How to write protobuf byte array from Flink to Kafka

Question

I'm new to Flink. All I want to do is put my protobuf POJO to kafka as byte array. So my FlinkKafkaProducer looks like this:

FlinkKafkaProducer<String> flinkKafkaProducer = createStringProducer(outputTopic, address);
        stringInputStream
                .map(//here returns byte[])
                .addSink(flinkKafkaProducer);

public static FlinkKafkaProducer<String> createStringProducer(String topic, String kafkaAddress) {
        return new FlinkKafkaProducer<>(kafkaAddress, topic, new SimpleStringSchema());
    }

And right now it works fine but my output is String. I've tryed to add TypeInformationSerializationSchema() instead of new SimpleStringSchema() to change output but I cant get how to adjust it correct. Cant find any tutorial. Could someone help?

Answer 1

So, I finally figure out how to write protobuf to kafka producer as byte array. The problem was with serialization. In case of POJO flink uses libery Kryo for custom de/serialization. Best way to write protobuf is use ProtobufSerializer.class . In this example I will read from kafka String message and write as byte array. Gradle dependencys:

 compile (group: 'com.twitter', name: 'chill-protobuf', version: '0.7.6'){
        exclude group: 'com.esotericsoftware.kryo', module: 'kryo'
    }
    implementation 'com.google.protobuf:protobuf-java:3.11.0'

Registration:

StreamExecutionEnvironment environment = StreamExecutionEnvironment.getExecutionEnvironment();
environment.getConfig().registerTypeWithKryoSerializer(MyProtobuf.class, ProtobufSerializer.class);

KafkaSerializerClass

@Data
@RequiredArgsConstructor
public class MyProtoKafkaSerializer implements KafkaSerializationSchema<MyProto> {
    private final String topic;
    private final byte[] key;

    @Override
    public ProducerRecord<byte[], byte[]> serialize(MyProto element, Long timestamp) {
                
        return new ProducerRecord<>(topic, key, element.toByteArray());
    }
}

Job

  public static FlinkKafkaProducer<MyProto> createProtoProducer(String topic, String kafkaAddress) {
        MyProtoKafkaSerializer myProtoKafkaSerializer = new MyProtoKafkaSerializer(topic);
        Properties props = new Properties();
        props.setProperty("bootstrap.servers", kafkaAddress);
        props.setProperty("group.id", consumerGroup);
        return new FlinkKafkaProducer<>(topic, myProtoKafkaSerializer, props, FlinkKafkaProducer.Semantic.AT_LEAST_ONCE);
    }

 public static FlinkKafkaConsumer<String> createProtoConsumerForTopic(String topic, String kafkaAddress, String kafkaGroup) {
        Properties props = new Properties();
        props.setProperty("bootstrap.servers", kafkaAddress);
        props.setProperty("group.id", kafkaGroup);
        return new FlinkKafkaConsumer<>(topic, new SimpleStringSchema(), props);
    }

DataStream<String> stringInputStream = environment.addSource(flinkKafkaConsumer);
        FlinkKafkaProducer<MyProto> flinkKafkaProducer = createProtoProducer(outputTopic, address);
        stringInputStream
                .map(hashtagMapFunction)
                .addSink(flinkKafkaProducer);

        environment.execute("My test job");

Sources :

1. https://ci.apache.org/projects/flink/flink-docs-release-1.10/dev/custom_serializers.html#register-a-custom-serializer-for-your-flink-program
2. https://flink.apache.org/news/2020/04/15/flink-serialization-tuning-vol-1.html#protobuf-via-kryo

Answer 2

It indeed seems tricky to find documentation in this matter. I'll assume you use Flink >= 1.9. In that case, the following should work:

private static class PojoKafkaSerializationSchema implements KafkaSerializationSchema<YourPojo> {
    @Override
    public void open(SerializationSchema.InitializationContext context) throws Exception {}

    @Override
    public ProducerRecord<byte[], byte[]> serialize(YourPojo element,@Nullable Long timestamp) {
        // serialize your POJO here and return a Kafka `ProducerRecord`
        return null;
    }
}

// Elsewhere: 
PojoKafkaSerializationSchema schema = new PojoKafkaSerializationSchema();
FlinkKafkaProducer<Integer> kafkaProducer = new FlinkKafkaProducer<>(
    "test-topic",
    schema,
    properties,
    FlinkKafkaProducer.Semantic.AT_LEAST_ONCE
);

This code is mostly inspired by this test case , but I didn't have time to actually run it.