zl程序教程

您现在的位置是:首页 >  其他

当前栏目

Akka(19): Stream:组合数据流,组合共用-Graph modular composition详解编程语言

编程语言 详解 组合 19 stream Graph 数据流 Akka
2023-06-13 09:20:38 时间

   akka-stream的Graph是一种运算方案,它可能代表某种简单的线性数据流图如:Source/Flow/Sink,也可能是由更基础的流图组合而成相对复杂点的某种复合流图,而这个复合流图本身又可以被当作组件来组合更大的Graph。因为Graph只是对数据流运算的描述,所以它是可以被重复利用的。所以我们应该尽量地按照业务流程需要来设计构建Graph。在更高的功能层面上实现Graph的模块化(modular)。按上回讨论,Graph又可以被描述成一种黑盒子,它的入口和出口就是Shape,而内部的作用即处理步骤Stage则是用GraphStage来形容的。下面是akka-stream预设的一些基础数据流图:

compose_shapes.png

上面Source,Sink,Flow代表具备线性步骤linear-stage的流图,属于最基础的组件,可以用来构建数据处理链条。而Fan-In合并型,Fan-Out扩散型则具备多个输入或输出端口,可以用来构建更复杂的数据流图。我们可以用以上这些基础Graph来构建更复杂的复合流图,而这些复合流图又可以被重复利用去构建更复杂的复合流图。下面就是一些常见的复合流图:

compose_composites.png

注意上面的Composite Flow(from Sink and Source)可以用Flow.fromSinkAndSource函数构建:

def fromSinkAndSource[I, O](sink: Graph[SinkShape[I], _], source: Graph[SourceShape[O], _]): Flow[I, O, NotUsed] = 

 fromSinkAndSourceMat(sink, source)(Keep.none)

这个Flow从流向来说先Sink再Source是反的,形成的Flow上下游间无法协调,即Source端终结信号无法到达Sink端,因为这两端是相互独立的。我们必须用CoupledTermination对象中的fromSinkAndSource函数构建的Flow来解决这个问题:

/** 

 * Allows coupling termination (cancellation, completion, erroring) of Sinks and Sources while creating a Flow them them. 

 * Similar to `Flow.fromSinkAndSource` however that API does not connect the completion signals of the wrapped stages. 

object CoupledTerminationFlow { 

 @deprecated("Use `Flow.fromSinkAndSourceCoupledMat(..., ...)(Keep.both)` instead", "2.5.2") 

 def fromSinkAndSource[I, O, M1, M2](in: Sink[I, M1], out: Source[O, M2]): Flow[I, O, (M1, M2)] = 

 Flow.fromSinkAndSourceCoupledMat(in, out)(Keep.both) 

 

从上面图列里的Composite BidiFlow可以看出:一个复合Graph的内部可以是很复杂的,但从外面看到的只是简单的几个输入输出端口。不过Graph内部构件之间的端口必须按照功能逻辑进行正确的连接,剩下的就变成直接向外公开的界面端口了。这种机制支持了层级式的模块化组合方式,如下面的图示:

compose_nested_flow.png

最后变成:

compose_nested_flow_opaque.png

在DSL里我们可以用name( ??? )来分割模块:

val nestedFlow = 

 Flow[Int].filter(_ != 0) // an atomic processing stage 

 .map(_ - 2) // another atomic processing stage 

 .named("nestedFlow") // wraps up the Flow, and gives it a name 

val nestedSink = 

 nestedFlow.to(Sink.fold(0)(_ + _)) // wire an atomic sink to the nestedFlow 

 .named("nestedSink") // wrap it up 

// Create a RunnableGraph 

val runnableGraph = nestedSource.to(nestedSink)

在下面这个示范里我们自定义一个某种功能的流图模块:它有2个输入和3个输出。然后我们再使用这个自定义流图模块组建一个完整的闭合流图:

import akka.actor._ 

import akka.stream._ 

import akka.stream.scaladsl._ 

import scala.collection.immutable 

object GraphModules { 

 def someProcess[I, O]: I = O = i = i.asInstanceOf[O] 

 case class TwoThreeShape[I, I2, O, O2, O3]( 

 in1: Inlet[I], 

 in2: Inlet[I2], 

 out1: Outlet[O], 

 out2: Outlet[O2], 

 out3: Outlet[O3]) extends Shape { 

 override def inlets: immutable.Seq[Inlet[_]] = in1 :: in2 :: Nil 

 override def outlets: immutable.Seq[Outlet[_]] = out1 :: out2 :: out3 :: Nil 

 override def deepCopy(): Shape = TwoThreeShape( 

 in1.carbonCopy(), 

 in2.carbonCopy(), 

 out1.carbonCopy(), 

 out2.carbonCopy(), 

 out3.carbonCopy() 

//a functional module with 2 input 3 output 

 def TwoThreeGraph[I, I2, O, O2, O3] = GraphDSL.create() { implicit builder = 

 val balancer = builder.add(Balance[I](2)) 

 val flow = builder.add(Flow[I2].map(someProcess[I2, O2])) 

 TwoThreeShape(balancer.in, flow.in, balancer.out(0), balancer.out(1), flow.out) 

 val closedGraph = GraphDSL.create() {implicit builder = 

 import GraphDSL.Implicits._ 

 val inp1 = builder.add(Source(List(1,2,3))).out 

 val inp2 = builder.add(Source(List(10,20,30))).out 

 val merge = builder.add(Merge[Int](2)) 

 val mod23 = builder.add(TwoThreeGraph[Int,Int,Int,Int,Int]) 

 inp1 ~ mod23.in1 

 inp2 ~ mod23.in2 

 mod23.out1 ~ merge.in(0) 

 mod23.out2 ~ merge.in(1) 

 mod23.out3 ~ Sink.foreach(println) 

 merge ~ Sink.foreach(println) 

 ClosedShape 

object TailorGraph extends App { 

 import GraphModules._ 

 implicit val sys = ActorSystem("streamSys") 

 implicit val ec = sys.dispatcher 

 implicit val mat = ActorMaterializer() 

 RunnableGraph.fromGraph(closedGraph).run() 

 scala.io.StdIn.readLine() 

 sys.terminate() 


这个自定义的TwoThreeGraph是一个复合的流图模块,是可以重复使用的。注意这个~ 符合的使用:akka-stream只提供了对预设定Shape作为连接对象的支持如:

 def ~ [Out](junction: UniformFanInShape[T, Out])(implicit b: Builder[_]): PortOps[Out] = {...} 

 def ~ [Out](junction: UniformFanOutShape[T, Out])(implicit b: Builder[_]): PortOps[Out] = {...} 

 def ~ [Out](flow: FlowShape[T, Out])(implicit b: Builder[_]): PortOps[Out] = {...} 

 def ~ (to: Graph[SinkShape[T], _])(implicit b: Builder[_]): Unit = 

 b.addEdge(importAndGetPort(b), b.add(to).in) 

 def ~ (to: SinkShape[T])(implicit b: Builder[_]): Unit = 

 b.addEdge(importAndGetPort(b), to.in) 

...

所以对于我们自定义的TwoThreeShape就只能使用直接的端口连接了:

 def ~ [U : T](to: Inlet[U])(implicit b: Builder[_]): Unit = 

 b.addEdge(importAndGetPort(b), to)

以上的过程显示:通过akka的GraphDSL,对复合型Graph的构建可以实现形象化,大部分工作都在如何对组件之间的端口进行连接。我们再来看个较复杂复合流图的构建过程,下面是这个流图的图示:

compose_graph.png

可以说这是一个相对复杂的数据处理方案,里面甚至包括了数据流回路(feedback)。无法想象如果用纯函数数据流如scalaz-stream应该怎样去实现这么复杂的流程,也可能根本是没有解决方案的。但用akka GraphDSL可以很形象的组合这个数据流图;

 import GraphDSL.Implicits._ 

 RunnableGraph.fromGraph(GraphDSL.create() { implicit builder = 

 val A: Outlet[Int] = builder.add(Source.single(0)).out 

 val B: UniformFanOutShape[Int, Int] = builder.add(Broadcast[Int](2)) 

 val C: UniformFanInShape[Int, Int] = builder.add(Merge[Int](2)) 

 val D: FlowShape[Int, Int] = builder.add(Flow[Int].map(_ + 1)) 

 val E: UniformFanOutShape[Int, Int] = builder.add(Balance[Int](2)) 

 val F: UniformFanInShape[Int, Int] = builder.add(Merge[Int](2)) 

 val G: Inlet[Any] = builder.add(Sink.foreach(println)).in 

 C ~ F 

 A ~ B ~ C ~ F 

 B ~ D ~ E ~ F 

 E ~ G 

 ClosedShape 

 })

另一个端口连接方式的版本如下:

RunnableGraph.fromGraph(GraphDSL.create() { implicit builder = 

 val B = builder.add(Broadcast[Int](2)) 

 val C = builder.add(Merge[Int](2)) 

 val E = builder.add(Balance[Int](2)) 

 val F = builder.add(Merge[Int](2)) 

 Source.single(0) ~ B.in; B.out(0) ~ C.in(1); C.out ~ F.in(0) 

 C.in(0) ~ F.out 

 B.out(1).map(_ + 1) ~ E.in; E.out(0) ~ F.in(1) 

 E.out(1) ~ Sink.foreach(println) 

 ClosedShape 

})

如果把上面这个复杂的Graph切分成模块的话,其中一部分是这样的:

compose_graph_partial.png

这个开放数据流复合图可以用GraphDSL这样构建:
val partial = GraphDSL.create() { implicit builder = 

 val B = builder.add(Broadcast[Int](2)) 

 val C = builder.add(Merge[Int](2)) 

 val E = builder.add(Balance[Int](2)) 

 val F = builder.add(Merge[Int](2)) 

 C ~ F 

 B ~ C ~ F 

 B ~ Flow[Int].map(_ + 1) ~ E ~ F 

 FlowShape(B.in, E.out(1)) 

 }.named("partial")

// Convert the partial graph of FlowShape to a Flow to get 

// access to the fluid DSL (for example to be able to call .filter()) 

val flow = Flow.fromGraph(partial) 

// Simple way to create a graph backed Source 

val source = Source.fromGraph( GraphDSL.create() { implicit builder = 

 val merge = builder.add(Merge[Int](2)) 

 Source.single(0) ~ merge 

 Source(List(2, 3, 4)) ~ merge 

 // Exposing exactly one output port 

 SourceShape(merge.out) 

// Building a Sink with a nested Flow, using the fluid DSL 

val sink = { 

 val nestedFlow = Flow[Int].map(_ * 2).drop(10).named("nestedFlow") 

 nestedFlow.to(Sink.head) 

// Putting all together 

val closed = source.via(flow.filter(_ 1)).to(sink)
和scalaz-stream不同的还有akka-stream的运算是在actor上进行的,除了大家都能对数据流元素进行处理之外,akka-stream还可以通过actor的内部状态来维护和返回运算结果。这个运算结果在复合流图中传播的过程是可控的,如下图示:

返回运算结果是通过viaMat, toMat来实现的。简写的via,to默认选择流图左边运算产生的结果。

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

12842.html

cjava