Build Status Gem Version

Avromatic generates Ruby models from Avro schemas and provides utilities to encode and decode them.

This README reflects Avromatic 2.0. Please see the 1-0-stable branch for Avromatic 1.0.


Add this line to your application's Gemfile:

gem 'avromatic'

And then execute:

$ bundle

Or install it yourself as:

$ gem install avromatic

See the Logical Types section below for details on using Avromatic with unreleased Avro features.



Avromatic supports the following configuration:

Model Generation

  • schema_store: A schema store is required to load Avro schemas from the filesystem. It should be an object that responds to find(name, namespace = nil) and returns an Avro::Schema object. An AvroTurf::SchemaStore can be used. The schema_store is unnecessary if models are generated directly from Avro::Schema objects. See Models.
  • nested_models: An optional ModelRegistry that is used to store, by full schema name, the generated models that are embedded within top-level models. By default a new Avromatic::ModelRegistry is created.
  • eager_load_models: An optional array of models, or strings with class names for models, that are added to nested_models at the end of Avromatic.configure and during code reloading in Rails applications. This option is useful for defining models that will be extended when the load order is important.
  • allow_unknown_attributes: Optionally allow model constructors to silently ignore unknown attributes. Defaults to false. WARNING: Setting this to true will result in incorrect union member coercions if an earlier union member is satisfied by a subset of the latter union member's attributes.

Custom Types

See the section below on configuring Custom Types.

Using a Schema Registry/Messaging API

The configuration options below are required when using a schema registry (see Confluent Schema Registry) and the Messaging API.

  • schema_registry: An AvroSchemaRegistry::Client or AvroTurf::ConfluentSchemaRegistry object used to store Avro schemas so that they can be referenced by id. Either schema_registry or registry_url must be configured.
  • registry_url: URL for the schema registry. Either schema_registry or registry_url must be configured.
  • use_schema_fingerprint_lookup: Avromatic supports a Schema Registry extension that provides an endpoint to lookup existing schema ids by fingerprint. A successful response from this GET request can be cached indefinitely. The use of this additional endpoint can be disabled by setting this option to false and this is recommended if using a Schema Registry that does not support the endpoint.
  • messaging: An AvroTurf::Messaging object to be shared by all generated models The build_messaging! method may be used to create a Avromatic::Messaging instance based on the other configuration values.
  • logger: The logger to use for the schema registry client.

Example using a schema registry:

Avromatic.configure do |config|
  config.schema_store = AvroTurf::SchemaStore.new(path: 'avro/schemas')
  config.registry_url = Rails.configuration.x.avro_schema_registry_url


  • use_custom_datum_reader: Avromatic includes a modified subclass of Avro::IO::DatumReader. This subclass returns additional information about the index of union members when decoding Avro messages. This information is used to optimize model creation when decoding. By default this information is included in the hash returned by the DatumReader but can be omitted by setting this option to false.


  • use_custom_datum_writer: Avromatic includes a modified subclass of Avro::IO::DatumWriter. This subclass supports caching avro encodings for immutable models and uses additional information about the index of union members to optimize the encoding of Avro messages. By default this information is included in the hash passed to the encoder but can be omitted by setting this option to false.


Models are defined based on an Avro schema for a record.

The Avro schema can be specified by name and loaded using the schema store:

class MyModel
  include Avromatic::Model.build(schema_name :my_model)

# Construct instances by passing in a hash of attributes
instance = MyModel.new(id: 123, name: 'Tesla Model 3', enabled: true)

# Access attribute values with readers
instance.name # => "Tesla Model 3"

# Models are immutable by default
instance.name = 'Tesla Model X' # => NoMethodError (private method `name=' called for #<MyModel:0x00007ff711e64e60>) 

# Booleans can also be accessed by '?' readers that coerce nil to false
instance.enabled? # => true

# Models implement ===, eql? and hash
instance == MyModel.new(id: 123, name: 'Tesla Model 3', enabled: true) # => true
instance.eql?(MyModel.new(id: 123, name: 'Tesla Model 3', enabled: true)) # => true
instance.hash # => -1279155042741869898

# Retrieve a hash of the model's attributes via to_h, to_hash or attributes
instance.to_h # => {:id=>123, :name=>"Tesla Model 3", :enabled=>true}

Or an Avro::Schema object can be specified directly:

class MyModel
  include Avromatic::Model.build(schema: schema_object)

Models are generated as immutable value objects by default, but can optionally be defined as mutable:

class MyModel
  include Avromatic::Model.build(schema_name :my_model, mutable: true)

Generated models include attributes for each field in the Avro schema including any default values defined in the schema.

A model may be defined with both a key and a value schema:

class MyTopic
  include Avromatic::Model.build(value_schema_name: :topic_value,
                                 key_schema_name: :topic_key)

When key and value schemas are both specified, attributes are added to the model for the union of the fields in the two schemas.

By default, optional fields are not allowed in key schemas since their values may be accidentally omitted leading to problems if data is partitioned based on the key values.

This behavior can be overridden by specifying the :allow_optional_key_fields option for the model:

class MyTopic
  include Avromatic::Model.build(value_schema_name: :topic_value,
                                 key_schema_name: :topic_key,
                                 allow_optional_key_fields: true)

A model can also be generated as an anonymous class that can be assigned to a constant:

MyModel = Avromatic::Model.model(schema_name :my_model)

Experimental: Union Support

Avromatic contains experimental support for unions containing more than one non-null member type. This feature is experimental because Avromatic may attempt to coerce between types too aggressively.

For now, if a union contains nested models then it is recommended that you assign model instances.

Some combination of the ordering of member types in the union and relying on model validation may be required so that the correct member is selected, especially when deserializing from Avro.

In the future, the type coercion used in the gem will be enhanced to better support the union use case.

Nested Models

Nested models are models that are embedded within top-level models generated using Avromatic. Normally these nested models are automatically generated.

By default, nested models are stored in Avromatic.nested_models. This is an Avromatic::ModelRegistry instance that provides access to previously generated nested models by the full name of their Avro schema.

#=> <model class>

The ModelRegistry can be customized to remove a namespace prefix:

Avromatic.nested_models =
  Avromatic::ModelRegistry.new(remove_namespace_prefix: 'com.my_company')

The :remove_namespace_prefix value can be a string or a regexp.

By default, top-level generated models reuse Avromatic.nested_models. This allows nested models to be shared across different generated models. A :nested_models option can be specified when generating a model. This allows the reuse of nested models to be scoped:

Avromatic::Model.model(schema_name, :my_model
                       nested_models: ModelRegistry.new)

Only models without a key schema can be used as nested models. When a model is generated with just a value schema then it is automatically registered so that it can be used as a nested model.

To extend a model that will be used as a nested model, you must ensure that it is defined, which will register it, prior it being referenced by another model.

Using the Avromatic.eager_load_models option allows models that are extended and will be used as nested models to be defined at the end of the .configure block. In Rails applications, these models are also re-registered after nested_models is cleared when code reloads to ensure that classes load in the correct order:

Avromatic.configure do |config|
  config.eager_load_models = [
    # reference any extended models that should be defined first

Custom Type Configuration

Custom types can be configured for fields of named types (record, enum, fixed). These customizations are registered on the Avromatic module. Once a custom type is registered, it is used for all models with a schema that references that type. It is recommended to register types within a block passed to Avromatic.configure:

Avromatic.configure do |config|
  config.register_type('com.example.my_string', MyString)

The full name of the type and an optional class may be specified. When a class is provided then values for attributes of that type are defined using the specified class.

If the provided class responds to the class methods from_avro and to_avro then those methods are used to convert values when assigning to the model and before encoding using Avro respectively.

from_avro and to_avro methods may be also be specified as Procs when registering the type:

Avromatic.configure do |config|
  config.register_type('com.example.updown_string') do |type|
    type.from_avro = ->(value) { value.upcase }
    type.to_avro = ->(value) { value.downcase }

Nil handling is not required as the conversion methods are not be called if the inbound or outbound value is nil.

If a custom type is registered for a record-type field, then any to_avro method/Proc should return a Hash with string keys for encoding using Avro.

Encoding and Decoding

Avromatic provides two different interfaces for encoding the key (optional) and value associated with a model.

Manually Managed Schemas

The attributes for the value schema used to define a model can be encoded using:

encoded_value = model.avro_raw_value

In order to decode this data, a copy of the value schema is required.

If a model also has an Avro schema for a key, then the key attributes can be encoded using:

encoded_key = model.avro_raw_key

If attributes were encoded using the same schema(s) used to define a model, then the data can be decoded to create a new model instance:

MyModel.avro_raw_decode(key: encoded_key, value: encoded_value)

If the attributes where encoded using a different version of the model's schemas, then a new model instance can be created by also providing the schemas used to encode the data:

MyModel.avro_raw_decode(key: encoded_key,
                        key_schema: writers_key_schema,
                        value: encoded_value,
                        value_schema: writers_value_schema)

Messaging API

The other interface for encoding and decoding attributes uses the AvroTurf::Messaging API. This interface leverages a schema registry and prefixes the encoded data with an id to identify the schema. In this approach, a schema registry is used to ensure that the correct schemas are available during decoding.

The attributes for the value schema can be encoded with a schema id prefix using:

message_value = model.avro_message_value

If a model has an Avro schema for a key, then those attributes can also be encoded prefixed with a schema id:

message_key = model.avro_message_key

A model instance can be created from a key and value encoded in this manner:

MyTopic.avro_message_decode(message_key, message_value)

Or just a value if only one schema is used:


The schemas associated with a model can also be added to a schema registry without encoding a message:



A stream of messages encoded from various models using the messaging approach can be decoded using Avromatic::Model::MessageDecoder. The decoder must be initialized with the list of models to decode:

decoder = Avromatic::Model::MessageDecoder.new(MyModel1, MyModel2)

decoder.decode(model1_messge_key, model1_message_value)
# => instance of MyModel1
# => instance of MyModel2

Validations and Coercions

An exception will be thrown if an attribute value cannot be coerced to the corresponding Avro schema field's type. The following coercions are supported:

Ruby Type Avro Type
String, Symbol string
Array array
Hash map
Integer, Float int
Integer long
Float float
Float double
String bytes
Date, Time, DateTime date
Time, DateTime timestamp-millis
Time, DateTime timestamp-micros
TrueClass, FalseClass boolean
NilClass null
Hash record

Validation of required fields is done automatically when serializing a model to Avro. It can also be done explicitly by calling the valid? or invalid? methods from the ActiveModel::Validations interface.

RSpec Support

This gem also includes an "avromatic/rspec" file that can be required to support using Avromatic with a fake schema registry during tests.

Requiring this file configures a RSpec before hook that directs any schema registry requests to a fake, in-memory schema registry and rebuilds the Avromatic::Messaging object for each example.


After checking out the repo, run bin/setup to install dependencies. Then, run rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to rubygems.org.


Bug reports and pull requests are welcome on GitHub at https://github.com/salsify/avromatic.


The gem is available as open source under the terms of the MIT License.