Avro Data Model

Introduction

Apache Avro is a data serialization framework. It is used in data serialization (especially in Hadoop ecosystem) and RPC protocols. It has libraries to support many languages. The library supports code generation with static languages like Java, while for dynamic languages for example python, code generation is not necessary.

When avro data is deserialized in Python environment, it was stored as a dictionary in memory. As a dictionary, it looses all the interesting features provided by the avro schema. For example, you can modify an integer field with a string without getting any errors. As a dictionary, it also doesn't provide any nice features from a normal class, for example, if an avro schema has firstName and lastName fields, it is not easy to define a fullName function to generate the full name.

Use Cases of the Library

In stream processing and RPC protocols, strict data types are required to make sure the system runs correctly. In Python, avro data is converted to a dictionary, which doesn't guarantee types and also doesn't provide a custom class hierarchy. I am looking to develop a way so that a class can be build on top of an avro schema, so that it can keep correct data type and also has a class structure.

My solution is similar to what SQLAlchemy ORM does. You need to manually create classes corresponding to avro schemas. However, fields of the avro schemas are all extracted from avsc file instead of being manually defined like SQLAlchemy. The classes allow defining methods to introduce new properties or new validations. Please check the following examples for how to use the library.

The purpose of the library is to bridge the gap between dynamical typed python and the use cases that requires strong types. This library should be restricted to places where static types are required. Otherwise, you will loose all the happiness playing with Python if applying this library everywhere.

Example

A Simple Example

User.avsc

{
  "type": "record",
  "name": "User",
  "fields": [
    {
      "name": "lastName",
      "type": "string"
    },
    {
      "name": "firstName",
      "type": "string"
    }
  ]
}

The following code defined a User class associated with the schema

@avro_schema(AvroDataNames(default_namespace="example.avro"), schema_file="User.avsc")
class User(object):
  def fullname(self):
    return "{} {}".format(self.firstName, self.lastName)

With this class definition, the full name can be obtained with the function call.

user = User({"firstName": "Alyssa", "lastName": "Yssa"})
print(user.fullname())
# Alyssa Yssa

Avro Schema with Extra Validation

In some use cases, some extra validations are required, for example: Date.avsc

{
  "name": "Date",
  "type": "record",
  "fields": [
    {
      "name": "year",
      "type": "int"
    },
    {
      "name": "month",
      "type": "int"
    },
    {
      "name": "day",
      "type": "int"
    }
  ]
}

The month and day of a date cannot be arbitrary integers. A extra validation can be done as following:

@avro_schema(AvroDataNames(default_namespace="example.avro"), schema_file="Date.avsc")
class Date(object):
  def __init__(self, value):
    if isinstance(value, datetime.date):
      value = {
          'year': value.year,
          'month': value.month,
          'day': value.day
      }
    super().__init__(value)

  def date(self):
    return datetime.date(self.year, self.month, self.day)

  def validate(self, data):
    return super().validate(data) \
        and datetime.date(data['year'], data['month'], data['day'])

The Date class can validate the input before assign it to then underlying avro schema

date = Date({"year": 2018, "month": 12, "date": 99})
# ValueError: day is out of range for month
date = Date(datetime.date(2018, 12, 12))
# No Error

Extract an avro schema defined in an outer schema

Sometimes an avro schema is defined in another schema Employee.avsc

{
  "type": "record",
  "name": "Employee",
  "namespace": "com.test",
  "fields": [
    {
      "name": "id"
      "type": "string"
    },
    {
      "name": "name",
      "type": {
        "type": "record",
        "name": "Name",
        "namespace": "com.test",
        "fields": [
          {
            "name": "lastName",
            "type": "string"
          },
          {
            "name": "firstName",
            "type": "string"
          }
        ]
      }
    }
  ]
}

The schema com.test.Name is defined in com.test.Employee. There is no Name.avsc, but you can still define a class for it the schema:

# Parent schema must be define first.
@avro_schema(
    EXAMPLE_NAMES,
    schema_file=os.path.join(DIRNAME, "Employee.avsc"))
class Employee(object):
    pass


# Full name is required
@avro_schema(EXAMPLE_NAMES, full_name="com.test.Name")
class Name(object):
    pass


name = Name({{"firstName": "Alyssa", "lastName": "Yssa"})
print(name)
# {'firstName': 'Alyssa', 'lastName': 'Yssa'}

Contributing

After cloning/forking the repo, navigate to the directory and run

source init.sh

The python environment should be ready for you.

Authors

• Kun Fang - (https://github.com/kun-fang)

See also the list of contributors who participated in this project.

License

This project is licensed under the MIT License - see the LICENSE.md file for details