Skip to content
This repository has been archived by the owner on Jul 14, 2023. It is now read-only.

Akka Reactive Streaming for Azure Event Hubs

License

Notifications You must be signed in to change notification settings

Azure/azure-event-hubs-reactive

Repository files navigation

Maven Central Build Issues

Reactive Event Hubs

Reactive Event Hubs is an Akka Stream library that can be used to read events from Azure Event Hubs, via a reactive stream with asynchronous back pressure. Azure Event Hubs is a service used to scale telemetry ingestion from websites, apps, and any streams of data.

The library can be used both in Java and Scala, providing a fluent DSL for both programming languages, similarly to the approach used by Akka.

The following is a simple example showing how to use the library in Scala. A stream of incoming telemetry data is read, parsed and converted to a Temperature object, and then filtered based on the temperature value:

EventHub().source()
    .to(console)
    .run()

and the equivalent code in Java:

TypeReference<Temperature> type = new TypeReference<Temperature>() {};

new EventHub().source()
    .to(console())
    .run(streamMaterializer);

Streaming from Event hubs to any

An interesting example is reading telemetry data from Azure Event Hubs, and sending it to a Kafka topic, so that it can be consumed by other services downstream:

...
import org.apache.kafka.common.serialization.StringSerializer
import org.apache.kafka.common.serialization.ByteArraySerializer
import org.apache.kafka.clients.producer.ProducerRecord
import akka.kafka.ProducerSettings
import akka.kafka.scaladsl.Producer

case class KafkaProducer(bootstrapServer: String)(implicit val system: ActorSystem) {

  protected val producerSettings = ProducerSettings(system, new ByteArraySerializer, new StringSerializer)
    .withBootstrapServers(bootstrapServer)

  def getSink() = Producer.plainSink(producerSettings)

  def packageMessage(elem: String, topic: String): ProducerRecord[Array[Byte], String] = {
    new ProducerRecord[Array[Byte], String](topic, elem)
  }
}
val kafkaProducer = KafkaProducer(bootstrapServer)

EventHub().source()
    .runWith(kafkaProducer.getSink())

Source options

Event hub partitions

The library supports reading from a subset of partitions, to enable the development of distributed applications. Consider for instance the scenario of a client application deployed to multiple nodes, where each node processes independently a subset of the incoming telemetry.

val p1 = 0
val p2 = 3

EventHub().source(Seq(p1, p2))
    .to(console)
    .run()

Starting point

Unless specified, the stream starts from the beginning of the data present in each partition. It's possible to start the stream from a given date and time too:

val start = java.time.Instant.now()

EventHub().source(start)
    .to(console)
    .run()

Multiple options

EventHub().source() provides a quick API to specify the start time or the partitions. To specify more options, you can use the SourceOptions class, combining multiple settings:

val options = SourceOptions()
  .partitions(0,2,3)
  .fromTime(java.time.Instant.now())
  .withRuntimeInfo()
  .saveOffsets()

EventHub().source(options)
    .to(console)
    .run()

Stream processing restart - saving the current position

The library provides a mechanism to restart the stream from a recent checkpoint, to be resilient to restarts and crashes. Checkpoints are saved automatically, with a configured frequency, on a storage provided. For instance, the stream position can be saved every 30 seconds and/or every 500 messages (these values are configurable), in a table in Cassandra or using Azure blobs.

Currently the position is saved in a concurrent thread, delayed by time and/or count, depending on the configuration settings. Given the current implementation it's possible that the position saved is ahead of your processing logic. While it's possible to mitigate the risk via the configuration settings, at-least-once cannot be guaranteed.

We are working to support at-least-once semantic in the next few months, providing more control on the checkpointing logic.

For more information about the checkpointing feature, please read here.

Build configuration

Reactive Event Hubs is available in Maven Central for Scala 2.12. To import the library into your project, add the following reference in your build.sbt file:

libraryDependencies += "com.microsoft.azure" %% "azure-eventhubs-reactive" % "0.5.0"

or this dependency in pom.xml file when working with Maven:

<dependency>
    <groupId>com.microsoft.azure</groupId>
    <artifactId>azure-eventhubs-reactive_2.12</artifactId>
    <version>0.5.0</version>
</dependency>

Reactive Event Hubs internally uses some libraries like Azure Storage SDK, Akka etc. If your project depends on these libraries too, your can override the versions, explicitly importing the packages in your build.sbt and pom.xml files. If you encounter some incompatibility with future versions of these, please let us know opening an issue, or sending a PR.

Event Hub configuration

By default Reactive Event Hubs uses an application.conf configuration file to fetch the parameters required to connect to Azure Event Hubs. The connection and authentication values to use, can be found in the Azure Portal:

Properties required to receive telemetry:

  • eventHubName:
  • eventHubEndpoint:
  • eventHubPartitions:
  • accessPolicy:
  • accessKey:

The values should be stored in your application.conf resource (or equivalent). Optionally you can reference environment settings if you prefer, for example to hide sensitive data.

reactive-eventhubs {

  connection {
    eventHubName        = "<Event Hub name>"
    eventHubEndpoint    = "<Event Hub endpoint>"
    eventHubPartitions  = <the number of partitions in your Event Hub>
    accessPolicy   = "<access policy name>"
    accessKey      = "<access policy key>"
  }

  [... other settings...]
}

Example using environment settings:

reactive-eventhubs {

  connection {
    eventHubName        = ${?EVENTHUB_NAME}
    eventHubEndpoint    = ${?EVENTHUB_ENDPOINT}
    eventHubPartitions  = ${?EVENTHUB_PARTITIONS}
    accessPolicy   = ${?EVENTHUB_ACCESS_POLICY}
    accessKey      = ${?EVENTHUB_ACCESS_KEY}
  }

  [... other settings...]
}

Note that the library will automatically use these exact environment variables, unless overridden in your configuration file (all the default settings are stored in reference.conf).

Although using a configuration file is the preferred approach, it's also possible to inject a different configuration at runtime, providing an object implementing the IConfiguration interface.

The logging level can be managed overriding Akka configuration, for example:

akka {
  # Options: OFF, ERROR, WARNING, INFO, DEBUG
  loglevel = "WARNING"
}

There are other settings, to tune performance and connection details:

  • streaming.consumerGroup: the consumer group used during the connection
  • streaming.receiverBatchSize: the number of messages retrieved on each call to Azure Event Hubs. The default (and maximum) value is 999.
  • streaming.receiverTimeout: timeout applied to calls while retrieving messages. The default value is 3 seconds.
  • streaming.retrieveRuntimeInfo: when enabled, the messages returned by EventHub.Source will contain some runtime information about the last message in each partition. You can use this information to calculate how many telemetry events remain to process.

The complete configuration reference (and default values) is available in reference.conf.

Samples

The project includes several demos in Java and Scala, showing some of the use cases and how the Reactive Event Hubs API works. All the demos require an instance of Azure Event Hubs, with some telemetry to stream.

  1. DisplayMessages [Java]: how to stream Azure Event Hubs telemetry within a Java application
  2. AllMessagesFromBeginning [Scala]: simple example streaming all the events in the hub.
  3. OnlyRecentMessages [Scala]: stream all the events, starting from the current time.
  4. OnlyTwoPartitions [Scala]: shows how to stream events from a subset of partitions.
  5. MultipleDestinations [Scala]: shows how to read once and deliver events to multiple destinations.
  6. CloseStream [Scala]: show how to close the stream
  7. Throughput [Scala]: stream all events and display statistics about the throughput.
  8. Throttling [Scala]: throttle the incoming stream to a defined speed of events/second.
  9. StoreOffsetsWhileStreaming [Scala]: demonstrates how the stream can be restarted without losing its position. The current position is stored in a Cassandra table (we suggest to run a docker container for the purpose of the demo, e.g. docker run -ip 9042:9042 --rm cassandra).
  10. StartFromStoredOffsetsButDontWriteNewOffsets [Scala]: shows how to use the saved checkpoints to start streaming from a known position, without changing the value in the storage. If the storage doesn't contain checkpoints, the stream starts from the beginning.
  11. StartFromStoredOffsetsIfAvailableOrByTimeOtherwise [Scala]: similar to the previous demo, with a fallback datetime when the storage doesn't contain checkpoints.
  12. StreamIncludingRuntimeInformation [Scala]: shows how runtime information works.

When ready, you should either edit the application.conf configuration files (scala and java) with your credentials, or set the corresponding environment variables. Follow the instructions described in the previous section on how to set the correct values.

The root folder includes also a script showing how to set the environment variables in Linux/MacOS and Windows.

The demos can be executed using the scripts included in the root folder (run_<language>_samples.sh and run_<language>_samples.cmd):

Running the tests

You can use the included build.sh script to execute all the unit and functional tests in the suite. The functional tests require an existing Azure Event Hub resource, that yous should setup. For the tests to connect to your Hub, configure your environment using the setup-env-vars.* scripts mentioned above in this page.