUM
	
See below for how the List, Map and Enum types are actually represented in a Tuple as a
nested Tuple.

===============================
Example
===============================

	final Fields cascadingFields = new Fields("id", "timestamp", "statusMsg", "content");
	final Class<?>[] schemeTypes = {Integer.class, Long.class, String.class, BytesWritable.class};
	Tap avroSource = new Hfs(new AvroScheme(cascadingFields, schemeTypes), "outputdir");
  
  	...hook up your Cascading workflow here...
  	
===============================
Limitations
===============================

There's currently no way to set the metadata for the generated Avro file, nor is there
any way to read that data back in. This is now supported in Avro, we just need to hook
it up in the Scheme.

The support for lists and maps is rudimentary and kludgy. Since these aren't natively
supported by Cascading yet (as of 1.1), we imitate this by nesting Tuples into the Tuple.
So a list item is a Tuple of 0...n primitive values, and a map is a Tuple that contains
alternative key/value pairs. Note that Avro maps always use a String as the key, so that
same constraint exists on the Cascading side.

The support for enums assumes that the Tuple has stored the enum value as the string
returned by enum.toString(), such that enum.valueOf(string) will return the original enum.

There's no support for using anything other than primitive types as values, so you can't
leverage Avro support for nesting records/maps/arrays as fields in the Avro record. While
Cascading does support using arbitrary types for fields, as long as they can be serialized,
we do not and probably will not attempt to synthesize mappings of arbitrarily complex data
types between Cascading and Avro.

===============================
Known Issues
===============================

Avro doesn't work properly with standard Hadoop installations prior to 0.20, since
Hadoop includes an older version (1.0.1) of the Jackson jar that is required by Avro,
and Avro uses some newer APIs. So you'll have to sculpt the Hadoop setup to exclude
that older Jackson jar, or play games with the classloader to ensure the newer jar
gets used with your job.

===============================
Building
===============================

You need Apache Ant 1.7 or higher, and a git client.

1. Download source from GitHub

% git clone git://github.com/bixolabs/cascading.avro.git
% cd cascading.avro

2. Build the jar

% ant clean jar

or to build and install the jar in your local Maven repo:

% ant clean install

3. Create Eclipse project files

% ant eclipse

Then, from Eclipse follow the standard procedure to import an existing Java project into your Workspace.