如何使用 Apache Beam 反序列化 Kafka AVRO 消息

Question

The main goal is the aggregate two Kafka topics, one compacted slow moving data and the other fast moving data which is received every second.

I have been able to consume messages in simple scenarios such as a KV (Long,String) using something like:

PCollection<KV<Long,String>> input = p.apply(KafkaIO.<Long, 
String>read()
.withKeyDeserializer(LongDeserializer.class)
.withValueDeserializer(StringDeserializer.class)

PCollection<String> output = input.apply(Values.<String>create());

But this doesn't seem to be the approach when you need to deserialise from AVRO. I have a KV(STRING, AVRO) which I need to consume.

I attempted generating the Java Classes from the AVRO schema and then including them in the “apply” for example:

PCollection<MyClass> output = input.apply(Values.<MyClass>create());

But this didn't seem to be the correct approach.

Is there any documentation/examples anyone could point me to, so I could get an understanding as to how you would work with Kafka AVRO and Beam. Any help would be much appreciated.

I have updated my code:

import io.confluent.kafka.serializers.KafkaAvroDeserializer;
import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.coders.AvroCoder;
import org.apache.beam.sdk.io.kafka.KafkaIO;
import org.apache.beam.sdk.options.PipelineOptions;
import org.apache.beam.sdk.options.PipelineOptionsFactory;
import org.apache.beam.sdk.values.KV;
import org.apache.beam.sdk.values.PCollection;
import org.apache.kafka.common.serialization.LongDeserializer;

public class Main {

public static void main(String[] args) {

    PipelineOptions options = PipelineOptionsFactory.create();

    Pipeline p = Pipeline.create(options);

    PCollection<KV<Long, Myclass>> input = p.apply(KafkaIO.<Long, String>read()
            .withKeyDeserializer(LongDeserializer.class)
            .withValueDeserializerAndCoder(KafkaAvroDeserializer.class, AvroCoder.of(Myclass.class))
    );

    p.run();

}
}
#######################################################
import org.apache.beam.sdk.coders.AvroCoder;
import org.apache.beam.sdk.coders.DefaultCoder;

@DefaultCoder(AvroCoder.class)
public class Myclass{
String name;
String age;

Myclass(){}
Myclass(String n, String a) {
    this.name= n;
    this.age= a;
}
}

But i now get the following error incompatible types: java.lang.Class < io.confluent.kafka.serializers.KafkaAvroDeserializer > cannot be converted to java.lang.Class < ? extends org.apache.kafka.common.serialization.Deserializer < java.lang.String > >

I must be importing the incorrect serializers?

Answer 1

You can use KafkaAvroDeserializer as following:

PCollection<KV<Long,MyClass>> input = p.apply(KafkaIO.<Long, String>read()
.withKeyDeserializer(LongDeserializer.class)
  .withValueDeserializerAndCoder(KafkaAvroDeserializer.class, AvroCoder.of(MyClass.class))

Where MyClass is the POJO class generated Avro Schema.

Make sure your POJO class has annotation AvroCoder as in below example :

@DefaultCoder(AvroCoder.class)
   public class MyClass{
      String name;
      String age;

      MyClass(){}
      MyClass(String n, String a) {
         this.name= n;
         this.age= a;
      }
  }

Answer 2

I have faced the same issue. Found the solution in this mail-archives. http://mail-archives.apache.org/mod_mbox/beam-user/201710.mbox/%3CCAMsy_NiVrT_9_xfxOtK1inHxb=x_yAdBcBN+4aquu_hn0GJ0nA@mail.gmail.com%3E

In your case, you need to defined your own KafkaAvroDeserializer like as follows.

public class MyClassKafkaAvroDeserializer extends
  AbstractKafkaAvroDeserializer implements Deserializer<MyClass> {

  @Override
  public void configure(Map<String, ?> configs, boolean isKey) {
      configure(new KafkaAvroDeserializerConfig(configs));
  }

  @Override
  public MyClass deserialize(String s, byte[] bytes) {
      return (MyClass) this.deserialize(bytes);
  }

  @Override
  public void close() {} }

Then specify your KafkaAvroDeserializer as ValueDeserializer.

p.apply(KafkaIO.<Long, MyClass>read()
 .withKeyDeserializer(LongDeserializer.class)
 .withValueDeserializer(MyClassKafkaAvroDeserializer.class) );

Answer 3

Change KafkaIO.<Long, String>read() to KafkaIO.<Long, Object>read() .

If you look into the implementation of KafkaAvroDeserializer, it implements Deserializer:

public class KafkaAvroDeserializer extends AbstractKafkaAvroDeserializer implements Deserializer<Object>

Answer 4

Yohei's answer is good, but I also found this to work

import io.confluent.kafka.streams.serdes.avro.SpecificAvroDeserializer;

...

public static class CustomKafkaAvroDeserializer extends SpecificAvroDeserializer<MyCustomClass> {}

...
.withValueDeserializerAndCoder(CustomKafkaAvroDeserializer.class, AvroCoder.of(MyCustomClass.class))
...

where MyCustomClass is code gen'd with Avro tools.

Answer 5

I had a similar issue today, and came across the following example which resolved it for me.

https://github.com/andrewrjones/debezium-kafka-beam-example/blob/master/src/main/java/com/andrewjones/KafkaAvroConsumerExample.java

the missing piece for me was (Class)KafkaAvroDeserializer

KafkaIO.<String, MyClass>read()
        .withBootstrapServers("kafka:9092")
        .withTopic("dbserver1.inventory.customers")
        .withKeyDeserializer(StringDeserializer.class)
        .withValueDeserializerAndCoder((Class)KafkaAvroDeserializer.class, AvroCoder.of(MyClass.class))