渔民旁的松树 2020-05-26
akka 2.6.x正式发布以来已经有好一段时间了。核心变化是typed-actor的正式启用,当然persistence,cluster等模块也有较大变化。一开始从名称估摸就是把传统any类型的消息改成强类型消息,所以想拖一段时间看看到底能对我们现有基于akka-classic的应用软件有什么深层次的影响。不过最近考虑的一些系统架构逼的我不得不立即开始akka-typed的调研,也就是说akka-classic已经无法或者很困难去实现新的系统架构,且听我道来:最近在考虑一个微服务中台。作为后台数据服务调用的唯一入口,平台应该是个分布式软件,那么采用akka-cluster目前是唯一的选择,毕竟前期搞过很多基于akka-cluster的应用软件。但是,akka-cluster-sharding只能支持一种entity actor。毕竟,由于akka-classic的消息是没有类型的,只能在收到消息后再通过类型模式匹配的方式确定应该运行的代码。所以,这个actor必须包括所有的业务逻辑处理运算。也就是说对于一个大型应用来说这就是一块巨型代码。还有,如果涉及到维护actor状态的话,比如persistenceActor,或者综合类型业务运算,那么又需要多少种类的数据结构,又怎样去维护、管理这些结构呢?对我来说这基本上是mission-impossible。实际上logom应该正符合这个中台的要求:cluster-sharding, CQRS... 抱着一种好奇的心态了解了一下lagom源码,忽然恍然大悟:这个东西是基于akka-typed的!想想看也是:如果我们可以把actor和消息类型绑在一起,那么我们就可以通过消息类型对应到某种actor。也就是说基于akka-typed,我们可以把综合性的业务划分成多个actor模块,然后我们可以指定那种actor做那些事情。当然,经过了功能细分,actor的设计也简单了许多。现在这个新的中台可以实现前台应用直接调用对应的actor处理业务了。不用多想了,这注定就是akka应用的将来,还等什么呢?
先从一个最简单的hello程序开始吧:基本上是两个actor相互交换消息。先用第一个来示范标准的actor构建过程:
object HelloActor { sealed trait Request case class Greeting(fromWhom: String, replyTo: ActorRef[Greeter.Greeted]) extends Request def apply(): Behavior[Greeting] = { Behaviors.receive { (ctx, greeter) => ctx.log.info("receive greeting from {}", greeter.fromWhom) greeter.replyTo ! Greeter.Greeted(s"hello ${greeter.fromWhom}!") Behaviors.same } } }
akka-typed的actor构建是通过定义它的Behavior行为实现的。特别的是类型参数Behavior[Greeting],代表这个actor只处理Greeting类型的消息,因而是个typed-actor。akka-typed已经不支持sender()了,在消息里自带,如Greeting.replyTo。Behavior定义是通过工厂模式Behaviors实现的,看看Behaviors的定义:
/** * Factories for [[akka.actor.typed.Behavior]]. */ object Behaviors { def setup[T](factory: ActorContext[T] => Behavior[T]): Behavior[T] def withStash[T](capacity: Int)(factory: StashBuffer[T] => Behavior[T]): Behavior[T] def same[T]: Behavior[T] def unhandled[T]: Behavior[T] def stopped[T]: Behavior[T] def stopped[T](postStop: () => Unit): Behavior[T] def empty[T]: Behavior[T] def ignore[T]: Behavior[T] def receive[T](onMessage: (ActorContext[T], T) => Behavior[T]): Receive[T] def receiveMessage[T](onMessage: T => Behavior[T]): Receive[T] def receivePartial[T](onMessage: PartialFunction[(ActorContext[T], T), Behavior[T]]): Receive[T] def receiveMessagePartial[T](onMessage: PartialFunction[T, Behavior[T]]): Receive[T] def receiveSignal[T](handler: PartialFunction[(ActorContext[T], Signal), Behavior[T]]): Behavior[T] def supervise[T](wrapped: Behavior[T]): Supervise[T] def withTimers[T](factory: TimerScheduler[T] => Behavior[T]): Behavior[T] ... }
上面的构建函数除返回Behavior[T]外还有Receive[T]和Supervise[T],这两个类型是什么?它们还是Behavior[T]:
trait Receive[T] extends Behavior[T] { def receiveSignal(onSignal: PartialFunction[(ActorContext[T], Signal), Behavior[T]]): Behavior[T] } def supervise[T](wrapped: Behavior[T]): Supervise[T] = new Supervise[T](wrapped) private final val ThrowableClassTag = ClassTag(classOf[Throwable]) final class Supervise[T] private[akka] (val wrapped: Behavior[T]) extends AnyVal { /** Specify the [[SupervisorStrategy]] to be invoked when the wrapped behavior throws. */ def onFailure[Thr <: Throwable: ClassTag](strategy: SupervisorStrategy): Behavior[T] = { val tag = classTag[Thr] val effectiveTag = if (tag == ClassTag.Nothing) ThrowableClassTag else tag Supervisor(Behavior.validateAsInitial(wrapped), strategy)(effectiveTag) } }
注意,Supervise.onFailure返回了Behavior[T]。
helloActor的Behavior是通过Behaviors.receive构建的。还可以用setup,receiveMessage来构建。注意:构建函数的入参数也是Behavior[T],所以这些构造器可以一层层嵌套着使用。setup,receive为函数内层提供了ActorContext, withTimers提供TimerScheduler[T]。那么我可以把HelloActor的功能再完善点,加个监管策略SupervisorStrategy:
object HelloActor { sealed trait Request case class Greeting(fromWhom: String, replyTo: ActorRef[Greeter.Greeted]) extends Request def apply(): Behavior[Greeting] = { Behaviors.supervise( Behaviors.receive[Greeting] { (ctx, greeter) => ctx.log.info("receive greeting from {}", greeter.fromWhom) greeter.replyTo ! Greeter.Greeted(s"hello ${greeter.fromWhom}!") Behaviors.same } ).onFailure(SupervisorStrategy.restartWithBackoff(10.seconds, 1.minute, 0.20)) } }
在akka-typed里,actor监管已经从父辈转到自身。再就是增加了BackOff-SupervisorStrategy,不需要独立的BackOffSupervisor actor了。
再看看另一个Greeter:
object Greeter { sealed trait Response case class Greeted(hello: String) extends Response def apply(): Behavior[Greeted] = { Behaviors.setup ( ctx => Behaviors.receiveMessage { message => ctx.log.info(message.hello) Behaviors.same } ) } }
这个跟HelloActor没什么不同,不过用了setup,receiveMessage套装。值得注意的是Greeter负责处理Greeted消息,这是一个不带sender ActorRef的类型,意味着处理这类消息后不需要答复消息发送者。
然后还需要一个actor来构建上面两个actor实例,启动对话:
object GreetStarter { sealed trait Command case class SayHiTo(whom: String) extends Command case class RepeatedGreeting(whom: String, interval: FiniteDuration) extends Command def apply(): Behavior[Command] = { Behaviors.setup[Command] { ctx => val helloActor = ctx.spawn(HelloActor(), "hello-actor") val greeter = ctx.spawn(Greeter(), "greeter") Behaviors.withTimers { timer => new GreetStarter( helloActor,greeter,ctx,timer) .repeatGreeting(1,3) } } } } class GreetStarter private ( helloActor: ActorRef[HelloActor.Greeting], greeter: ActorRef[Greeter.Greeted], ctx: ActorContext[GreetStarter.Command], timer: TimerScheduler[GreetStarter.Command]){ import GreetStarter._ private def repeatGreeting(count: Int, max: Int): Behavior[Command] = Behaviors.receiveMessage { msg => msg match { case RepeatedGreeting(whom, interval) => ctx.log.info2("start greeting to {} with interval {}", whom, interval) timer.startSingleTimer(SayHiTo(whom), interval) Behaviors.same case SayHiTo(whom) => ctx.log.info2("{}th time greeting to {}",count,whom) if (max == count) Behaviors.stopped else { helloActor ! HelloActor.Greeting(whom, greeter) repeatGreeting(count + 1, max) } } } }
上面这个例子有点复杂,逻辑也有些问题,主要是为了示范一种函数式actor构建模式及actor状态转换虚构出来的。akka-typed已经不再支持become方法了。
最后,需要一个相当于main这么一个顶层的程序:
def main(args: Array[String]) { val man: ActorSystem[GreetStarter.Command] = ActorSystem(GreetStarter(), "greetDemo") man ! GreetStarter.RepeatedGreeting("Tiger",5.seconds) man ! GreetStarter.RepeatedGreeting("Peter",5.seconds) man ! GreetStarter.RepeatedGreeting("Susanna",5.seconds) }
akka-classic的顶级actor,即: /users是由系统默认创建的。akka-typed需要用户提供这个顶层actor。这个是在ActorSystem的第一个参数指定的。我们再看看akka-typed的ActorSystem的构建函数:
object ActorSystem { /** * Scala API: Create an ActorSystem */ def apply[T](guardianBehavior: Behavior[T], name: String): ActorSystem[T] = createInternal(name, guardianBehavior, Props.empty, ActorSystemSetup.create(BootstrapSetup())) /** * Scala API: Create an ActorSystem */ def apply[T](guardianBehavior: Behavior[T], name: String, config: Config): ActorSystem[T] = createInternal(name, guardianBehavior, Props.empty, ActorSystemSetup.create(BootstrapSetup(config))) /** * Scala API: Create an ActorSystem */ def apply[T](guardianBehavior: Behavior[T], name: String, config: Config, guardianProps: Props): ActorSystem[T] = createInternal(name, guardianBehavior, guardianProps, ActorSystemSetup.create(BootstrapSetup(config))) ... }
其中一个apply与akka-classic的ActorSystem构建方式很相似:
def main(args: Array[String]) { val config = ConfigFactory.load("application.conf") val man: ActorSystem[GreetStarter.Command] = ActorSystem(GreetStarter(), "greetDemo",config) man ! GreetStarter.RepeatedGreeting("Tiger",5.seconds) man ! GreetStarter.RepeatedGreeting("Peter",5.seconds) man ! GreetStarter.RepeatedGreeting("Susanna",5.seconds) }
下面是本次讨论的完整源代码:
build.sbt
name := "learn-akka-typed" version := "0.1" scalaVersion := "2.13.2" lazy val akkaVersion = "2.6.5" libraryDependencies ++= Seq( "com.typesafe.akka" %% "akka-actor-typed" % akkaVersion, "ch.qos.logback" % "logback-classic" % "1.2.3" ) fork in Test := true
Lesson01.scala
import akka.actor.typed._ import scaladsl._ import scala.concurrent.duration._ import com.typesafe.config._ object Lesson01 { object HelloActor { sealed trait Request case class Greeting(fromWhom: String, replyTo: ActorRef[Greeter.Greeted]) extends Request def apply(): Behavior[Greeting] = { Behaviors.supervise( Behaviors.receive[Greeting] { (ctx, greeter) => ctx.log.info("receive greeting from {}", greeter.fromWhom) greeter.replyTo ! Greeter.Greeted(s"hello ${greeter.fromWhom}!") Behaviors.same } ).onFailure(SupervisorStrategy.restartWithBackoff(10.seconds, 1.minute, 0.20)) } } object Greeter { sealed trait Response case class Greeted(hello: String) extends Response def apply(): Behavior[Greeted] = { Behaviors.setup ( ctx => Behaviors.receiveMessage { message => ctx.log.info(message.hello) Behaviors.same } ) } } object GreetStarter { sealed trait Command case class SayHiTo(whom: String) extends Command case class RepeatedGreeting(whom: String, interval: FiniteDuration) extends Command def apply(): Behavior[Command] = { Behaviors.setup[Command] { ctx => val helloActor = ctx.spawn(HelloActor(), "hello-actor") val greeter = ctx.spawn(Greeter(), "greeter") Behaviors.withTimers { timer => new GreetStarter( helloActor,greeter,ctx,timer) .repeatGreeting(1,3) } } } } class GreetStarter private ( helloActor: ActorRef[HelloActor.Greeting], greeter: ActorRef[Greeter.Greeted], ctx: ActorContext[GreetStarter.Command], timer: TimerScheduler[GreetStarter.Command]){ import GreetStarter._ private def repeatGreeting(count: Int, max: Int): Behavior[Command] = Behaviors.receiveMessage { msg => msg match { case RepeatedGreeting(whom, interval) => ctx.log.info2("start greeting to {} with interval {}", whom, interval) timer.startSingleTimer(SayHiTo(whom), interval) Behaviors.same case SayHiTo(whom) => ctx.log.info2("{}th time greeting to {}",count,whom) if (max == count) Behaviors.stopped else { helloActor ! HelloActor.Greeting(whom, greeter) repeatGreeting(count + 1, max) } } } } def main(args: Array[String]) { val config = ConfigFactory.load("application.conf") val man: ActorSystem[GreetStarter.Command] = ActorSystem(GreetStarter(), "greetDemo",config) man ! GreetStarter.RepeatedGreeting("Tiger",5.seconds) man ! GreetStarter.RepeatedGreeting("Peter",5.seconds) man ! GreetStarter.RepeatedGreeting("Susanna",5.seconds) } }