Persistence

Identifiers

A persistent actor must have an identifier that doesn't change across different actor incarnations. The identifier must be defined with the persistenceId method.

@Override
public String persistenceId() {
    return "my-stable-persistence-id";
}

Recovery

By default, a persistent actor is automatically recovered on start and on restart by replaying journaled messages. New messages sent to a persistent actor during recovery do not interfere with replayed messages. They are stashed and received by a persistent actor after recovery phase completes.

The number of concurrent recoveries of recoveries that can be in progress at the same time is limited to not overload the system and the backend data store. When exceeding the limit the actors will wait until other recoveries have been completed. This is configured by:

akka.persistence.max-concurrent-recoveries = 50

@Override
public Recovery recovery() {
    return Recovery.create(SnapshotSelectionCriteria.none());
}

@Override
public Recovery recovery() {
    return Recovery.create(457L);
}

@Override
public Recovery recovery() {
    return Recovery.none();
}

public boolean recoveryRunning();
public boolean recoveryFinished();

@Override
public void onReceiveRecover(Object message) {
  if (message instanceof RecoveryCompleted) {
    // perform init after recovery, before any other messages
  }
}
    
@Override
public void onReceiveCommand(Object message) throws Exception {
  if (message instanceof String) {
    // ...
  } else {
    unhandled(message);
  }
}

Internal stash

The persistent actor has a private stash for internally caching incoming messages during recovery or the persist\persistAll method persisting events. You can still use/inherit from the Stash interface. The internal stash cooperates with the normal stash by hooking into unstashAll method and making sure messages are unstashed properly to the internal stash to maintain ordering guarantees.

You should be careful to not send more messages to a persistent actor than it can keep up with, otherwise the number of stashed messages will grow without bounds. It can be wise to protect against OutOfMemoryError by defining a maximum stash capacity in the mailbox configuration:

akka.actor.default-mailbox.stash-capacity=10000

Note that the stash capacity is per actor. If you have many persistent actors, e.g. when using cluster sharding, you may need to define a small stash capacity to ensure that the total number of stashed messages in the system don't consume too much memory. Additionally, The persistent actor defines three strategies to handle failure when the internal stash capacity is exceeded. The default overflow strategy is the ThrowOverflowExceptionStrategy, which discards the current received message and throws a StashOverflowException, causing actor restart if default supervision strategy is used. you can override the internalStashOverflowStrategy method to return DiscardToDeadLetterStrategy or ReplyToStrategy for any "individual" persistent actor, or define the "default" for all persistent actors by providing FQCN, which must be a subclass of StashOverflowStrategyConfigurator, in the persistence configuration:

akka.persistence.internal-stash-overflow-strategy=
  "akka.persistence.ThrowExceptionConfigurator"

The DiscardToDeadLetterStrategy strategy also has a pre-packaged companion configurator akka.persistence.DiscardConfigurator.

You can also query default strategy via the Akka persistence extension singleton:

Persistence.get(context().system()).defaultInternalStashOverflowStrategy();

Relaxed local consistency requirements and high throughput use-cases

If faced with relaxed local consistency requirements and high throughput demands sometimes PersistentActor and its persist may not be enough in terms of consuming incoming Commands at a high rate, because it has to wait until all Events related to a given Command are processed in order to start processing the next Command. While this abstraction is very useful for most cases, sometimes you may be faced with relaxed requirements about consistency – for example you may want to process commands as fast as you can, assuming that the Event will eventually be persisted and handled properly in the background, retroactively reacting to persistence failures if needed.

The persistAsync method provides a tool for implementing high-throughput persistent actors. It will not stash incoming Commands while the Journal is still working on persisting and/or user code is executing event callbacks.

In the below example, the event callbacks may be called "at any time", even after the next Command has been processed. The ordering between events is still guaranteed ("evt-b-1" will be sent after "evt-a-2", which will be sent after "evt-a-1" etc.).

class MyPersistentActor extends UntypedPersistentActor {
    @Override
    public String persistenceId() { return "some-persistence-id"; }

    @Override
    public void onReceiveRecover(Object msg) {
        // handle recovery here
    }

    @Override
    public void onReceiveCommand(Object msg) {
        sender().tell(msg, self());

        persistAsync(String.format("evt-%s-1", msg), new Procedure<String>(){
            @Override
            public void apply(String event) throws Exception {
                sender().tell(event, self());
            }
        });
        persistAsync(String.format("evt-%s-2", msg), new Procedure<String>(){
            @Override
            public void apply(String event) throws Exception {
                sender().tell(event, self());
            }
        });
    }
}

Deferring actions until preceding persist handlers have executed

Sometimes when working with persistAsync you may find that it would be nice to define some actions in terms of ''happens-after the previous persistAsync handlers have been invoked''. PersistentActor provides an utility method called deferAsync, which works similarly to persistAsync yet does not persist the passed in event. It is recommended to use it for read operations, and actions which do not have corresponding events in your domain model.

Using this method is very similar to the persist family of methods, yet it does not persist the passed in event. It will be kept in memory and used when invoking the handler.

class MyPersistentActor extends UntypedPersistentActor {
    @Override
    public String persistenceId() { return "some-persistence-id"; }

    @Override
    public void onReceiveRecover(Object msg) {
        // handle recovery here
    }

    @Override
    public void onReceiveCommand(Object msg) {
        final Procedure<String> replyToSender = new Procedure<String>() {
            @Override
            public void apply(String event) throws Exception {
                sender().tell(event, self());
            }
        };

        persistAsync(String.format("evt-%s-1", msg), replyToSender);
        persistAsync(String.format("evt-%s-2", msg), replyToSender);
        deferAsync(String.format("evt-%s-3", msg), replyToSender);
    }
}

Notice that the sender() is safe to access in the handler callback, and will be pointing to the original sender of the command for which this deferAsync handler was called.

final ActorRef persistentActor = system.actorOf(Props.create(MyPersistentActor.class));
persistentActor.tell("a", null);
persistentActor.tell("b", null);

// order of received messages:
// a
// b
// evt-a-1
// evt-a-2
// evt-a-3
// evt-b-1
// evt-b-2
// evt-b-3

Nested persist calls

It is possible to call persist and persistAsync inside their respective callback blocks and they will properly retain both the thread safety (including the right value of sender()) as well as stashing guarantees.

In general it is encouraged to create command handlers which do not need to resort to nested event persisting, however there are situations where it may be useful. It is important to understand the ordering of callback execution in those situations, as well as their implication on the stashing behaviour (that persist() enforces). In the following example two persist calls are issued, and each of them issues another persist inside its callback:

@Override
public void onReceiveCommand(Object msg) {
    final Procedure<String> replyToSender = new Procedure<String>() {
        @Override
        public void apply(String event) throws Exception {
            sender().tell(event, self());
        }
    };

    final Procedure<String> outer1Callback = new Procedure<String>() {
        @Override
        public void apply(String event) throws Exception {
            sender().tell(event, self());
            persist(String.format("%s-inner-1", msg), replyToSender);
        }
    };
    final Procedure<String> outer2Callback = new Procedure<String>() {
        @Override
        public void apply(String event) throws Exception {
            sender().tell(event, self());
            persist(String.format("%s-inner-2", msg), replyToSender);
        }
    };

    persist(String.format("%s-outer-1", msg), outer1Callback);
    persist(String.format("%s-outer-2", msg), outer2Callback);
}

When sending two commands to this PersistentActor, the persist handlers will be executed in the following order:

persistentActor.tell("a", self());
persistentActor.tell("b", self());

// order of received messages:
// a
// a-outer-1
// a-outer-2
// a-inner-1
// a-inner-2
// and only then process "b"
// b
// b-outer-1
// b-outer-2
// b-inner-1
// b-inner-2

First the "outer layer" of persist calls is issued and their callbacks are applied. After these have successfully completed, the inner callbacks will be invoked (once the events they are persisting have been confirmed to be persisted by the journal). Only after all these handlers have been successfully invoked will the next command be delivered to the persistent Actor. In other words, the stashing of incoming commands that is guaranteed by initially calling persist() on the outer layer is extended until all nested persist callbacks have been handled.

It is also possible to nest persistAsync calls, using the same pattern:

@Override
public void onReceiveCommand(Object msg) {
    final Procedure<String> replyToSender = new Procedure<String>() {
        @Override
        public void apply(String event) throws Exception {
            sender().tell(event, self());
        }
    };

    final Procedure<String> outer1Callback = new Procedure<String>() {
        @Override
        public void apply(String event) throws Exception {
            sender().tell(event, self());
            persistAsync(String.format("%s-inner-1", msg), replyToSender);
        }
    };
    final Procedure<String> outer2Callback = new Procedure<String>() {
        @Override
        public void apply(String event) throws Exception {
            sender().tell(event, self());
            persistAsync(String.format("%s-inner-1", msg), replyToSender);
        }
    };

    persistAsync(String.format("%s-outer-1", msg), outer1Callback);
    persistAsync(String.format("%s-outer-2", msg), outer2Callback);
}

In this case no stashing is happening, yet events are still persisted and callbacks are executed in the expected order:

persistentActor.tell("a", ActorRef.noSender());
persistentActor.tell("b", ActorRef.noSender());

// order of received messages:
// a
// b
// a-outer-1
// a-outer-2
// b-outer-1
// b-outer-2
// a-inner-1
// a-inner-2
// b-inner-1
// b-inner-2

// which can be seen as the following causal relationship:
// a -> a-outer-1 -> a-outer-2 -> a-inner-1 -> a-inner-2
// b -> b-outer-1 -> b-outer-2 -> b-inner-1 -> b-inner-2

While it is possible to nest mixed persist and persistAsync with keeping their respective semantics it is not a recommended practice, as it may lead to overly complex nesting.

Failures

If persistence of an event fails, onPersistFailure will be invoked (logging the error by default), and the actor will unconditionally be stopped.

The reason that it cannot resume when persist fails is that it is unknown if the event was actually persisted or not, and therefore it is in an inconsistent state. Restarting on persistent failures will most likely fail anyway since the journal is probably unavailable. It is better to stop the actor and after a back-off timeout start it again. The akka.pattern.BackoffSupervisor actor is provided to support such restarts.

@Override
public void preStart() throws Exception {
  final Props childProps = Props.create(MyPersistentActor1.class);
  final Props props = BackoffSupervisor.props(
    childProps,
    "myActor",
    Duration.create(3, TimeUnit.SECONDS),
    Duration.create(30, TimeUnit.SECONDS),
    0.2);
  getContext().actorOf(props, "mySupervisor");
  super.preStart();
}

If persistence of an event is rejected before it is stored, e.g. due to serialization error, onPersistRejected will be invoked (logging a warning by default), and the actor continues with next message.

If there is a problem with recovering the state of the actor from the journal when the actor is started, onRecoveryFailure is called (logging the error by default), and the actor will be stopped. Note that failure to load snapshot is also treated like this, but you can disable loading of snapshots if you for example know that serialization format has changed in an incompatible way, see Recovery customization.

Atomic writes

Each event is of course stored atomically, but it is also possible to store several events atomically by using the persistAll or persistAllAsync method. That means that all events passed to that method are stored or none of them are stored if there is an error.

The recovery of a persistent actor will therefore never be done partially with only a subset of events persisted by persistAll.

Some journals may not support atomic writes of several events and they will then reject the persistAll command, i.e. onPersistRejected is called with an exception (typically UnsupportedOperationException).

Batch writes

In order to optimize throughput when using persistAsync, a persistent actor internally batches events to be stored under high load before writing them to the journal (as a single batch). The batch size is dynamically determined by how many events are emitted during the time of a journal round-trip: after sending a batch to the journal no further batch can be sent before confirmation has been received that the previous batch has been written. Batch writes are never timer-based which keeps latencies at a minimum.

Message deletion

It is possible to delete all messages (journaled by a single persistent actor) up to a specified sequence number; Persistent actors may call the deleteMessages method to this end.

Deleting messages in event sourcing based applications is typically either not used at all, or used in conjunction with snapshotting, i.e. after a snapshot has been successfully stored, a deleteMessages(toSequenceNr) up until the sequence number of the data held by that snapshot can be issued to safely delete the previous events while still having access to the accumulated state during replays - by loading the snapshot.

The result of the deleteMessages request is signaled to the persistent actor with a DeleteMessagesSuccess message if the delete was successful or a DeleteMessagesFailure message if it failed.

Message deletion doesn't affect the highest sequence number of the journal, even if all messages were deleted from it after deleteMessages invocation.

Persistence status handling

Persisting, deleting, and replaying messages can either succeed or fail.

The most important operations (persist and recovery) have failure handlers modelled as explicit callbacks which the user can override in the PersistentActor. The default implementations of these handlers emit a log message (error for persist/recovery failures, and warning for others), logging the failure cause and information about which message caused the failure.

For critical failures such as recovery or persisting events failing the persistent actor will be stopped after the failure handler is invoked. This is because if the underlying journal implementation is signalling persistence failures it is most likely either failing completely or overloaded and restarting right-away and trying to persist the event again will most likely not help the journal recover – as it would likely cause a Thundering herd problem, as many persistent actors would restart and try to persist their events again. Instead, using a BackoffSupervisor (as described in Failures) which implements an exponential-backoff strategy which allows for more breathing room for the journal to recover between restarts of the persistent actor.

Safely shutting down persistent actors

Special care should be given when shutting down persistent actors from the outside. With normal Actors it is often acceptable to use the special PoisonPill message to signal to an Actor that it should stop itself once it receives this message – in fact this message is handled automatically by Akka, leaving the target actor no way to refuse stopping itself when given a poison pill.

This can be dangerous when used with PersistentActor due to the fact that incoming commands are stashed while the persistent actor is awaiting confirmation from the Journal that events have been written when persist() was used. Since the incoming commands will be drained from the Actor's mailbox and put into its internal stash while awaiting the confirmation (thus, before calling the persist handlers) the Actor may receive and (auto)handle the PoisonPill before it processes the other messages which have been put into its stash, causing a pre-mature shutdown of the Actor.

The example below highlights how messages arrive in the Actor's mailbox and how they interact with its internal stashing mechanism when persist() is used. Notice the early stop behaviour that occurs when PoisonPill is used:

final class Shutdown {}

class MyPersistentActor extends UntypedPersistentActor {
    @Override
    public String persistenceId() {
        return "some-persistence-id";
    }

    @Override
    public void onReceiveCommand(Object msg) throws Exception {
        if (msg instanceof Shutdown) {
            context().stop(self());
        } else if (msg instanceof String) {
            System.out.println(msg);
            persist("handle-" + msg, new Procedure<String>() {
                @Override
                public void apply(String param) throws Exception {
                    System.out.println(param);
                }
            });
        } else unhandled(msg);
    }

    @Override
    public void onReceiveRecover(Object msg) throws Exception {
        // handle recovery...
    }
}

// UN-SAFE, due to PersistentActor's command stashing:
persistentActor.tell("a", ActorRef.noSender());
persistentActor.tell("b", ActorRef.noSender());
persistentActor.tell(PoisonPill.getInstance(), ActorRef.noSender());
// order of received messages:
// a
//   # b arrives at mailbox, stashing;        internal-stash = [b]
//   # PoisonPill arrives at mailbox, stashing; internal-stash = [b, Shutdown]
// PoisonPill is an AutoReceivedMessage, is handled automatically
// !! stop !!
// Actor is stopped without handling `b` nor the `a` handler!

// SAFE:
persistentActor.tell("a", ActorRef.noSender());
persistentActor.tell("b", ActorRef.noSender());
persistentActor.tell(new Shutdown(), ActorRef.noSender());
// order of received messages:
// a
//   # b arrives at mailbox, stashing;        internal-stash = [b]
//   # Shutdown arrives at mailbox, stashing; internal-stash = [b, Shutdown]
// handle-a
//   # unstashing;                            internal-stash = [Shutdown]
// b
// handle-b
//   # unstashing;                            internal-stash = []
// Shutdown
// -- stop --

Replay Filter

There could be cases where event streams are corrupted and multiple writers (i.e. multiple persistent actor instances) journaled different messages with the same sequence number. In such a case, you can configure how you filter replayed messages from multiple writers, upon recovery.

In your configuration, under the akka.persistence.journal.xxx.replay-filter section (where xxx is your journal plugin id), you can select the replay filter mode from one of the following values:

• repair-by-discard-old
• fail
• warn
• off

For example, if you configure the replay filter for leveldb plugin, it looks like this:

# The replay filter can detect a corrupt event stream by inspecting
# sequence numbers and writerUuid when replaying events.
akka.persistence.journal.leveldb.replay-filter {
  # What the filter should do when detecting invalid events.
  # Supported values:
  # `repair-by-discard-old` : discard events from old writers,
  #                           warning is logged
  # `fail` : fail the replay, error is logged
  # `warn` : log warning but emit events untouched
  # `off` : disable this feature completely
  mode = repair-by-discard-old
}

Updates

The default update interval of all persistent views of an actor system is configurable:

akka.persistence.view.auto-update-interval = 5s

UntypedPersistentView implementation classes may also override the autoUpdateInterval method to return a custom update interval for a specific view class or view instance. Applications may also trigger additional updates at any time by sending a view an Update message.

final ActorRef view = system.actorOf(Props.create(MyView.class));
view.tell(Update.create(true), null);

If the await parameter is set to true, messages that follow the Update request are processed when the incremental message replay, triggered by that update request, completed. If set to false (default), messages following the update request may interleave with the replayed message stream. Automated updates always run with await = false.

Automated updates of all persistent views of an actor system can be turned off by configuration:

akka.persistence.view.auto-update = off

Implementation classes may override the configured default value by overriding the autoUpdate method. To limit the number of replayed messages per update request, applications can configure a custom akka.persistence.view.auto-update-replay-max value or override the autoUpdateReplayMax method. The number of replayed messages for manual updates can be limited with the replayMax parameter of the Update message.

Recovery

Initial recovery of persistent views works the very same way as for persistent actors (i.e. by sending a Recover message to self). The maximum number of replayed messages during initial recovery is determined by autoUpdateReplayMax. Further possibilities to customize initial recovery are explained in section Recovery.

Identifiers

A persistent view must have an identifier that doesn't change across different actor incarnations. The identifier must be defined with the viewId method.

The viewId must differ from the referenced persistenceId, unless Snapshots of a view and its persistent actor should be shared (which is what applications usually do not want).

Snapshot deletion

A persistent actor can delete individual snapshots by calling the deleteSnapshot method with the sequence number of when the snapshot was taken.

To bulk-delete a range of snapshots matching SnapshotSelectionCriteria, persistent actors should use the deleteSnapshots method.

Snapshot status handling

Saving or deleting snapshots can either succeed or fail – this information is reported back to the persistent actor via status messages as illustrated in the following table.

Relationship between deliver and confirmDelivery

To send messages to the destination path, use the deliver method after you have persisted the intent to send the message.

The destination actor must send back a confirmation message. When the sending actor receives this confirmation message you should persist the fact that the message was delivered successfully and then call the confirmDelivery method.

If the persistent actor is not currently recovering, the deliver method will send the message to the destination actor. When recovering, messages will be buffered until they have been confirmed using confirmDelivery. Once recovery has completed, if there are outstanding messages that have not been confirmed (during the message replay), the persistent actor will resend these before sending any other messages.

Deliver requires a deliveryIdToMessage function to pass the provided deliveryId into the message so that the correlation between deliver and confirmDelivery is possible. The deliveryId must do the round trip. Upon receipt of the message, the destination actor will send the same``deliveryId`` wrapped in a confirmation message back to the sender. The sender will then use it to call the confirmDelivery method to complete the delivery routine.

class Msg implements Serializable {
  public final long deliveryId;
  public final String s;
  
  public Msg(long deliveryId, String s) {
    this.deliveryId = deliveryId;
    this.s = s;
  }
}

class Confirm implements Serializable {
  public final long deliveryId;
  
  public Confirm(long deliveryId) {
    this.deliveryId = deliveryId;
  }
}

  
class MsgSent implements Serializable {
  public final String s;
  
  public MsgSent(String s) {
    this.s = s;
  }
}
class MsgConfirmed implements Serializable {
  public final long deliveryId;
  
  public MsgConfirmed(long deliveryId) {
    this.deliveryId = deliveryId;
  }
}

class MyPersistentActor extends UntypedPersistentActorWithAtLeastOnceDelivery {
  private final ActorSelection destination;
  
  @Override
  public String persistenceId() { return "persistence-id"; }

  public MyPersistentActor(ActorSelection destination) {
      this.destination = destination;
  }

  @Override
  public void onReceiveCommand(Object message) {
    if (message instanceof String) {
      String s = (String) message;
      persist(new MsgSent(s), new Procedure<MsgSent>() {
        public void apply(MsgSent evt) {
          updateState(evt);
        }
      });
    } else if (message instanceof Confirm) {
      Confirm confirm = (Confirm) message;
      persist(new MsgConfirmed(confirm.deliveryId), new Procedure<MsgConfirmed>() {
        public void apply(MsgConfirmed evt) {
          updateState(evt);
        }
      });
    } else {
      unhandled(message);
    }
  }
  
  @Override
  public void onReceiveRecover(Object event) {
    updateState(event);
  }
  
  void updateState(Object event) {
    if (event instanceof MsgSent) {
      final MsgSent evt = (MsgSent) event;
      deliver(destination, new Function<Long, Object>() {
        public Object apply(Long deliveryId) {
          return new Msg(deliveryId, evt.s);
        }
      });
    } else if (event instanceof MsgConfirmed) {
      final MsgConfirmed evt = (MsgConfirmed) event;
      confirmDelivery(evt.deliveryId);
    }
  }
}

class MyDestination extends UntypedActor {
  public void onReceive(Object message) throws Exception {
    if (message instanceof Msg) {
      Msg msg = (Msg) message;
      // ...
      getSender().tell(new Confirm(msg.deliveryId), getSelf());
    } else {
      unhandled(message);
    }
  }
}

The deliveryId generated by the persistence module is a strictly monotonically increasing sequence number without gaps. The same sequence is used for all destinations of the actor, i.e. when sending to multiple destinations the destinations will see gaps in the sequence. It is not possible to use custom deliveryId. However, you can send a custom correlation identifier in the message to the destination. You must then retain a mapping between the internal deliveryId (passed into the deliveryIdToMessage function) and your custom correlation id (passed into the message). You can do this by storing such mapping in a Map(correlationId -> deliveryId) from which you can retrieve the deliveryId to be passed into the confirmDelivery method once the receiver of your message has replied with your custom correlation id.

The UntypedPersistentActorWithAtLeastOnceDelivery class has a state consisting of unconfirmed messages and a sequence number. It does not store this state itself. You must persist events corresponding to the deliver and confirmDelivery invocations from your PersistentActor so that the state can be restored by calling the same methods during the recovery phase of the PersistentActor. Sometimes these events can be derived from other business level events, and sometimes you must create separate events. During recovery, calls to deliver will not send out messages, those will be sent later if no matching confirmDelivery will have been performed.

Support for snapshots is provided by getDeliverySnapshot and setDeliverySnapshot. The AtLeastOnceDeliverySnapshot contains the full delivery state, including unconfirmed messages. If you need a custom snapshot for other parts of the actor state you must also include the AtLeastOnceDeliverySnapshot. It is serialized using protobuf with the ordinary Akka serialization mechanism. It is easiest to include the bytes of the AtLeastOnceDeliverySnapshot as a blob in your custom snapshot.

The interval between redelivery attempts is defined by the redeliverInterval method. The default value can be configured with the akka.persistence.at-least-once-delivery.redeliver-interval configuration key. The method can be overridden by implementation classes to return non-default values.

The maximum number of messages that will be sent at each redelivery burst is defined by the redeliveryBurstLimit method (burst frequency is half of the redelivery interval). If there's a lot of unconfirmed messages (e.g. if the destination is not available for a long time), this helps to prevent an overwhelming amount of messages to be sent at once. The default value can be configured with the akka.persistence.at-least-once-delivery.redelivery-burst-limit configuration key. The method can be overridden by implementation classes to return non-default values.

After a number of delivery attempts a AtLeastOnceDelivery.UnconfirmedWarning message will be sent to self. The re-sending will still continue, but you can choose to call confirmDelivery to cancel the re-sending. The number of delivery attempts before emitting the warning is defined by the warnAfterNumberOfUnconfirmedAttempts method. The default value can be configured with the akka.persistence.at-least-once-delivery.warn-after-number-of-unconfirmed-attempts configuration key. The method can be overridden by implementation classes to return non-default values.

The UntypedPersistentActorWithAtLeastOnceDelivery class holds messages in memory until their successful delivery has been confirmed. The maximum number of unconfirmed messages that the actor is allowed to hold in memory is defined by the maxUnconfirmedMessages method. If this limit is exceed the deliver method will not accept more messages and it will throw AtLeastOnceDelivery.MaxUnconfirmedMessagesExceededException. The default value can be configured with the akka.persistence.at-least-once-delivery.max-unconfirmed-messages configuration key. The method can be overridden by implementation classes to return non-default values.

Eager initialization of persistence plugin

By default, persistence plugins are started on-demand, as they are used. In some case, however, it might be beneficial to start a certain plugin eagerly. In order to do that, you should first add the akka.persistence.Persistence under the akka.extensions key. Then, specify the IDs of plugins you wish to start automatically under akka.persistence.journal.auto-start-journals and akka.persistence.snapshot-store.auto-start-snapshot-stores.

Journal plugin API

A journal plugin extends AsyncWriteJournal.

AsyncWriteJournal is an actor and the methods to be implemented are:

/**
 * Java API, Plugin API: asynchronously writes a batch (`Iterable`) of
 * persistent messages to the journal.
 *
 * The batch is only for performance reasons, i.e. all messages don't have to
 * be written atomically. Higher throughput can typically be achieved by using
 * batch inserts of many records compared to inserting records one-by-one, but
 * this aspect depends on the underlying data store and a journal
 * implementation can implement it as efficient as possible. Journals should
 * aim to persist events in-order for a given `persistenceId` as otherwise in
 * case of a failure, the persistent state may be end up being inconsistent.
 * 
 * Each `AtomicWrite` message contains the single `PersistentRepr` that
 * corresponds to the event that was passed to the `persist` method of the
 * `PersistentActor`, or it contains several `PersistentRepr` that corresponds
 * to the events that were passed to the `persistAll` method of the
 * `PersistentActor`. All `PersistentRepr` of the `AtomicWrite` must be
 * written to the data store atomically, i.e. all or none must be stored. If
 * the journal (data store) cannot support atomic writes of multiple events it
 * should reject such writes with an `Optional` with an
 * `UnsupportedOperationException` describing the issue. This limitation
 * should also be documented by the journal plugin.
 *
 * If there are failures when storing any of the messages in the batch the
 * returned `Future` must be completed with failure. The `Future` must only be
 * completed with success when all messages in the batch have been confirmed
 * to be stored successfully, i.e. they will be readable, and visible, in a
 * subsequent replay. If there is uncertainty about if the messages were
 * stored or not the `Future` must be completed with failure.
 *
 * Data store connection problems must be signaled by completing the `Future`
 * with failure.
 *
 * The journal can also signal that it rejects individual messages
 * (`AtomicWrite`) by the returned
 * `Iterable<Optional<Exception>>`. The returned `Iterable` must
 * have as many elements as the input `messages` `Iterable`. Each `Optional`
 * element signals if the corresponding `AtomicWrite` is rejected or not, with
 * an exception describing the problem. Rejecting a message means it was not
 * stored, i.e. it must not be included in a later replay. Rejecting a message
 * is typically done before attempting to store it, e.g. because of
 * serialization error.
 *
 * Data store connection problems must not be signaled as rejections.
 *
 * Note that it is possible to reduce number of allocations by caching some
 * result `Iterable` for the happy path, i.e. when no messages are rejected.
 *
 * Calls to this method are serialized by the enclosing journal actor. If you spawn
 * work in asynchronous tasks it is alright that they complete the futures in any order,
 * but the actual writes for a specific persistenceId should be serialized to avoid 
 * issues such as events of a later write are visible to consumers (query side, or replay)
 * before the events of an earlier write are visible. This can also be done with
 * consistent hashing if it is too fine grained to do it on the persistenceId level.
 * Normally a `PersistentActor` will only have one outstanding write request to the journal but 
 * it may emit several write requests when `persistAsync` is used and the max batch size
 * is reached.
 *
 * This call is protected with a circuit-breaker.
 */
Future<Iterable<Optional<Exception>>> doAsyncWriteMessages(Iterable<AtomicWrite> messages);

/**
 * Java API, Plugin API: synchronously deletes all persistent messages up to
 * `toSequenceNr`.
 *
 * This call is protected with a circuit-breaker.
 *
 * @see AsyncRecoveryPlugin
 */
Future<Void> doAsyncDeleteMessagesTo(String persistenceId, long toSequenceNr);

If the storage backend API only supports synchronous, blocking writes, the methods should be implemented as:

@Override
public Future<Iterable<Optional<Exception>>> doAsyncWriteMessages(
  Iterable<AtomicWrite> messages) {
  try {
    Iterable<Optional<Exception>> result = new ArrayList<Optional<Exception>>();
    // blocking call here...
    // result.add(..)
    return Futures.successful(result);
  } catch (Exception e) {
    return Futures.failed(e);
  }
}

A journal plugin must also implement the methods defined in AsyncRecovery for replays and sequence number recovery:

/**
 * Java API, Plugin API: asynchronously replays persistent messages.
 * Implementations replay a message by calling `replayCallback`. The returned
 * future must be completed when all messages (matching the sequence number
 * bounds) have been replayed. The future must be completed with a failure if
 * any of the persistent messages could not be replayed.
 *
 * The `replayCallback` must also be called with messages that have been
 * marked as deleted. In this case a replayed message's `deleted` method must
 * return `true`.
 *
 * The `toSequenceNr` is the lowest of what was returned by
 * {@link #doAsyncReadHighestSequenceNr} and what the user specified as
 * recovery {@link akka.persistence.Recovery} parameter.
 *
 * @param persistenceId
 *          id of the persistent actor.
 * @param fromSequenceNr
 *          sequence number where replay should start (inclusive).
 * @param toSequenceNr
 *          sequence number where replay should end (inclusive).
 * @param max
 *          maximum number of messages to be replayed.
 * @param replayCallback
 *          called to replay a single message. Can be called from any thread.
 */
Future<Void> doAsyncReplayMessages(String persistenceId, long fromSequenceNr,
    long toSequenceNr, long max, Consumer<PersistentRepr> replayCallback);

/**
 * Java API, Plugin API: asynchronously reads the highest stored sequence
 * number for the given `persistenceId`. The persistent actor will use the
 * highest sequence number after recovery as the starting point when
 * persisting new events. This sequence number is also used as `toSequenceNr`
 * in subsequent call to [[#asyncReplayMessages]] unless the user has
 * specified a lower `toSequenceNr`.
 *
 * @param persistenceId
 *          id of the persistent actor.
 * @param fromSequenceNr
 *          hint where to start searching for the highest sequence number.
 */
Future<Long> doAsyncReadHighestSequenceNr(String persistenceId, long fromSequenceNr);

A journal plugin can be activated with the following minimal configuration:

# Path to the journal plugin to be used
akka.persistence.journal.plugin = "my-journal"

# My custom journal plugin
my-journal {
  # Class name of the plugin.
  class = "docs.persistence.MyJournal"
  # Dispatcher for the plugin actor.
  plugin-dispatcher = "akka.actor.default-dispatcher"
}

The specified plugin class must have a no-arg constructor. The plugin-dispatcher is the dispatcher used for the plugin actor. If not specified, it defaults to akka.persistence.dispatchers.default-plugin-dispatcher.

The journal plugin instance is an actor so the methods corresponding to requests from persistent actors are executed sequentially. It may delegate to asynchronous libraries, spawn futures, or delegate to other actors to achive parallelism.

The journal plugin class must have a constructor with one of these signatures:

• constructor with one com.typesafe.config.Config parameter and a String parameter for the config path
• constructor with one com.typesafe.config.Config parameter
• constructor without parameters

The plugin section of the actor system's config will be passed in the config constructor parameter. The config path of the plugin is passed in the String parameter.

Don't run journal tasks/futures on the system default dispatcher, since that might starve other tasks.

Snapshot store plugin API

A snapshot store plugin must extend the SnapshotStore actor and implement the following methods:

/**
 * Java API, Plugin API: asynchronously loads a snapshot.
 *
 * @param persistenceId
 *          id of the persistent actor.
 * @param criteria
 *          selection criteria for loading.
 */
Future<Optional<SelectedSnapshot>> doLoadAsync(String persistenceId, 
    SnapshotSelectionCriteria criteria);

/**
 * Java API, Plugin API: asynchronously saves a snapshot.
 *
 * @param metadata
 *          snapshot metadata.
 * @param snapshot
 *          snapshot.
 */
Future<Void> doSaveAsync(SnapshotMetadata metadata, Object snapshot);

/**
 * Java API, Plugin API: deletes the snapshot identified by `metadata`.
 *
 * @param metadata
 *          snapshot metadata.
 */
Future<Void> doDeleteAsync(SnapshotMetadata metadata);

/**
 * Java API, Plugin API: deletes all snapshots matching `criteria`.
 *
 * @param persistenceId
 *          id of the persistent actor.
 * @param criteria
 *          selection criteria for deleting.
 */
Future<Void> doDeleteAsync(String persistenceId, SnapshotSelectionCriteria criteria);

A snapshot store plugin can be activated with the following minimal configuration:

# Path to the snapshot store plugin to be used
akka.persistence.snapshot-store.plugin = "my-snapshot-store"

# My custom snapshot store plugin
my-snapshot-store {
  # Class name of the plugin.
  class = "docs.persistence.MySnapshotStore"
  # Dispatcher for the plugin actor.
  plugin-dispatcher = "akka.persistence.dispatchers.default-plugin-dispatcher"
}

The snapshot store instance is an actor so the methods corresponding to requests from persistent actors are executed sequentially. It may delegate to asynchronous libraries, spawn futures, or delegate to other actors to achive parallelism.

The snapshot store plugin class must have a constructor with one of these signatures:

• constructor with one com.typesafe.config.Config parameter and a String parameter for the config path
• constructor with one com.typesafe.config.Config parameter
• constructor without parameters

The plugin section of the actor system's config will be passed in the config constructor parameter. The config path of the plugin is passed in the String parameter.

The plugin-dispatcher is the dispatcher used for the plugin actor. If not specified, it defaults to akka.persistence.dispatchers.default-plugin-dispatcher.

Don't run snapshot store tasks/futures on the system default dispatcher, since that might starve other tasks.

Plugin TCK

In order to help developers build correct and high quality storage plugins, we provide a Technology Compatibility Kit (TCK for short).

The TCK is usable from Java as well as Scala projects. For Java you need to include the akka-persistence-tck dependency:

<dependency>
  <groupId>com.typesafe.akka</groupId>
  <artifactId>akka-persistence-tck_${scala.version}</artifactId>
  <version>2.4.20</version>
  <scope>test</scope>
</dependency>

To include the Journal TCK tests in your test suite simply extend the provided JavaJournalSpec:

class MyJournalSpecTest extends JavaJournalSpec {

  public MyJournalSpecTest() {
    super(ConfigFactory.parseString(
      "persistence.journal.plugin = " +
        "\"akka.persistence.journal.leveldb-shared\""));
  }

  @Override
  public CapabilityFlag supportsRejectingNonSerializableObjects() {
    return CapabilityFlag.off();
  }
}

Please note that some of the tests are optional, and by overriding the supports... methods you give the TCK the needed information about which tests to run. You can implement these methods using the provided CapabilityFlag.on / CapabilityFlag.off values.

We also provide a simple benchmarking class JavaJournalPerfSpec which includes all the tests that JavaJournalSpec has, and also performs some longer operations on the Journal while printing its performance stats. While it is NOT aimed to provide a proper benchmarking environment it can be used to get a rough feel about your journal's performance in the most typical scenarios.

In order to include the SnapshotStore TCK tests in your test suite simply extend the SnapshotStoreSpec:

class MySnapshotStoreTest extends JavaSnapshotStoreSpec {

  public MySnapshotStoreTest() {
    super(ConfigFactory.parseString(
      "akka.persistence.snapshot-store.plugin = " +
        "\"akka.persistence.snapshot-store.local\""));
  }
}

In case your plugin requires some setting up (starting a mock database, removing temporary files etc.) you can override the beforeAll and afterAll methods to hook into the tests lifecycle:

class MyJournalSpecTest extends JavaJournalSpec {

  List<File> storageLocations = new ArrayList<File>();

  public MyJournalSpecTest() {
    super(ConfigFactory.parseString(
      "persistence.journal.plugin = " +
        "\"akka.persistence.journal.leveldb-shared\""));

    Config config = system().settings().config();
    storageLocations.add(new File(
      config.getString("akka.persistence.journal.leveldb.dir")));
    storageLocations.add(new File(
      config.getString("akka.persistence.snapshot-store.local.dir")));
  }


  @Override
  public CapabilityFlag supportsRejectingNonSerializableObjects() {
    return CapabilityFlag.on();
  }

  @Override
  public void beforeAll() {
    for (File storageLocation : storageLocations) {
      FileUtils.deleteRecursively(storageLocation);
    }
    super.beforeAll();
  }

  @Override
  public void afterAll() {
    super.afterAll();
    for (File storageLocation : storageLocations) {
      FileUtils.deleteRecursively(storageLocation);
    }
  }
}

We highly recommend including these specifications in your test suite, as they cover a broad range of cases you might have otherwise forgotten to test for when writing a plugin from scratch.

Local LevelDB journal

The LevelDB journal plugin config entry is akka.persistence.journal.leveldb. It writes messages to a local LevelDB instance. Enable this plugin by defining config property:

# Path to the journal plugin to be used
akka.persistence.journal.plugin = "akka.persistence.journal.leveldb"

LevelDB based plugins will also require the following additional dependency declaration:

<dependency>
  <groupId>org.iq80.leveldb</groupId>
  <artifactId>leveldb</artifactId>
  <version>0.7</version>
</dependency>
<dependency>
  <groupId>org.fusesource.leveldbjni</groupId>
  <artifactId>leveldbjni-all</artifactId>
  <version>1.8</version>
</dependency>

The default location of the LevelDB files is a directory named journal in the current working directory. This location can be changed by configuration where the specified path can be relative or absolute:

akka.persistence.journal.leveldb.dir = "target/journal"

With this plugin, each actor system runs its own private LevelDB instance.

Shared LevelDB journal

A LevelDB instance can also be shared by multiple actor systems (on the same or on different nodes). This, for example, allows persistent actors to failover to a backup node and continue using the shared journal instance from the backup node.

A shared LevelDB instance is started by instantiating the SharedLeveldbStore actor.

final ActorRef store = system.actorOf(Props.create(SharedLeveldbStore.class), "store");

By default, the shared instance writes journaled messages to a local directory named journal in the current working directory. The storage location can be changed by configuration:

akka.persistence.journal.leveldb-shared.store.dir = "target/shared"

Actor systems that use a shared LevelDB store must activate the akka.persistence.journal.leveldb-shared plugin.

akka.persistence.journal.plugin = "akka.persistence.journal.leveldb-shared"

This plugin must be initialized by injecting the (remote) SharedLeveldbStore actor reference. Injection is done by calling the SharedLeveldbJournal.setStore method with the actor reference as argument.

class SharedStorageUsage extends UntypedActor {
  @Override
  public void preStart() throws Exception {
    String path = "akka.tcp://[email protected]:2552/user/store";
    ActorSelection selection = getContext().actorSelection(path);
    selection.tell(new Identify(1), getSelf());
  }

  @Override
  public void onReceive(Object message) throws Exception {
    if (message instanceof ActorIdentity) {
      ActorIdentity identity = (ActorIdentity) message;
      if (identity.correlationId().equals(1)) {
        ActorRef store = identity.getRef();
        if (store != null) {
          SharedLeveldbJournal.setStore(store, getContext().system());
        }
      }
    }
  }
}

Internal journal commands (sent by persistent actors) are buffered until injection completes. Injection is idempotent i.e. only the first injection is used.

Local snapshot store

The local snapshot store plugin config entry is akka.persistence.snapshot-store.local. It writes snapshot files to the local filesystem. Enable this plugin by defining config property:

# Path to the snapshot store plugin to be used
akka.persistence.snapshot-store.plugin = "akka.persistence.snapshot-store.local"

The default storage location is a directory named snapshots in the current working directory. This can be changed by configuration where the specified path can be relative or absolute:

akka.persistence.snapshot-store.local.dir = "target/snapshots"

Note that it is not mandatory to specify a snapshot store plugin. If you don't use snapshots you don't have to configure it.

Persistence Plugin Proxy

A persistence plugin proxy allows sharing of journals and snapshot stores across multiple actor systems (on the same or on different nodes). This, for example, allows persistent actors to failover to a backup node and continue using the shared journal instance from the backup node. The proxy works by forwarding all the journal/snapshot store messages to a single, shared, persistence plugin instance, and therefor supports any use case supported by the proxied plugin.

The journal and snapshot store proxies are controlled via the akka.persistence.journal.proxy and akka.persistence.snapshot-store.proxy configuration entries, respectively. Set the target-journal-plugin or target-snapshot-store-plugin keys to the underlying plugin you wish to use (for example: akka.persistence.journal.leveldb). The start-target-journal and start-target-snapshot-store keys should be set to on in exactly one actor system - this is the system that will instantiate the shared persistence plugin. Next, the proxy needs to be told how to find the shared plugin. This can be done by setting the target-journal-address and target-snapshot-store-address configuration keys, or programmatically by calling the PersistencePluginProxy.setTargetLocation method.