gogen-avro

README

gogen-avro

Generate Go structures and serializer / deserializer methods from Avro schemas. Generated serializers/deserializers are 2-8x faster than goavro, and you get compile-time safety for getting and setting fields.

Installation

gogen-avro is a tool which you install on your system (usually on your GOPATH), and run as part of your build process. To install gogen-avro to $GOPATH/bin/, run:

go get gopkg.in/alanctgardner/gogen-avro.v2/...

Usage

To generate Go source files from one or more Avro schema files, run:

gogen-avro [--container] [--package=<package name>] <output directory> <avro schema files>

You can also use a go:generate directive in a source file (example):

//go:generate $GOPATH/bin/gogen-avro . primitives.avsc

The generated source files contain structs for each schema, plus a function Serialize(io.Writer) to encode the contents into the given io.Writer, and Deserialize<RecordType>(io.Reader) to read a struct from the given io.Reader.

Container File Support

Container file support is still being worked on - please report any bugs you find

Supplying the --container flag generates a container file writer for each schema in addition to the serializers.

The container file writer should be constructed with an existing io.Writer, a Codec (Null, Snappy, or Deflate) and a block size in records. Records are encoded and buffered when you call writer.writeRecord(record), and synchronously flushed when the block size is hit. You can also manually flush a block by calling writer.Flush(). You must call writer.Flush() before closing the underlying io.Writer, to ensure all the records in the last block are written.

An example of how to write a container file can be found in example/container/example.go.

Example

The example directory contains simple example projects with an Avro schema. Once you've installed gogen-avro on your GOPATH, you can install the example projects:

# Build the Go source files from the Avro schema using the generate directive
go generate github.com/alanctgardner/gogen-avro/example

# Install the example projects on the gopath
go install github.com/alanctgardner/gogen-avro/example/record
go install github.com/alanctgardner/gogen-avro/example/container

Type Conversion

Gogen-avro produces a Go struct which reflects the structure of your Avro schema. Most Go types map neatly onto Avro types:

Avro Type	Go Type	Notes
null	interface{}	This is just a placeholder, nothing is encoded/decoded
boolean	bool
int, long	int32,int64
float, double	float32, float64
bytes	[]byte
string	string
enum	custom type	Generates a type with a constant for each symbol
array	[]
map	map[string]
fixed	[]byte	Fixed fields are given a custom type, which is an alias for an appropriately sized byte array
union	custom type	Unions are handled as a struct with one field per possible type, and an enum field to dictate which field to read

union is more complicated than primitive types. We generate a struct and enum whose name is uniquely determined by the types in the union. For a field whose type is ["null", "int"] we generate the following:

type UnionNullInt struct {
	// All the possible types the union could take on
	Null               interface{}
	Int                int32
	// Which field actually has data in it - defaults to the first type in the list, "null"
	UnionType          UnionNullTypeEnum
}

type UnionNullIntTypeEnum int

const (
	UnionNullIntTypeEnumNull            UnionNullIntTypeEnum = 0
	UnionNullIntTypeEnumInt             UnionNullIntTypeEnum = 1
)

TODO / Caveats

This package doesn't implement the entire Avro 1.7.7 specification, specifically:

• Schema resolution
• Framing - generate RPCs and container format readers/writers

Versioning

This tool is versioned using gopkg.in. The API is guaranteed to be stable within a release. This guarantee applies to:

• the public members of generated structures
• the public methods attached to generated structures
• the command-line arguments of the tool itself

Only bugfixes will be backported to existing major releases. This means that source files generated with the same major release may differ, but they will never break your build.

3.0

• Experimental support for writing object container files
• Improved variable and type names
• Support for custom package names as a command line argument

2.0

• Bug fixes for arrays and maps with record members
• Refactored internals significantly

1.0

• Initial release
• No longer supported - no more bugfixes are being backported

Thanks

Thanks to LinkedIn's goavro library, for providing the encoders for primitives.