Reactive Streams for Kafka

If you have questions or are working on a pull request or just curious, please feel welcome to join the chat room:

Akka Streams connector for Apache Kafka.

Created and maintained by

New API: 0.11-M4

Supports Kafka 0.10.0.x

This version of akka-stream-kafka depends on Akka 2.4.8, Scala 2.11.8.

Available at Maven Central for Scala 2.11:

libraryDependencies += "com.typesafe.akka" %% "akka-stream-kafka" % "0.11-M4"

Changes

0.11-M4

Consumer implementation is completely rewritten. Now we support partition assignment for consumer groups. More details about it here. Also we may reuse one kafka connection in case of manual topic-partition assignment.

Consumer API slightly changed. In 0.11-M3 you set topic-partition to subscribe in ConsumerSettings. Now ConsumerSettings represents a kafka connection. The connection potentially can serve multiple Sources and you specify topic-partition for every source.

Example usage

Scala

Producer Settings:

import akka.kafka._
import akka.kafka.scaladsl._
import org.apache.kafka.common.serialization.StringSerializer
import org.apache.kafka.common.serialization.ByteArraySerializer

  val producerSettings = ProducerSettings(system, new ByteArraySerializer, new StringSerializer)
    .withBootstrapServers("localhost:9092")

Produce messages:

  Source(1 to 10000)
    .map(_.toString)
    .map(elem => new ProducerRecord[Array[Byte], String]("topic1", elem))
    .to(Producer.plainSink(producerSettings))

Produce messages in a flow:

  Source(1 to 10000)
    .map(elem => ProducerMessage.Message(new ProducerRecord[Array[Byte], String]("topic1", elem.toString), elem))
    .via(Producer.flow(producerSettings))
    .map { result =>
      val record = result.message.record
      println(s"${record.topic}/${record.partition} ${result.offset}: ${record.value} (${result.message.passThrough}")
      result
    }

Consumer Settings:

import akka.kafka._
import akka.kafka.scaladsl._
import org.apache.kafka.common.serialization.StringDeserializer
import org.apache.kafka.common.serialization.ByteArrayDeserializer
import org.apache.kafka.clients.consumer.ConsumerConfig

val consumerSettings = ConsumerSettings(system, new ByteArrayDeserializer, new StringDeserializer)
  .withBootstrapServers("localhost:9092")
  .withGroupId("group1")
  .withProperty(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, "earliest")

Consume messages and store a representation, including offset, in DB:

  db.loadOffset().foreach { fromOffset =>
    val subscription = Subscriptions.assignmentWithOffset(new TopicPartition("topic1", 1) -> fromOffset)
    Consumer.plainSource(consumerSettings, subscription)
      .mapAsync(1)(db.save)
  }

Consume messages at-most-once:

  Consumer.atMostOnceSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
    .mapAsync(1) { record =>
      rocket.launch(record.value)
    }

Consume messages at-least-once:

  Consumer.committableSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
    .mapAsync(1) { msg =>
      db.update(msg.value).flatMap(_ => msg.committableOffset.commitScaladsl())
    }

Connect a Consumer to Producer:

  Consumer.committableSource(consumerSettings.withClientId("client1"))
    .map(msg =>
      ProducerMessage.Message(new ProducerRecord[Array[Byte], String]("topic2", msg.value), msg.committableOffset))
    .to(Producer.commitableSink(producerSettings))

Consume messages at-least-once, and commit in batches:

  Consumer.committableSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
    .mapAsync(1) { msg =>
      db.update(msg.value).map(_ => msg.committableOffset)
    }
    .batch(max = 10, first => CommittableOffsetBatch.empty.updated(first)) { (batch, elem) =>
      batch.updated(elem)
    }
    .mapAsync(1)(_.commitScaladsl())

Reusable kafka consumer:

  //Consumer is represented by actor
  //Create new consumer
  val consumer: ActorRef = system.actorOf(KafkaConsumerActor.props(consumerSettings))

  //Manually assign topic partition to it
  val stream1 = Consumer
    .plainExternalSource[Array[Byte], String](consumer, Subscriptions.assignment(new TopicPartition("topic1", 1)))
    .via(business)
    .to(Sink.ignore)

  //Manually assign another topic partition
  val stream2 = Consumer
    .plainExternalSource[Array[Byte], String](consumer, Subscriptions.assignment(new TopicPartition("topic1", 2)))
    .via(business)
    .to(Sink.ignore)

Consumer group:

  //Consumer group represented as Source[(TopicPartition, Source[Messages])]
  val consumerGroup = Consumer.committablePartitionedSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
  //Process each assigned partition separately
  consumerGroup.map {
    case (topicPartition, source) =>
      source
        .via(business)
        .toMat(Sink.ignore)(Keep.both)
        .run()
  }
  .mapAsyncUnordered(maxPartitions)(_._2)

Additional examples are available in ConsumerExamples.scala

Java

Producer Settings:

import akka.kafka.*;
import akka.kafka.javadsl.*;
import org.apache.kafka.common.serialization.StringSerializer;
import org.apache.kafka.common.serialization.ByteArraySerializer;

final ProducerSettings<byte[], String> producerSettings = ProducerSettings
  .create(system, new ByteArraySerializer(), new StringSerializer())
  .withBootstrapServers("localhost:9092");

Produce messages:

Source.range(1, 10000)
  .map(n -> n.toString()).map(elem -> new ProducerRecord<byte[], String>("topic1", elem))
  .to(Producer.plainSink(producerSettings));

Produce messages in a flow:

Source.range(1, 10000)
  .map(n -> new ProducerMessage.Message<byte[], String, Integer>(
    new ProducerRecord<>("topic1", n.toString()), n))
  .via(Producer.flow(producerSettings))
  .map(result -> {
    ProducerRecord record = result.message().record();
    System.out.println(record);
    return result;
  });

Consumer Settings:

import akka.kafka.*;
import akka.kafka.javadsl.*;
import org.apache.kafka.common.serialization.StringSerializer;
import org.apache.kafka.common.serialization.ByteArraySerializer;
import org.apache.kafka.clients.consumer.ConsumerConfig;

final ConsumerSettings<byte[], String> consumerSettings =
    ConsumerSettings.create(system, new ByteArrayDeserializer(), new StringDeserializer())
  .withBootstrapServers("localhost:9092")
  .withGroupId("group1")
  .withProperty(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, "earliest");

Consume messages and store a representation, including offset, in DB:

db.loadOffset().thenAccept(fromOffset -> {
  Consumer.plainSource(
          consumerSettings,
          Subscriptions.assignmentWithOffset(new TopicPartition("topic1", 1), fromOffset)
  ).mapAsync(1, record -> db.save(record));
});

Consume messages at-most-once:

Consumer.atMostOnceSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
  .mapAsync(1, record -> rocket.launch(record.value()));

Consume messages at-least-once:

Consumer.committableSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
  .mapAsync(1, msg -> db.update(msg.value())
    .thenCompose(done -> msg.committableOffset().commitJavadsl()));

Connect a Consumer to Producer:

Consumer.committableSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
  .map(msg ->
    new ProducerMessage.Message<byte[], String, ConsumerMessage.Committable>(
        new ProducerRecord<>("topic2", msg.value()), msg.committableOffset()))
  .to(Producer.commitableSink(producerSettings));

Consume messages at-least-once, and commit in batches:

Consumer.committableSource(consumerSettings.withClientId("client1"), Subscriptions.topics("topic1"))
  .mapAsync(1, msg ->
    db.update(msg.value()).thenApply(done -> msg.committableOffset()))
  .batch(10,
    first -> ConsumerMessage.emptyCommittableOffsetBatch().updated(first),
    (batch, elem) -> batch.updated(elem))
  .mapAsync(1, c -> c.commitJavadsl());

Additional examples are available in ConsumerExamples.java

Configuration

The configuration properties are defined in reference.conf

Testing

Tests require Apache Kafka and Zookeeper to be available on localhost:9092 and localhost:2181. Note that auto.create.topics.enable should be true.

Old API: 0.10.0

Supports Kafka 0.9.0.x For Kafka 0.8 see this branch.

Available at Maven Central for Scala 2.11:

libraryDependencies += "com.softwaremill.reactivekafka" %% "reactive-kafka-core" % "0.10.0"

Example usage

Scala

import akka.actor.ActorSystem
import akka.stream.ActorMaterializer
import akka.stream.scaladsl.{Sink, Source}
import com.softwaremill.react.kafka.KafkaMessages._
import org.apache.kafka.common.serialization.{StringSerializer, StringDeserializer}
import com.softwaremill.react.kafka.{ProducerMessage, ConsumerProperties, ProducerProperties, ReactiveKafka}
import org.reactivestreams.{ Publisher, Subscriber }

implicit val actorSystem = ActorSystem("ReactiveKafka")
implicit val materializer = ActorMaterializer()

val kafka = new ReactiveKafka()
val publisher: Publisher[StringConsumerRecord] = kafka.consume(ConsumerProperties(
 bootstrapServers = "localhost:9092",
 topic = "lowercaseStrings",
 groupId = "groupName",
 valueDeserializer = new StringDeserializer()
))
val subscriber: Subscriber[StringProducerMessage] = kafka.publish(ProducerProperties(
  bootstrapServers = "localhost:9092",
  topic = "uppercaseStrings",
  valueSerializer = new StringSerializer()
))

Source.fromPublisher(publisher).map(m => ProducerMessage(m.value().toUpperCase))
  .to(Sink.fromSubscriber(subscriber)).run()

Java

import akka.actor.ActorSystem;
import akka.stream.ActorMaterializer;
import akka.stream.javadsl.Sink;
import akka.stream.javadsl.Source;
import org.apache.kafka.clients.consumer.ConsumerRecord;
import org.apache.kafka.common.serialization.StringDeserializer;
import org.apache.kafka.common.serialization.StringSerializer;
import org.reactivestreams.Publisher;
import org.reactivestreams.Subscriber;

public void run() {
String brokerList = "localhost:9092";

ReactiveKafka kafka = new ReactiveKafka();
ActorSystem system = ActorSystem.create("ReactiveKafka");
ActorMaterializer materializer = ActorMaterializer.create(system);

StringDeserializer deserializer = new StringDeserializer();
ConsumerProperties<String> cp =
   new PropertiesBuilder.Consumer(brokerList, "topic", "groupId", deserializer)
      .build();

Publisher<ConsumerRecord<String, String>> publisher = kafka.consume(cp, system);

StringSerializer serializer = new StringSerializer();
ProducerProperties<String, String> pp = new PropertiesBuilder.Producer(
   brokerList,
   "topic",
   serializer,
   serializer).build();

Subscriber<ProducerMessage<String, String>> subscriber = kafka.publish(pp, system);
Source.fromPublisher(publisher).map(this::toProdMessage)
  .to(Sink.fromSubscriber(subscriber)).run(materializer);
}

private ProducerMessage<String, String> toProdMessage(ConsumerRecord<String, String> record) {
  return KeyValueProducerMessage.apply(record.key(), record.value());
}

Passing configuration properties to Kafka

In order to set your own custom Kafka parameters, you can construct ConsumerProperties and ProducerProperties using some of their provided methods in a builder-pattern-style DSL, for example:

import org.apache.kafka.common.serialization.StringDeserializer
import com.softwaremill.react.kafka.ConsumerProperties

val consumerProperties = ConsumerProperties(
  "localhost:2181",
  "topic",
  "groupId",
  new StringDeserializer()
)
  .readFromEndOfStream()
  .consumerTimeoutMs(300)
  .commitInterval(2 seconds)
  .setProperty("some.kafka.property", "value")

The ProducerProperties class offers a similar API.

Controlling consumer start offset

By default a new consumer will start reading from the beginning of a topic, fetching all uncommitted messages. If you want to start reading from the end, you can specify this on your ConsumerProperties:

  val consumerProperties = ConsumerProperties(...).readFromEndOfStream()

Working with actors

Since we are based upon akka-stream, the best way to handle errors is to leverage Akka's error handling and lifecycle management capabilities. Producers and consumers are in fact actors.

Obtaining actor references

ReactiveKafka comes with a few methods allowing working on the actor level. You can let it create Props to let your own supervisor create these actors as children, or you can directly create actors at the top level of supervision. Here are some examples:

import akka.actor.{Props, ActorRef, Actor, ActorSystem}
import akka.stream.ActorMaterializer
import org.apache.kafka.common.serialization.{StringSerializer, StringDeserializer}
import com.softwaremill.react.kafka.{ReactiveKafka, ProducerProperties, ConsumerProperties}

// inside an Actor:
implicit val materializer = ActorMaterializer()

val kafka = new ReactiveKafka()
// consumer
val consumerProperties = ConsumerProperties(
  bootstrapServers = "localhost:9092",
  topic = "lowercaseStrings",
  groupId = "groupName",
  valueDeserializer = new StringDeserializer()
)
val consumerActorProps: Props = kafka.consumerActorProps(consumerProperties)
val publisherActor: ActorRef = context.actorOf(consumerActorProps)
// or:
val topLevelPublisherActor: ActorRef = kafka.consumerActor(consumerActorProps)

// subscriber
val producerProperties = ProducerProperties(
  bootstrapServers = "localhost:9092",
  topic = "uppercaseStrings",
  new StringSerializer()
)
val producerActorProps: Props = kafka.producerActorProps(producerProperties)
val subscriberActor: ActorRef = context.actorOf(producerActorProps)
// or:
val topLevelSubscriberActor: ActorRef = kafka.producerActor(producerProperties)

Handling errors

When a consumer or a producer fails to read/write from Kafka, the error is unrecoverable and requires that the connection be terminated. This will be performed automatically and the KafkaActorSubscriber / KafkaActorPublisher which failed will be stopped. You can use DeathWatch to detect such failures in order to restart your stream. Additionally, when a producer fails, it will signal onError() to stop the rest of stream.

Example of monitoring routine:

import akka.actor.{Actor, ActorRef, ActorSystem, Props}
import akka.stream.ActorMaterializer
import com.softwaremill.react.kafka.KafkaMessages._
import com.softwaremill.react.kafka.{ConsumerProperties, ProducerProperties, ReactiveKafka}

class Handler extends Actor {
  implicit val materializer = ActorMaterializer()

  def createSupervisedSubscriberActor() = {
    val kafka = new ReactiveKafka()

    // subscriber
    val subscriberProperties = ProducerProperties(
      bootstrapServers = "localhost:9092",
      topic = "uppercaseStrings",
      valueSerializer = new StringSerializer()
    )
    val subscriberActorProps: Props = kafka.producerActorProps(subscriberProperties)
    val subscriberActor = context.actorOf(subscriberActorProps)
    context.watch(subscriberActor)
  }

  override def receive: Receive = {
    case Terminated(actorRef) => // your custom handling
  }

  // Rest of the Actor's body
}

Cleaning up

If you want to manually stop a publisher or a subscriber, you have to send an appropriate message to the underlying actor. KafkaActorPublisher must receive a KafkaActorPublisher.Stop, whereas KafkaActorSubscriber must receive a ActorSubscriberMessage.OnComplete. If you're using a PublisherWithCommitSink returned from ReactiveKafka.consumeWithOffsetSink(), you must call its cancel() method in order to gracefully close all underlying resources.

Manual Commit (version 0.8 and above)

In order to be able to achieve "at-least-once" delivery, you can use following API to obtain an additional Sink, where you can stream back messages that you processed. An underlying actor will periodically flush offsets of these messages as committed. Example:

import scala.concurrent.duration._
import akka.actor.ActorSystem
import akka.stream.ActorMaterializer
import com.softwaremill.react.kafka.KafkaMessages._
import akka.stream.scaladsl.Source
import com.softwaremill.react.kafka.{ConsumerProperties, ReactiveKafka}

implicit val actorSystem = ActorSystem("ReactiveKafka")
implicit val materializer = ActorMaterializer()

val kafka = new ReactiveKafka()
val consumerProperties = ConsumerProperties(
  bootstrapServers = "localhost:9092",
  topic = "lowercaseStrings",
  groupId = "groupName",
  valueDeserializer = new StringDeserializer())
.commitInterval(5 seconds) // flush interval

val consumerWithOffsetSink = kafka.consumeWithOffsetSink(consumerProperties)
Source.fromPublisher(consumerWithOffsetSink.publisher)
  .map(processMessage(_)) // your message processing
  .to(consumerWithOffsetSink.offsetCommitSink) // stream back for commit
  .run()

Tuning

KafkaActorSubscriber and KafkaActorPublisher have their own thread pools, configured in reference.conf. You can tune them by overriding kafka-publisher-dispatcher.thread-pool-executor and kafka-subscriber-dispatcher.thread-pool-executor in your application.conf file.
Alternatively, you can provide your own dispatcher name. It can be passed to appropriate variants of factory methods in ReactiveKafka: publish(), producerActor(), producerActorProps() or consume(), consumerActor(), consumerActorProps().

Testing

Tests require Apache Kafka and Zookeeper to be available on localhost:9092 and localhost:2181. Note that auto.create.topics.enable should be true.

13h3r/reactive-kafka

Reactive Streams for Kafka

New API: 0.11-M4

Changes

0.11-M4

Example usage

Scala

Java

Configuration

Testing

Old API: 0.10.0

Example usage

Scala

Java

Passing configuration properties to Kafka

Controlling consumer start offset

Working with actors

Obtaining actor references

Handling errors

Cleaning up

Manual Commit (version 0.8 and above)

Tuning

Testing