Interaction Patterns
You are viewing the documentation for the new actor APIs, to view the Akka Classic documentation, see Classic Actors.
Dependency
To use Akka Actor Typed, you must add the following dependency in your project:
- sbt
val AkkaVersion = "2.6.21" libraryDependencies += "com.typesafe.akka" %% "akka-actor-typed" % AkkaVersion
- Maven
- Gradle
Introduction
Interacting with an Actor in Akka is done through an ActorRef[T]
where T
is the type of messages the actor accepts, also known as the “protocol”. This ensures that only the right kind of messages can be sent to an actor and also that no one else but the Actor itself can access the Actor instance internals.
Message exchange with Actors follow a few common patterns, let’s go through each one of them.
Fire and Forget
The fundamental way to interact with an actor is through “tell”, which is so common that it has a special symbolic method name: actorRef
!
message
. Sending a message with tell can safely be done from any thread.
Tell is asynchronous which means that the method returns right away. After the statement is executed there is no guarantee that the message has been processed by the recipient yet. It also means there is no way to know if the message was received, the processing succeeded or failed.
Example:
With the given protocol and actor behavior:
- Scala
-
source
object Printer { case class PrintMe(message: String) def apply(): Behavior[PrintMe] = Behaviors.receive { case (context, PrintMe(message)) => context.log.info(message) Behaviors.same } }
- Java
Fire and forget looks like this:
- Scala
-
source
val system = ActorSystem(Printer(), "fire-and-forget-sample") // note how the system is also the top level actor ref val printer: ActorRef[Printer.PrintMe] = system // these are all fire and forget printer ! Printer.PrintMe("message 1") printer ! Printer.PrintMe("not message 2")
- Java
Useful when:
- It is not critical to be sure that the message was processed
- There is no way to act on non successful delivery or processing
- We want to minimize the number of messages created to get higher throughput (sending a response would require creating twice the number of messages)
Problems:
- If the inflow of messages is higher than the actor can process the inbox will fill up and can in the worst case cause the JVM crash with an
OutOfMemoryError
- If the message gets lost, the sender will not know
Request-Response
Many interactions between actors require one or more response message being sent back from the receiving actor. A response message can be a result of a query, some form of acknowledgment that the message was received and processed or events that the request subscribed to.
In Akka the recipient of responses has to be encoded as a field in the message itself, which the recipient can then use to send (tell) a response back.
Example:
With the following protocol:
- Scala
-
source
case class Request(query: String, replyTo: ActorRef[Response]) case class Response(result: String)
- Java
The sender would use its own ActorRef[Response]
, which it can access through ActorContext.self
, for the replyTo
.
On the receiving side the ActorRef[Response]
can then be used to send one or more responses back:
- Scala
-
source
def apply(): Behaviors.Receive[Request] = Behaviors.receiveMessage[Request] { case Request(query, replyTo) => // ... process query ... replyTo ! Response(s"Here are the cookies for [$query]!") Behaviors.same }
- Java
Useful when:
- Subscribing to an actor that will send many response messages back
Problems:
- Actors seldom have a response message from another actor as a part of their protocol (see adapted response)
- It is hard to detect that a message request was not delivered or processed (see ask)
- Unless the protocol already includes a way to provide context, for example a request id that is also sent in the response, it is not possible to tie an interaction to some specific context without introducing a new, separate, actor (see ask or per session child actor)
Adapted Response
Most often the sending actor does not, and should not, support receiving the response messages of another actor. In such cases we need to provide an ActorRef
of the right type and adapt the response message to a type that the sending actor can handle.
Example:
- Scala
-
source
object Backend { sealed trait Request final case class StartTranslationJob(taskId: Int, site: URI, replyTo: ActorRef[Response]) extends Request sealed trait Response final case class JobStarted(taskId: Int) extends Response final case class JobProgress(taskId: Int, progress: Double) extends Response final case class JobCompleted(taskId: Int, result: URI) extends Response } object Frontend { sealed trait Command final case class Translate(site: URI, replyTo: ActorRef[URI]) extends Command private final case class WrappedBackendResponse(response: Backend.Response) extends Command def apply(backend: ActorRef[Backend.Request]): Behavior[Command] = Behaviors.setup[Command] { context => val backendResponseMapper: ActorRef[Backend.Response] = context.messageAdapter(rsp => WrappedBackendResponse(rsp)) def active(inProgress: Map[Int, ActorRef[URI]], count: Int): Behavior[Command] = { Behaviors.receiveMessage[Command] { case Translate(site, replyTo) => val taskId = count + 1 backend ! Backend.StartTranslationJob(taskId, site, backendResponseMapper) active(inProgress.updated(taskId, replyTo), taskId) case wrapped: WrappedBackendResponse => wrapped.response match { case Backend.JobStarted(taskId) => context.log.info("Started {}", taskId) Behaviors.same case Backend.JobProgress(taskId, progress) => context.log.info2("Progress {}: {}", taskId, progress) Behaviors.same case Backend.JobCompleted(taskId, result) => context.log.info2("Completed {}: {}", taskId, result) inProgress(taskId) ! result active(inProgress - taskId, count) } } } active(inProgress = Map.empty, count = 0) } } - Java
You can register several message adapters for different message classes. It’s only possible to have one message adapter per message class to make sure that the number of adapters are not growing unbounded if registered repeatedly. That also means that a registered adapter will replace an existing adapter for the same message class.
A message adapter will be used if the message class matches the given class or is a subclass thereof. The registered adapters are tried in reverse order of their registration order, i.e. the last registered first.
A message adapter (and the returned ActorRef
) has the same lifecycle as the receiving actor. It’s recommended to register the adapters in a top level Behaviors.setup
or constructor of AbstractBehavior
but it’s possible to register them later if needed.
The adapter function is running in the receiving actor and can safely access its state, but if it throws an exception the actor is stopped.
Useful when:
- Translating between different actor message protocols
- Subscribing to an actor that will send many response messages back
Problems:
- It is hard to detect that a message request was not delivered or processed (see ask)
- Only one adaption can be made per response message type, if a new one is registered the old one is replaced, for example different target actors can’t have different adaption if they use the same response types, unless some correlation is encoded in the messages
- Unless the protocol already includes a way to provide context, for example a request id that is also sent in the response, it is not possible to tie an interaction to some specific context without introducing a new, separate, actor
Request-Response with ask between two actors
In an interaction where there is a 1:1 mapping between a request and a response we can use ask
on the ActorContext
to interact with another actor.
The interaction has two steps, first we need to construct the outgoing message, to do that we need an ActorRef[Response]
to put as recipient in the outgoing message. The second step is to transform the successful Response
or failure into a message that is part of the protocol of the sending actor. See also the Generic response wrapper for replies that are either a success or an error.
Example:
- Scala
-
source
object Hal { sealed trait Command case class OpenThePodBayDoorsPlease(replyTo: ActorRef[Response]) extends Command case class Response(message: String) def apply(): Behaviors.Receive[Hal.Command] = Behaviors.receiveMessage[Command] { case OpenThePodBayDoorsPlease(replyTo) => replyTo ! Response("I'm sorry, Dave. I'm afraid I can't do that.") Behaviors.same } } object Dave { sealed trait Command // this is a part of the protocol that is internal to the actor itself private case class AdaptedResponse(message: String) extends Command def apply(hal: ActorRef[Hal.Command]): Behavior[Dave.Command] = Behaviors.setup[Command] { context => // asking someone requires a timeout, if the timeout hits without response // the ask is failed with a TimeoutException implicit val timeout: Timeout = 3.seconds // Note: The second parameter list takes a function `ActorRef[T] => Message`, // as OpenThePodBayDoorsPlease is a case class it has a factory apply method // that is what we are passing as the second parameter here it could also be written // as `ref => OpenThePodBayDoorsPlease(ref)` context.ask(hal, Hal.OpenThePodBayDoorsPlease.apply) { case Success(Hal.Response(message)) => AdaptedResponse(message) case Failure(_) => AdaptedResponse("Request failed") } // we can also tie in request context into an interaction, it is safe to look at // actor internal state from the transformation function, but remember that it may have // changed at the time the response arrives and the transformation is done, best is to // use immutable state we have closed over like here. val requestId = 1 context.ask(hal, Hal.OpenThePodBayDoorsPlease.apply) { case Success(Hal.Response(message)) => AdaptedResponse(s"$requestId: $message") case Failure(_) => AdaptedResponse(s"$requestId: Request failed") } Behaviors.receiveMessage { // the adapted message ends up being processed like any other // message sent to the actor case AdaptedResponse(message) => context.log.info("Got response from hal: {}", message) Behaviors.same } } }
- Java
The response adapting function is running in the receiving actor and can safely access its state, but if it throws an exception the actor is stopped.
Useful when:
- Single response queries
- An actor needs to know that the message was processed before continuing
- To allow an actor to resend if a timely response is not produced
- To keep track of outstanding requests and not overwhelm a recipient with messages (“backpressure”)
- Context should be attached to the interaction but the protocol does not support that (request id, what query the response was for)
Problems:
- There can only be a single response to one
ask
(see per session child Actor) - When
ask
times out, the receiving actor does not know and may still process it to completion, or even start processing it after the fact - Finding a good value for the timeout, especially when
ask
triggers chainedask
s in the receiving actor. You want a short timeout to be responsive and answer back to the requester, but at the same time you do not want to have many false positives
Request-Response with ask from outside an Actor
Sometimes you need to interact with actors from the outside of the actor system, this can be done with fire-and-forget as described above or through another version of ask
that returns a Future[Response]
that is either completed with a successful response or failed with a TimeoutException
if there was no response within the specified timeout.
To do this we use ask
(or the symbolic ?
) implicitly added to ActorRef
by akka.actor.typed.scaladsl.AskPattern._
to send a message to an actor and get a Future[Response]
back. ask
takes implicit Timeout
and ActorSystem
parameters.
Example:
- Scala
-
source
object CookieFabric { sealed trait Command case class GiveMeCookies(count: Int, replyTo: ActorRef[Reply]) extends Command sealed trait Reply case class Cookies(count: Int) extends Reply case class InvalidRequest(reason: String) extends Reply def apply(): Behaviors.Receive[CookieFabric.GiveMeCookies] = Behaviors.receiveMessage { message => if (message.count >= 5) message.replyTo ! InvalidRequest("Too many cookies.") else message.replyTo ! Cookies(message.count) Behaviors.same } } import akka.actor.typed.scaladsl.AskPattern._ import akka.util.Timeout // asking someone requires a timeout if the timeout hits without response // the ask is failed with a TimeoutException implicit val timeout: Timeout = 3.seconds // implicit ActorSystem in scope implicit val system: ActorSystem[_] = theSystem val result: Future[CookieFabric.Reply] = cookieFabric.ask(ref => CookieFabric.GiveMeCookies(3, ref)) // the response callback will be executed on this execution context implicit val ec = system.executionContext result.onComplete { case Success(CookieFabric.Cookies(count)) => println(s"Yay, $count cookies!") case Success(CookieFabric.InvalidRequest(reason)) => println(s"No cookies for me. $reason") case Failure(ex) => println(s"Boo! didn't get cookies: ${ex.getMessage}") }
- Java
Note that validation errors are also explicit in the message protocol. The GiveMeCookies
request can reply with Cookies
or InvalidRequest
. The requestor has to decide how to handle an InvalidRequest
reply. Sometimes it should be treated as a failed Future
and for that the reply can be mapped on the requestor side. See also the Generic response wrapper for replies that are either a success or an error.
- Scala
-
source
val cookies: Future[CookieFabric.Cookies] = cookieFabric.ask[CookieFabric.Reply](ref => CookieFabric.GiveMeCookies(3, ref)).flatMap { case c: CookieFabric.Cookies => Future.successful(c) case CookieFabric.InvalidRequest(reason) => Future.failed(new IllegalArgumentException(reason)) } cookies.onComplete { case Success(CookieFabric.Cookies(count)) => println(s"Yay, $count cookies!") case Failure(ex) => println(s"Boo! didn't get cookies: ${ex.getMessage}") }
- Java
Useful when:
- Querying an actor from outside of the actor system
Problems:
- It is easy to accidentally close over and unsafely mutable state with the callbacks on the returned
Future
as those will be executed on a different thread - There can only be a single response to one
ask
(see per session child Actor) - When
ask
times out, the receiving actor does not know and may still process it to completion, or even start processing it after the fact
Generic response wrapper
In many cases the response can either be a successful result or an error (a validation error that the command was invalid for example). Having to define two response classes and a shared supertype for every request type can be repetitive, especially in a cluster context where you also have to make sure the messages can be serialized to be sent over the network.
To help with this a generic status-response type is included in Akka: StatusReply
, everywhere where ask
can be used there is also a second method askWithStatus
which, given that the response is a StatusReply
will unwrap successful responses and help with handling validation errors. Akka includes pre-built serializers for the type, so in the normal use case a clustered application only needs to provide a serializer for the successful result.
For the case where the successful reply does not contain an actual value but is more of an acknowledgment there is a pre defined StatusReply.Ack
of type StatusReply[Done]
.
Errors are preferably sent as a text describing what is wrong, but using exceptions to attach a type is also possible.
Example actor to actor ask:
- Scala
-
source
object Hal { sealed trait Command case class OpenThePodBayDoorsPlease(replyTo: ActorRef[StatusReply[String]]) extends Command def apply(): Behaviors.Receive[Hal.Command] = Behaviors.receiveMessage[Command] { case OpenThePodBayDoorsPlease(replyTo) => // reply with a validation error description replyTo ! StatusReply.Error("I'm sorry, Dave. I'm afraid I can't do that.") Behaviors.same } } object Dave { sealed trait Command // this is a part of the protocol that is internal to the actor itself private case class AdaptedResponse(message: String) extends Command def apply(hal: ActorRef[Hal.Command]): Behavior[Dave.Command] = Behaviors.setup[Command] { context => // asking someone requires a timeout, if the timeout hits without response // the ask is failed with a TimeoutException implicit val timeout: Timeout = 3.seconds // A StatusReply.Success(m) ends up as a Success(m) here, while a // StatusReply.Error(text) becomes a Failure(ErrorMessage(text)) context.askWithStatus(hal, Hal.OpenThePodBayDoorsPlease.apply) { case Success(message) => AdaptedResponse(message) case Failure(StatusReply.ErrorMessage(text)) => AdaptedResponse(s"Request denied: $text") case Failure(_) => AdaptedResponse("Request failed") } Behaviors.receiveMessage { // the adapted message ends up being processed like any other // message sent to the actor case AdaptedResponse(message) => context.log.info("Got response from hal: {}", message) Behaviors.same } } }
- Java
A validation error is turned into a Failure
for the message adapter. In this case we are explicitly handling the validation error separately from other ask failures.
Example ask from the outside:
- Scala
-
source
object CookieFabric { sealed trait Command case class GiveMeCookies(count: Int, replyTo: ActorRef[StatusReply[Cookies]]) extends Command case class Cookies(count: Int) def apply(): Behaviors.Receive[CookieFabric.GiveMeCookies] = Behaviors.receiveMessage { message => if (message.count >= 5) message.replyTo ! StatusReply.Error("Too many cookies.") else message.replyTo ! StatusReply.Success(Cookies(message.count)) Behaviors.same } } import akka.actor.typed.scaladsl.AskPattern._ import akka.util.Timeout // asking someone requires a timeout if the timeout hits without response // the ask is failed with a TimeoutException implicit val timeout: Timeout = 3.seconds // implicit ActorSystem in scope implicit val system: ActorSystem[_] = theSystem val result: Future[CookieFabric.Cookies] = cookieFabric.askWithStatus(ref => CookieFabric.GiveMeCookies(3, ref)) // the response callback will be executed on this execution context implicit val ec = system.executionContext result.onComplete { case Success(CookieFabric.Cookies(count)) => println(s"Yay, $count cookies!") case Failure(StatusReply.ErrorMessage(reason)) => println(s"No cookies for me. $reason") case Failure(ex) => println(s"Boo! didn't get cookies: ${ex.getMessage}") }
- Java
Note that validation errors are also explicit in the message protocol, but encoded as the wrapper type, constructed using StatusReply.Error(text)
:
- Scala
-
source
val cookies: Future[CookieFabric.Cookies] = cookieFabric.askWithStatus[CookieFabric.Cookies](ref => CookieFabric.GiveMeCookies(3, ref)).flatMap { case c: CookieFabric.Cookies => Future.successful(c) } cookies.onComplete { case Success(CookieFabric.Cookies(count)) => println(s"Yay, $count cookies!") case Failure(ex) => println(s"Boo! didn't get cookies: ${ex.getMessage}") }
- Java
Ignoring replies
In some situations an actor has a response for a particular request message but you are not interested in the response. In this case you can pass system.ignoreRef
turning the request-response into a fire-and-forget.
system.ignoreRef
, as the name indicates, returns an ActorRef
that ignores any message sent to it.
With the same protocol as the request response above, if the sender would prefer to ignore the reply it could pass system.ignoreRef
for the replyTo
, which it can access through ActorContext.system.ignoreRef
.
- Scala
-
source
cookieFabric ! CookieFabric.Request("don't send cookies back", context.system.ignoreRef)
- Java
Useful when:
- Sending a message for which the protocol defines a reply, but you are not interested in getting the reply
Problems:
The returned ActorRef
ignores all messages sent to it, therefore it should be used carefully.
- Passing it around inadvertently as if it was a normal
ActorRef
may result in broken actor-to-actor interactions. - Using it when performing an
ask
from outside the Actor System will cause theFuture
returned by theask
to timeout since it will never complete. - Finally, it’s legal to
watch
it, but since it’s of a special kind, it never terminates and therefore you will never receive aTerminated
signal from it.
Send Future result to self
When using an API that returns a Future
from an actor it’s common that you would like to use the value of the response in the actor when the Future
is completed. For this purpose the ActorContext
provides a pipeToSelf
method.
Example:
An actor, CustomerRepository
, is invoking a method on CustomerDataAccess
that returns a Future
.
- Scala
-
source
trait CustomerDataAccess { def update(value: Customer): Future[Done] } final case class Customer(id: String, version: Long, name: String, address: String) object CustomerRepository { sealed trait Command final case class Update(value: Customer, replyTo: ActorRef[UpdateResult]) extends Command sealed trait UpdateResult final case class UpdateSuccess(id: String) extends UpdateResult final case class UpdateFailure(id: String, reason: String) extends UpdateResult private final case class WrappedUpdateResult(result: UpdateResult, replyTo: ActorRef[UpdateResult]) extends Command private val MaxOperationsInProgress = 10 def apply(dataAccess: CustomerDataAccess): Behavior[Command] = { next(dataAccess, operationsInProgress = 0) } private def next(dataAccess: CustomerDataAccess, operationsInProgress: Int): Behavior[Command] = { Behaviors.receive { (context, command) => command match { case Update(value, replyTo) => if (operationsInProgress == MaxOperationsInProgress) { replyTo ! UpdateFailure(value.id, s"Max $MaxOperationsInProgress concurrent operations supported") Behaviors.same } else { val futureResult = dataAccess.update(value) context.pipeToSelf(futureResult) { // map the Future value to a message, handled by this actor case Success(_) => WrappedUpdateResult(UpdateSuccess(value.id), replyTo) case Failure(e) => WrappedUpdateResult(UpdateFailure(value.id, e.getMessage), replyTo) } // increase operationsInProgress counter next(dataAccess, operationsInProgress + 1) } case WrappedUpdateResult(result, replyTo) => // send result to original requestor replyTo ! result // decrease operationsInProgress counter next(dataAccess, operationsInProgress - 1) } } } } - Java
It could be tempting to just use onComplete on the Future
, but that introduces the risk of accessing internal state of the actor that is not thread-safe from an external thread. For example, the numberOfPendingOperations
counter in above example can’t be accessed from such callback. Therefore it is better to map the result to a message and perform further processing when receiving that message.
Useful when:
- Accessing APIs that are returning
Future
from an actor, such as a database or an external service - The actor needs to continue processing when the
Future
has completed - Keep context from the original request and use that when the
Future
has completed, for example anreplyTo
actor reference
Problems:
- Boilerplate of adding wrapper messages for the results
Per session child Actor
In some cases a complete response to a request can only be created and sent back after collecting multiple answers from other actors. For these kinds of interaction it can be good to delegate the work to a per “session” child actor. The child could also contain arbitrary logic to implement retrying, failing on timeout, tail chopping, progress inspection etc.
Note that this is essentially how ask
is implemented, if all you need is a single response with a timeout it is better to use ask
.
The child is created with the context it needs to do the work, including an ActorRef
that it can respond to. When the complete result is there the child responds with the result and stops itself.
As the protocol of the session actor is not a public API but rather an implementation detail of the parent actor, it may not always make sense to have an explicit protocol and adapt the messages of the actors that the session actor interacts with. For this use case it is possible to express that the actor can receive any message (Any
).
Example:
- Scala
-
source
// dummy data types just for this sample case class Keys() case class Wallet() object Home { sealed trait Command case class LeaveHome(who: String, replyTo: ActorRef[ReadyToLeaveHome]) extends Command case class ReadyToLeaveHome(who: String, keys: Keys, wallet: Wallet) def apply(): Behavior[Command] = { Behaviors.setup[Command] { context => val keyCabinet: ActorRef[KeyCabinet.GetKeys] = context.spawn(KeyCabinet(), "key-cabinet") val drawer: ActorRef[Drawer.GetWallet] = context.spawn(Drawer(), "drawer") Behaviors.receiveMessage[Command] { case LeaveHome(who, replyTo) => context.spawn(prepareToLeaveHome(who, replyTo, keyCabinet, drawer), s"leaving-$who") Behaviors.same } } } // per session actor behavior def prepareToLeaveHome( whoIsLeaving: String, replyTo: ActorRef[ReadyToLeaveHome], keyCabinet: ActorRef[KeyCabinet.GetKeys], drawer: ActorRef[Drawer.GetWallet]): Behavior[NotUsed] = { // we don't _really_ care about the actor protocol here as nobody will send us // messages except for responses to our queries, so we just accept any kind of message // but narrow that to more limited types when we interact Behaviors .setup[AnyRef] { context => var wallet: Option[Wallet] = None var keys: Option[Keys] = None // we narrow the ActorRef type to any subtype of the actual type we accept keyCabinet ! KeyCabinet.GetKeys(whoIsLeaving, context.self.narrow[Keys]) drawer ! Drawer.GetWallet(whoIsLeaving, context.self.narrow[Wallet]) def nextBehavior(): Behavior[AnyRef] = (keys, wallet) match { case (Some(w), Some(k)) => // we got both, "session" is completed! replyTo ! ReadyToLeaveHome(whoIsLeaving, w, k) Behaviors.stopped case _ => Behaviors.same } Behaviors.receiveMessage { case w: Wallet => wallet = Some(w) nextBehavior() case k: Keys => keys = Some(k) nextBehavior() case _ => Behaviors.unhandled } } .narrow[NotUsed] // we don't let anyone else know we accept anything } }
- Java
In an actual session child you would likely want to include some form of timeout as well (see scheduling messages to self).
Useful when:
- A single incoming request should result in multiple interactions with other actors before a result can be built, for example aggregation of several results
- You need to handle acknowledgement and retry messages for at-least-once delivery
Problems:
- Children have life cycles that must be managed to not create a resource leak, it can be easy to miss a scenario where the session actor is not stopped
- It increases complexity, since each such child can execute concurrently with other children and the parent
General purpose response aggregator
This is similar to above Per session child Actor pattern. Sometimes you might end up repeating the same way of aggregating replies and want to extract that to a reusable actor.
There are many variations of this pattern and that is the reason this is provided as a documentation example rather than a built in Behavior
in Akka. It is intended to be adjusted to your specific needs.
Example:
This example is an aggregator of expected number of replies. Requests for quotes are sent with the given sendRequests
function to the two hotel actors, which both speak different protocols. When both expected replies have been collected they are aggregated with the given aggregateReplies
function and sent back to the replyTo
. If replies don’t arrive within the timeout
the replies so far are aggregated and sent back to the replyTo
.
- Scala
-
source
object Hotel1 { final case class RequestQuote(replyTo: ActorRef[Quote]) final case class Quote(hotel: String, price: BigDecimal) } object Hotel2 { final case class RequestPrice(replyTo: ActorRef[Price]) final case class Price(hotel: String, price: BigDecimal) } // Any since no common type between Hotel1 and Hotel2 type Reply = Any object HotelCustomer { sealed trait Command final case class Quote(hotel: String, price: BigDecimal) final case class AggregatedQuotes(quotes: List[Quote]) extends Command def apply(hotel1: ActorRef[Hotel1.RequestQuote], hotel2: ActorRef[Hotel2.RequestPrice]): Behavior[Command] = { Behaviors.setup[Command] { context => context.spawnAnonymous( Aggregator[Reply, AggregatedQuotes]( sendRequests = { replyTo => hotel1 ! Hotel1.RequestQuote(replyTo) hotel2 ! Hotel2.RequestPrice(replyTo) }, expectedReplies = 2, context.self, aggregateReplies = replies => // The hotels have different protocols with different replies, // convert them to `HotelCustomer.Quote` that this actor understands. AggregatedQuotes( replies .map { case Hotel1.Quote(hotel, price) => Quote(hotel, price) case Hotel2.Price(hotel, price) => Quote(hotel, price) case unknown => throw new RuntimeException(s"Unknown reply $unknown") } .sortBy(_.price) .toList), timeout = 5.seconds)) Behaviors.receiveMessage { case AggregatedQuotes(quotes) => context.log.info("Best {}", quotes.headOption.getOrElse("Quote N/A")) Behaviors.same } } } }
- Java
The implementation of the Aggregator
:
- Scala
-
source
import scala.collection.immutable import scala.concurrent.duration.FiniteDuration import scala.reflect.ClassTag import akka.actor.typed.ActorRef import akka.actor.typed.Behavior import akka.actor.typed.scaladsl.Behaviors object Aggregator { sealed trait Command private case object ReceiveTimeout extends Command private case class WrappedReply[R](reply: R) extends Command def apply[Reply: ClassTag, Aggregate]( sendRequests: ActorRef[Reply] => Unit, expectedReplies: Int, replyTo: ActorRef[Aggregate], aggregateReplies: immutable.IndexedSeq[Reply] => Aggregate, timeout: FiniteDuration): Behavior[Command] = { Behaviors.setup { context => context.setReceiveTimeout(timeout, ReceiveTimeout) val replyAdapter = context.messageAdapter[Reply](WrappedReply(_)) sendRequests(replyAdapter) def collecting(replies: immutable.IndexedSeq[Reply]): Behavior[Command] = { Behaviors.receiveMessage { case WrappedReply(reply) => val newReplies = replies :+ reply.asInstanceOf[Reply] if (newReplies.size == expectedReplies) { val result = aggregateReplies(newReplies) replyTo ! result Behaviors.stopped } else collecting(newReplies) case ReceiveTimeout => val aggregate = aggregateReplies(replies) replyTo ! aggregate Behaviors.stopped } } collecting(Vector.empty) } } }
- Java
Useful when:
- Aggregating replies are performed in the same way at multiple places and should be extracted to a more general purpose actor.
- A single incoming request should result in multiple interactions with other actors before a result can be built, for example aggregation of several results
- You need to handle acknowledgement and retry messages for at-least-once delivery
Problems:
- Message protocols with generic types are difficult since the generic types are erased in runtime
- Children have life cycles that must be managed to not create a resource leak, it can be easy to miss a scenario where the session actor is not stopped
- It increases complexity, since each such child can execute concurrently with other children and the parent
Latency tail chopping
This is a variation of above General purpose response aggregator pattern.
The goal of this algorithm is to decrease tail latencies (“chop off the tail latency”) in situations where multiple destination actors can perform the same piece of work, and where an actor may occasionally respond more slowly than expected. In this case, sending the same work request (also known as a “backup request”) to another actor results in decreased response time - because it’s less probable that multiple actors are under heavy load simultaneously. This technique is explained in depth in Jeff Dean’s presentation on Achieving Rapid Response Times in Large Online Services.
There are many variations of this pattern and that is the reason this is provided as a documentation example rather than a built in Behavior
in Akka. It is intended to be adjusted to your specific needs.
Example:
- Scala
-
source
import scala.concurrent.duration.FiniteDuration import scala.reflect.ClassTag import akka.actor.typed.ActorRef import akka.actor.typed.Behavior import akka.actor.typed.scaladsl.Behaviors object TailChopping { sealed trait Command private case object RequestTimeout extends Command private case object FinalTimeout extends Command private case class WrappedReply[R](reply: R) extends Command def apply[Reply: ClassTag]( sendRequest: (Int, ActorRef[Reply]) => Boolean, nextRequestAfter: FiniteDuration, replyTo: ActorRef[Reply], finalTimeout: FiniteDuration, timeoutReply: Reply): Behavior[Command] = { Behaviors.setup { context => Behaviors.withTimers { timers => val replyAdapter = context.messageAdapter[Reply](WrappedReply(_)) def waiting(requestCount: Int): Behavior[Command] = { Behaviors.receiveMessage { case WrappedReply(reply) => replyTo ! reply.asInstanceOf[Reply] Behaviors.stopped case RequestTimeout => sendNextRequest(requestCount + 1) case FinalTimeout => replyTo ! timeoutReply Behaviors.stopped } } def sendNextRequest(requestCount: Int): Behavior[Command] = { if (sendRequest(requestCount, replyAdapter)) { timers.startSingleTimer(RequestTimeout, nextRequestAfter) } else { timers.startSingleTimer(FinalTimeout, finalTimeout) } waiting(requestCount) } sendNextRequest(1) } } } }
- Java
Useful when:
- Reducing higher latency percentiles and variations of latency are important
- The “work” can be done more than once with the same result, e.g. a request to retrieve information
Problems:
- Increased load since more messages are sent and “work” is performed more than once
- Can’t be used when the “work” is not idempotent and must only be performed once
- Message protocols with generic types are difficult since the generic types are erased in runtime
- Children have life cycles that must be managed to not create a resource leak, it can be easy to miss a scenario where the session actor is not stopped
Scheduling messages to self
The following example demonstrates how to use timers to schedule messages to an actor.
Example:
The Buncher
actor buffers a burst of incoming messages and delivers them as a batch after a timeout or when the number of batched messages exceeds a maximum size.
- Scala
-
source
object Buncher { sealed trait Command final case class ExcitingMessage(message: String) extends Command final case class Batch(messages: Vector[Command]) private case object Timeout extends Command private case object TimerKey def apply(target: ActorRef[Batch], after: FiniteDuration, maxSize: Int): Behavior[Command] = { Behaviors.withTimers(timers => new Buncher(timers, target, after, maxSize).idle()) } } class Buncher( timers: TimerScheduler[Buncher.Command], target: ActorRef[Buncher.Batch], after: FiniteDuration, maxSize: Int) { import Buncher._ private def idle(): Behavior[Command] = { Behaviors.receiveMessage[Command] { message => timers.startSingleTimer(TimerKey, Timeout, after) active(Vector(message)) } } def active(buffer: Vector[Command]): Behavior[Command] = { Behaviors.receiveMessage[Command] { case Timeout => target ! Batch(buffer) idle() case m => val newBuffer = buffer :+ m if (newBuffer.size == maxSize) { timers.cancel(TimerKey) target ! Batch(newBuffer) idle() } else active(newBuffer) } } }
- Java
There are a few things worth noting here:
- To get access to the timers you start with
Behaviors.withTimers
that will pass aTimerScheduler
instance to the function. This can be used with any type ofBehavior
, includingreceive
,receiveMessage
, but alsosetup
or any other behavior. - Each timer has a key and if a new timer with the same key is started, the previous is cancelled. It is guaranteed that a message from the previous timer is not received, even if it was already enqueued in the mailbox when the new timer was started.
- Both periodic and single message timers are supported.
- The
TimerScheduler
is mutable in itself, because it performs and manages the side effects of registering the scheduled tasks. - The
TimerScheduler
is bound to the lifecycle of the actor that owns it and is cancelled automatically when the actor is stopped. Behaviors.withTimers
can also be used insideBehaviors.supervise
and it will automatically cancel the started timers correctly when the actor is restarted, so that the new incarnation will not receive scheduled messages from a previous incarnation.
Schedule periodically
Scheduling of recurring messages can have two different characteristics:
- fixed-delay - The delay between sending subsequent messages will always be (at least) the given
delay
. UsestartTimerWithFixedDelay
. - fixed-rate - The frequency of execution over time will meet the given
interval
. UsestartTimerAtFixedRate
.
If you are uncertain of which one to use you should pick startTimerWithFixedDelay
.
When using fixed-delay it will not compensate the delay between messages if the scheduling is delayed longer than specified for some reason. The delay between sending subsequent messages will always be (at least) the given delay
. In the long run, the frequency of messages will generally be slightly lower than the reciprocal of the specified delay
.
Fixed-delay execution is appropriate for recurring activities that require “smoothness.” In other words, it is appropriate for activities where it is more important to keep the frequency accurate in the short run than in the long run.
When using fixed-rate it will compensate the delay for a subsequent task if the previous messages were delayed too long. In such cases, the actual sending interval will differ from the interval passed to the scheduleAtFixedRate
method.
If the tasks are delayed longer than the interval
, the subsequent message will be sent immediately after the prior one. This also has the consequence that after long garbage collection pauses or other reasons when the JVM was suspended all “missed” tasks will execute when the process wakes up again. For example, scheduleAtFixedRate
with an interval of 1 second and the process is suspended for 30 seconds will result in 30 messages being sent in rapid succession to catch up. In the long run, the frequency of execution will be exactly the reciprocal of the specified interval
.
Fixed-rate execution is appropriate for recurring activities that are sensitive to absolute time or where the total time to perform a fixed number of executions is important, such as a countdown timer that ticks once every second for ten seconds.
scheduleAtFixedRate
can result in bursts of scheduled messages after long garbage collection pauses, which may in worst case cause undesired load on the system. scheduleWithFixedDelay
is often preferred.
Responding to a sharded actor
When Akka Cluster is used to shard actors you need to take into account that an actor may move or get passivated.
The normal pattern for expecting a reply is to include an ActorRef
in the message, typically a message adapter. This can be used for a sharded actor but if ctx.self
is sent and the sharded actor is moved or passivated then the reply will sent to dead letters.
An alternative is to send the entityId
in the message and have the reply sent via sharding.
Example:
- Scala
-
source
// a sharded actor that needs counter updates object CounterConsumer { sealed trait Command final case class NewCount(count: Long) extends Command val TypeKey: EntityTypeKey[Command] = EntityTypeKey[Command]("example-sharded-response") } // a sharded counter that sends responses to another sharded actor object Counter { trait Command case object Increment extends Command final case class GetValue(replyToEntityId: String) extends Command val TypeKey: EntityTypeKey[Command] = EntityTypeKey[Command]("example-sharded-counter") private def apply(): Behavior[Command] = Behaviors.setup { context => counter(ClusterSharding(context.system), 0) } private def counter(sharding: ClusterSharding, value: Long): Behavior[Command] = Behaviors.receiveMessage { case Increment => counter(sharding, value + 1) case GetValue(replyToEntityId) => val replyToEntityRef = sharding.entityRefFor(CounterConsumer.TypeKey, replyToEntityId) replyToEntityRef ! CounterConsumer.NewCount(value) Behaviors.same } }
- Java
A disadvantage is that a message adapter can’t be used so the response has to be in the protocol of the actor being responded to. Additionally the EntityTypeKey
could be included in the message if it is not known statically.
As an “alternative to the alternative”, an EntityRef
can be included in the messages. The EntityRef
transparently wraps messages in a ShardingEnvelope
and sends them via sharding. If the target sharded entity has been passivated, it will be delivered to a new incarnation of that entity; if the target sharded entity has been moved to a different cluster node, it will be routed to that new node. If using this approach, be aware that at this time, a custom serializer is required.
As with directly including the entityId
and EntityTypeKey
in the message, EntityRef
s do not support message adaptation: the response has to be in the protocol of the entity being responded to.
In some cases, it may be useful to define messages with a RecipientRef
which is a common supertype of ActorRef
and EntityRef
. At this time, serializing a RecipientRef
requires a custom serializer.