@ovotech/avro-stream NPM

Avro Stream

Serialize/deserialize kafka-node streams with avro data, using confluent schema-registry to hold the schemas.

Using

yarn add @ovotech/avro-stream

Where sourceStream is a node readble stream, producing kafka-node produce objects. With an additional "schema" key holding the avro schema.

If the schema for the topic does not exist inside the schema registry, it would be created. Unless the auto create topic has been set for kafka, it would not create the topic automatically. You'll need to create it yourself.

import { AvroSerializer, AvroProduceRequest } from '@ovotech/avro-stream';
import { ReadableMock } from 'stream-mock';
import { ProducerStream } from 'kafka-node';

const data: AvroProduceRequest[] = [
  {
    topic: 'migration-completed',
    partition: 0,
    key: 'some-key',
    schema: {
      type: 'record',
      name: 'TestSchema',
      fields: [{ name: 'accountId', type: 'string' }],
    },
    messages: [{ accountId: '6666666' }, { accountId: '5555555' }],
  },
];

const sourceStream = new ReadableMock(data, { objectMode: true });
const producerStream = new ProducerStream({ kafkaClient: { kafkaHost: 'localhost:29092' } });
const serializer = new AvroSerializer('http://localhost:8081');

sourceStream.pipe(serializer).pipe(producerStream);

For deserializing avro kafka events:

import { AvroDeserializer } from '@ovotech/avro-stream';
import { WritableMock } from 'stream-mock';
import { ConsumerGroupStream } from 'kafka-node';

const consumerStream = new ConsumerGroupStream(
  {
    kafkaHost: 'localhost:29092',
    groupId: 'my-group',
    encoding: 'buffer',
    fromOffset: 'earliest',
  },
  ['migration-completed'],
);
const deserializer = new AvroDeserializer('http://localhost:8081');
const sinkStream = new WritableMock({ objectMode: true });

consumerStream.pipe(deserializer).pipe(sinkStream);

Custom schema registry implementations

The way avro serialization for kafka works is to embed the schema id as the first 5 bytes of the buffer so the buffer becomes <id><avro serialized buffer>. For that to work we need a resolver service that can do id->schema for deserializing and schema->id to serializing kafka events.

The default provided resolver is SchemaRegistryresolver using confluent schema-registry but you can write your own:

import { AvroSerializer, AvroDeserializer, SchemaResolver } from '@ovotech/avro-stream';

class MyResolver implements SchemaResolver {
  async toId(topic: string, schema: Schema) {
    return ...
  }

  async fromId(id: number) {
    return ...
  }
}

const resolver = new MyResolver();
const serializer = new AvroSerializer(resolver);
const deserializer = new AvroDeserializer(resolver);

Passing avro schema options

Sometimes you'll want to pass some options to the creation of the avro type from the schema, for example to pass in logical type resolvers. You can do that with the second argument to the constructors.

import { AvroSerializer, AvroDeserializer } from '@ovotech/avro-stream';

const serializer new AvroSerializer('...', { logicalTypes: ... });
const deserializer new AvroDeserializer('...', { logicalTypes: ... });

Errors

AvroSerializer can emit an AvroSerializerError, and subsequently AvroDeserializer - AvroDeserializerError. They are as follows:

AvroSerializerError

Property	Description
message	Original error message
chunk	The event sent from the previous stream to be serialized (AvroProduceRequest)
encoding	The buffer encoding
originalError	The original error object that was triggered

AvroDeserializer

Property	Description
message	Original error message
chunk	The event sent from the previous stream to be deserialized from kafka-node
encoding	The buffer encoding
originalError	The original error object that was triggered

Example error handling:

import { AvroSerializer, AvroSerializerError } from '@ovotech/avro-stream';

const serializer new AvroSerializer('...');

serializer.on('error', (error: AvroSerializerError) => {
  console.log(error.chunk);
})

AvroTopicSender

You can use an AvroTopicSender to produce ad-hock kafka messages.

import { AvroSerializer, AvroTopicSender } from '@ovotech/avro-stream';
import { ProducerStream } from 'kafka-node';

interface Message {
  accountId: string;
}

const sender = new AvroTopicSender<Message>({
  topic: 'test-topic-1',
  partition: 0,
  key: 'key-1',
  schema: {
    type: 'record',
    name: 'TestSchema1',
    fields: [{ name: 'accountId', type: 'string' }],
  },
});

const producerStream = new ProducerStream({ kafkaClient: { kafkaHost: 'localhost:29092' } });
const serializer = new AvroSerializer('http://localhost:8081');

sender.pipe(serializer).pipe(producerStream);

sender.send({ accountId: '222' }, { accountId: '111' });
sender.close();

Mocks

If you are writing tests for your kafka node streams, you would want to mock a stream of serialized objects, but you have only some produce requests. That's where you can use the MockAvroSerializer and MockSchemaRegistryResolver.

import { DateType } from '@ovotech/avro-logical-types';
import { ReadableMock, WritableMock } from 'stream-mock';
import { AvroProduceRequest, MockAvroSerializer, MockSchemaRegistryResolver } from '@ovotech/avro-stream';

const sourceData: AvroProduceRequest[] = [];

const mockSchemaResolver = new MockSchemaRegistryResolver(sourceData);
const sourceStream = new ReadableMock(sourceData, { objectMode: true });
const sinkStream = new WritableMock({ objectMode: true });
const serializer = new MockAvroSerializer(mockSchemaResolver);

sourceStream.pipe(serializer).pipe(sinkStream);

You will get a serialized stream of kafka messages as if its coming from kafka-node itself.

Gotchas

A thing to be aware of is that node streams unpipe in an event of an error, which means that you'll need to provide your own error handling and repipe the streams if you want it to be resilient to errors.

Running the tests

The tests require a running schema registry service, kafka and zookeeper. This is setup easily with a docker-compose:

docker-compose up

Then you can run the tests with:

yarn test

Coding style (linting, etc) tests

Style is maintained with prettier and tslint

yarn lint

Deployment

Deployment is preferment by lerna automatically on merge / push to master, but you'll need to bump the package version numbers yourself. Only updated packages with newer versions will be pushed to the npm registry.