Event Bus
Loading

Event Bus

Originally conceived as a way to send messages to groups of actors, the EventBus has been generalized into a set of abstract base classes implementing a simple interface:

/**
 * Attempts to register the subscriber to the specified Classifier
 * @return true if successful and false if not (because it was already
 *   subscribed to that Classifier, or otherwise)
 */
public boolean subscribe(Subscriber subscriber, Classifier to);

/**
 * Attempts to deregister the subscriber from the specified Classifier
 * @return true if successful and false if not (because it wasn't subscribed
 *   to that Classifier, or otherwise)
 */
public boolean unsubscribe(Subscriber subscriber, Classifier from);

/**
 * Attempts to deregister the subscriber from all Classifiers it may be subscribed to
 */
public void unsubscribe(Subscriber subscriber);

/**
 * Publishes the specified Event to this bus
 */
public void publish(Event event);

Note

Please note that the EventBus does not preserve the sender of the published messages. If you need a reference to the original sender you have to provide it inside the message.

This mechanism is used in different places within Akka, e.g. the Event Stream. Implementations can make use of the specific building blocks presented below.

An event bus must define the following three type parameters:

  • Event (E) is the type of all events published on that bus
  • Subscriber (S) is the type of subscribers allowed to register on that event bus
  • Classifier (C) defines the classifier to be used in selecting subscribers for dispatching events

The traits below are still generic in these types, but they need to be defined for any concrete implementation.

Classifiers

The classifiers presented here are part of the Akka distribution, but rolling your own in case you do not find a perfect match is not difficult, check the implementation of the existing ones on github

Lookup Classification

The simplest classification is just to extract an arbitrary classifier from each event and maintaining a set of subscribers for each possible classifier. This can be compared to tuning in on a radio station. The trait LookupClassification is still generic in that it abstracts over how to compare subscribers and how exactly to classify.

The necessary methods to be implemented are illustrated with the following example:

import akka.event.japi.LookupEventBus;
import java.util.concurrent.TimeUnit;

  public class MsgEnvelope {
    public final String topic;
    public final Object payload;

    public MsgEnvelope(String topic, Object payload) {
      this.topic = topic;
      this.payload = payload;
    }
  }
  
  /**
   * Publishes the payload of the MsgEnvelope when the topic of the
   * MsgEnvelope equals the String specified when subscribing.
   */
  public class LookupBusImpl extends LookupEventBus<MsgEnvelope, ActorRef, String> {

    // is used for extracting the classifier from the incoming events
    @Override public String classify(MsgEnvelope event) {
      return event.topic;
    }

    // will be invoked for each event for all subscribers which registered themselves
    // for the event’s classifier
    @Override public void publish(MsgEnvelope event, ActorRef subscriber) {
      subscriber.tell(event.payload, ActorRef.noSender());
    }

    // must define a full order over the subscribers, expressed as expected from
    // `java.lang.Comparable.compare`
    @Override public int compareSubscribers(ActorRef a, ActorRef b) {
      return a.compareTo(b);
    }

    // determines the initial size of the index data structure
    // used internally (i.e. the expected number of different classifiers)
    @Override public int mapSize() {
      return 128;
    }
    
  }

A test for this implementation may look like this:

LookupBusImpl lookupBus = new LookupBusImpl();
lookupBus.subscribe(getTestActor(), "greetings");
lookupBus.publish(new MsgEnvelope("time", System.currentTimeMillis()));
lookupBus.publish(new MsgEnvelope("greetings", "hello"));
expectMsgEquals("hello");

This classifier is efficient in case no subscribers exist for a particular event.

Subchannel Classification

If classifiers form a hierarchy and it is desired that subscription be possible not only at the leaf nodes, this classification may be just the right one. It can be compared to tuning in on (possibly multiple) radio channels by genre. This classification has been developed for the case where the classifier is just the JVM class of the event and subscribers may be interested in subscribing to all subclasses of a certain class, but it may be used with any classifier hierarchy.

The necessary methods to be implemented are illustrated with the following example:

import akka.event.japi.SubchannelEventBus;
import akka.util.Subclassification;

  public class StartsWithSubclassification implements Subclassification<String> {
    @Override public boolean isEqual(String x, String y) {
      return x.equals(y);
    }

    @Override public boolean isSubclass(String x, String y) {
      return x.startsWith(y);
    }
  }
  
  /**
   * Publishes the payload of the MsgEnvelope when the topic of the
   * MsgEnvelope starts with the String specified when subscribing.
   */
  public class SubchannelBusImpl extends SubchannelEventBus<MsgEnvelope, ActorRef, String> {

    // Subclassification is an object providing `isEqual` and `isSubclass`
    // to be consumed by the other methods of this classifier
    @Override public Subclassification<String> subclassification() {
      return new StartsWithSubclassification();
    }
    
    // is used for extracting the classifier from the incoming events
    @Override public String classify(MsgEnvelope event) {
      return event.topic;
    }

    // will be invoked for each event for all subscribers which registered themselves
    // for the event’s classifier
    @Override public void publish(MsgEnvelope event, ActorRef subscriber) {
      subscriber.tell(event.payload, ActorRef.noSender());
    }

  }

A test for this implementation may look like this:

SubchannelBusImpl subchannelBus = new SubchannelBusImpl();
subchannelBus.subscribe(getTestActor(), "abc");
subchannelBus.publish(new MsgEnvelope("xyzabc", "x"));
subchannelBus.publish(new MsgEnvelope("bcdef", "b"));
subchannelBus.publish(new MsgEnvelope("abc", "c"));
expectMsgEquals("c");
subchannelBus.publish(new MsgEnvelope("abcdef", "d"));
expectMsgEquals("d");

This classifier is also efficient in case no subscribers are found for an event, but it uses conventional locking to synchronize an internal classifier cache, hence it is not well-suited to use cases in which subscriptions change with very high frequency (keep in mind that “opening” a classifier by sending the first message will also have to re-check all previous subscriptions).

Scanning Classification

The previous classifier was built for multi-classifier subscriptions which are strictly hierarchical, this classifier is useful if there are overlapping classifiers which cover various parts of the event space without forming a hierarchy. It can be compared to tuning in on (possibly multiple) radio stations by geographical reachability (for old-school radio-wave transmission).

The necessary methods to be implemented are illustrated with the following example:

import akka.event.japi.ScanningEventBus;

  /**
   * Publishes String messages with length less than or equal to the length
   * specified when subscribing.
   */
  public class ScanningBusImpl extends ScanningEventBus<String, ActorRef, Integer> {

    // is needed for determining matching classifiers and storing them in an
    // ordered collection
    @Override public int compareClassifiers(Integer a, Integer b) {
      return a.compareTo(b);
    }

    // is needed for storing subscribers in an ordered collection  
    @Override public int compareSubscribers(ActorRef a, ActorRef b) {
      return a.compareTo(b);
    }

    // determines whether a given classifier shall match a given event; it is invoked
    // for each subscription for all received events, hence the name of the classifier
    @Override public boolean matches(Integer classifier, String event) {
      return event.length() <= classifier;
    }

    // will be invoked for each event for all subscribers which registered themselves
    // for the event’s classifier
    @Override public void publish(String event, ActorRef subscriber) {
      subscriber.tell(event, ActorRef.noSender());
    }

  }

A test for this implementation may look like this:

ScanningBusImpl scanningBus = new ScanningBusImpl();
scanningBus.subscribe(getTestActor(), 3);
scanningBus.publish("xyzabc");
scanningBus.publish("ab");
expectMsgEquals("ab");
scanningBus.publish("abc");
expectMsgEquals("abc");

This classifier takes always a time which is proportional to the number of subscriptions, independent of how many actually match.

Actor Classification

This classification was originally developed specifically for implementing DeathWatch: subscribers as well as classifiers are of type ActorRef.

This classification requires an ActorSystem in order to perform book-keeping operations related to the subscribers being Actors, which can terminate without first unsubscribing from the EventBus. ActorClassification maitains a system Actor which takes care of unsubscribing terminated actors automatically.

The necessary methods to be implemented are illustrated with the following example:

import akka.event.japi.ActorEventBus;

  public class Notification {
    public final ActorRef ref;
    public final int id;

    public Notification(ActorRef ref, int id) {
      this.ref = ref;
      this.id = id;
    }
  }
  
  public class ActorBusImpl extends ActorEventBus<Notification> {

    // the ActorSystem will be used for book-keeping operations, such as subscribers terminating
    public ActorBusImpl(ActorSystem system) {
      super(system);
    }

    // is used for extracting the classifier from the incoming events
    @Override public ActorRef classify(Notification event) {
      return event.ref;
    }
    
    // determines the initial size of the index data structure
    // used internally (i.e. the expected number of different classifiers)
    @Override public int mapSize() {
      return 128;
    }

  }

A test for this implementation may look like this:

ActorRef observer1 = new JavaTestKit(system).getRef();
ActorRef observer2 = new JavaTestKit(system).getRef();
JavaTestKit probe1 = new JavaTestKit(system);
JavaTestKit probe2 = new JavaTestKit(system);
ActorRef subscriber1 = probe1.getRef();
ActorRef subscriber2 = probe2.getRef();
ActorBusImpl actorBus = new ActorBusImpl(system);
actorBus.subscribe(subscriber1, observer1);
actorBus.subscribe(subscriber2, observer1);
actorBus.subscribe(subscriber2, observer2);
Notification n1 = new Notification(observer1, 100);
actorBus.publish(n1);
probe1.expectMsgEquals(n1);
probe2.expectMsgEquals(n1);
Notification n2 = new Notification(observer2, 101);
actorBus.publish(n2);
probe2.expectMsgEquals(n2);
probe1.expectNoMsg(FiniteDuration.create(500, TimeUnit.MILLISECONDS));

This classifier is still is generic in the event type, and it is efficient for all use cases.

Event Stream

The event stream is the main event bus of each actor system: it is used for carrying log messages and Dead Letters and may be used by the user code for other purposes as well. It uses Subchannel Classification which enables registering to related sets of channels (as is used for RemotingLifecycleEvent). The following example demonstrates how a simple subscription works. Given a simple actor:

import akka.actor.Props;
import akka.actor.ActorRef;
import akka.actor.ActorSystem;
import akka.actor.UntypedActor;
import akka.actor.DeadLetter;
public class DeadLetterActor extends UntypedActor {
  public void onReceive(Object message) {
    if (message instanceof DeadLetter) {
      System.out.println(message);
    }
  }
}

it can be subscribed like this:

final ActorSystem system = ActorSystem.create("DeadLetters");
final ActorRef actor = system.actorOf(Props.create(DeadLetterActor.class));
system.eventStream().subscribe(actor, DeadLetter.class);

Similarily to Actor Classification, EventStream will automatically remove subscibers when they terminate.

Note

The event stream is a local facility, meaning that it will not distribute events to other nodes in a clustered environment (unless you subscribe a Remote Actor to the stream explicitly). If you need to broadcast events in an Akka cluster, without knowing your recipients explicitly (i.e. obtaining their ActorRefs), you may want to look into: distributed-pub-sub.

Default Handlers

Upon start-up the actor system creates and subscribes actors to the event stream for logging: these are the handlers which are configured for example in application.conf:

akka {
  loggers = ["akka.event.Logging$DefaultLogger"]
}

The handlers listed here by fully-qualified class name will be subscribed to all log event classes with priority higher than or equal to the configured log-level and their subscriptions are kept in sync when changing the log-level at runtime:

system.eventStream.setLogLevel(Logging.DebugLevel());

This means that log events for a level which will not be logged are typically not dispatched at all (unless manual subscriptions to the respective event class have been done)

Dead Letters

As described at Stopping actors, messages queued when an actor terminates or sent after its death are re-routed to the dead letter mailbox, which by default will publish the messages wrapped in DeadLetter. This wrapper holds the original sender, receiver and message of the envelope which was redirected.

Other Uses

The event stream is always there and ready to be used, just publish your own events (it accepts Object) and subscribe listeners to the corresponding JVM classes.

Contents