Scalaz(57)- scalaz-stream: fs2-多线程编程,fs2 concurrency

    fs2的多线程编程模式不但提供了无阻碍I/O(java nio)能力,更为并行运算提供了良好的编程工具。在进入并行运算讨论前我们先示范一下fs2 pipe2对象里的一些Stream合并功能。我们先设计两个帮助函数(helper)来跟踪运算及模拟运算环境:

  def log[A](prompt: String): Pipe[Task,A,A] = _.evalMap {a =>
    Task.delay { println(prompt + a); a}}         //> log: [A](prompt: String)fs2.Pipe[fs2.Task,A,A]

  Stream(1,2,3).through(log(">")).run.unsafeRun   //> >1
                                                  //| >2
                                                  //| >3


log是个运算跟踪函数。

 implicit val strategy = Strategy.fromFixedDaemonPool(4)
  //> strategy  : fs2.Strategy = Strategy
 implicit val scheduler = Scheduler.fromFixedDaemonPool(2)
  //> scheduler  : fs2.Scheduler = Scheduler(java.util.concurrent.ScheduledThreadPoolExecutor@16022d9d[Running, pool size = 0, active threads = 0, queued tasks = 0, completed tasks = 0])
  def randomDelay[A](max: FiniteDuration): Pipe[Task, A, A] = _.evalMap { a => {
    val delay: Task[Int] = Task.delay {
      scala.util.Random.nextInt(max.toMillis.toInt)
    }
    delay.flatMap { d => Task.now(a).schedule(d.millis) }
   }
  }      //> randomDelay: [A](max: scala.concurrent.duration.FiniteDuration)fs2.Pipe[fs2.Task,A,A]
  Stream(1,2,3).through(randomDelay(1.second)).through(log("delayed>")).run.unsafeRun
                                                  //> delayed>1
                                                  //| delayed>2
                                                  //| delayed>3


randomDelay是一个模拟任意延迟运算环境的函数。我们也可以在连接randomDelay前后进行跟踪:

Stream(1,2,3).through(log("befor delay>"))
               .through(randomDelay(1.second))
               .through(log("after delay>")).run.unsafeRun
                                                  //> befor delay>1
                                                  //| after delay>1
                                                  //| befor delay>2
                                                  //| after delay>2
                                                  //| befor delay>3
                                                  //| after delay>3


值得注意的是randomDelay并不会阻碍(block)当前运算。

下面我们来看看pipe2对象里的合并函数interleave:

val sa = Stream(1,2,3).through(randomDelay(1.second)).through(log("A>"))
        //> sa  : fs2.Stream[fs2.Task,Int] = Segment(Emit(Chunk(1, 2, 3))).flatMap(<function1>).flatMap(<function1>)
val sb = Stream(1,2,3).through(randomDelay(1.second)).through(log("B>"))
        //> sb  : fs2.Stream[fs2.Task,Int] = Segment(Emit(Chunk(1, 2, 3))).flatMap(<function1>).flatMap(<function1>)
(sa interleave sb).through(log("AB")).run.unsafeRun
                                                  //> A>1
                                                  //| B>1
                                                  //| AB>1
                                                  //| AB>1
                                                  //| A>2
                                                  //| B>2
                                                  //| AB>2
                                                  //| AB>2
                                                  //| A>3
                                                  //| B>3
                                                  //| AB>3
                                                  //| AB>3


我们看到合并后的数据发送必须等待sa,sb完成了元素发送之后。这是一种固定顺序的合并操作。merge是一种不定顺序的合并方式,我们看看它的使用示范:

(sa merge sb).through(log("AB>")).run.unsafeRun   //> B>1
                                                  //| AB>1
                                                  //| B>2
                                                  //| AB>2
                                                  //| B>3
                                                  //| AB>3
                                                  //| A>1
                                                  //| AB>1
                                                  //| A>2
                                                  //| AB>2
                                                  //| A>3
                                                  //| AB>3


我们看到merge不会同时等待sa,sb完成后再发送结果,只要其中一个完成发送就开始发送结果了。换言之merge合并基本上是跟着跑的快的那个,所以结果顺序是不规则不可确定的(nondeterministic)。那么从运算时间上来讲:interleave合并所花费时间就是确定的sa+sb,而merge则选sa,sb之间最快的时间。当然总体运算所需时间是相当的,但在merge时我们可以对发出的元素进行并行运算,能大大缩短运算时间。用merge其中一个问题是我们无法确定当前的元素是从那里发出的,我们可以用either来解决这个问题:

(sa either sb).through(log("AB>")).run.unsafeRun  //> A>1
                                                  //| AB>Left(1)
                                                  //| B>1
                                                  //| AB>Right(1)
                                                  //| A>2
                                                  //| AB>Left(2)
                                                  //| B>2
                                                  //| AB>Right(2)
                                                  //| B>3
                                                  //| AB>Right(3)
                                                  //| A>3
                                                  //| AB>Left(3)


我们通过left,right分辨数据源头。如果再增多一个Stream源头,我们还是可以用merge来合并三个Stream:

val sc = Stream.range(1,10).through(randomDelay(1.second)).through(log("C>"))
    //> sc  : fs2.Stream[fs2.Task,Int] = Segment(Emit(Chunk(()))).flatMap(<function1>).flatMap(<function1>).flatMap(<function1>)
((sa merge sb) merge sc).through(log("ABC>")).run.unsafeRun
                                                  //> B>1
                                                  //| ABC>1
                                                  //| C>1
                                                  //| ABC>1
                                                  //| A>1
                                                  //| ABC>1
                                                  //| B>2
                                                  //| ABC>2
                                                  //| A>2
                                                  //| ABC>2
                                                  //| B>3
                                                  //| ABC>3
                                                  //| C>2
                                                  //| ABC>2
                                                  //| A>3
                                                  //| ABC>3
                                                  //| C>3
                                                  //| ABC>3
                                                  //| C>4
                                                  //| ABC>4
                                                  //| C>5
                                                  //| ABC>5
                                                  //| C>6
                                                  //| ABC>6
                                                  //| C>7
                                                  //| ABC>7
                                                  //| C>8
                                                  //| ABC>8
                                                  //| C>9
                                                  //| ABC>9


如果我们无法确定数据源头数量的话,那么我们可以用以下的类型款式来表示:

Stream[Task,Stream[Task,A]]


这个类型代表的是Stream of Streams。在外部的Stream里包含了不确定数量的Streams。用具体的例子可以解释:外部的Stream代表客端数据连接(connection),内部的Stream代表每个客端读取的数据。把上面的三个Stream用这种类型来表示的话:

val streams:Stream[Task,Stream[Task,Int]] = Stream(sa,sb,sc)
     //> streams  : fs2.Stream[fs2.Task,fs2.Stream[fs2.Task,Int]] = Segment(Emit(Chunk(Segment(Emit(Chunk(1, 2, 3))).flatMap(<function1>).flatMap(<function1>),Segment(Emit(Chunk(1, 2, 3))).flatMap(<function1>).flatMap(<function1>), S
egment(Emit(Chunk(()))).flatMap(<function1>).flatMap(<function1>).flatMap(<function1>))))


现在我们不但需要对内部Stream进行运算还需要把结果打平成Stream[Task,A]。在fs2.concurrent包里就有这样一个组件(combinator):

  def join[F[_],O](maxOpen: Int)(outer: Stream[F,Stream[F,O]])(implicit F: Async[F]): Stream[F,O] = {...}


输入参数outer和运算结果类型都对得上。maxOpen代表最多并行运算数。我们可以用join运算上面合并sa,sb,sc的例子:

val ms = concurrent.join(3)(streams)              //> ms  : fs2.Stream[fs2.Task,Int] = attemptEval(Task).flatMap(<function1>).flatMap(<function1>)
ms.through(log("ABC>")).run.unsafeRun             //> C>1
                                                  //| ABC>1
                                                  //| A>1
                                                  //| ABC>1
                                                  //| C>2
                                                  //| ABC>2
                                                  //| B>1
                                                  //| ABC>1
                                                  //| C>3
                                                  //| ABC>3
                                                  //| A>2
                                                  //| ABC>2
                                                  //| B>2
                                                  //| ABC>2
                                                  //| C>4
                                                  //| ABC>4
                                                  //| A>3
                                                  //| ABC>3
                                                  //| B>3
                                                  //| ABC>3
                                                  //| C>5
                                                  //| ABC>5
                                                  //| C>6
                                                  //| ABC>6
                                                  //| C>7
                                                  //| ABC>7
                                                  //| C>8
                                                  //| ABC>8
                                                  //| C>9
                                                  //| ABC>9


结果就是我们预料的。上面提到过maxOpen是最大并行运算数。我们用另一个例子来观察:

val rangedStreams = Stream.range(0,5).map {id =>
      Stream.range(1,5).through(randomDelay(1.second)).through(log((('A'+id).toChar).toString +">")) }
      //> rangedStreams  : fs2.Stream[Nothing,fs2.Stream[fs2.Task,Int]] = Segment(Emit(Chunk(()))).flatMap(<function1>).mapChunks(<function1>)
concurrent.join(3)(rangedStreams).run.unsafeRun   //> B>1
                                                  //| A>1
                                                  //| C>1
                                                  //| B>2
                                                  //| C>2
                                                  //| A>2
                                                  //| B>3
                                                  //| C>3
                                                  //| C>4
                                                  //| D>1
                                                  //| A>3
                                                  //| A>4
                                                  //| B>4
                                                  //| E>1
                                                  //| E>2
                                                  //| E>3
                                                  //| D>2
                                                  //| D>3
                                                  //| E>4
                                                  //| D>4


可以看到一共只有三个运算过程同时存在,如:ABC, ED...

当我们的程序需要与外界程序交互时,可能会以下面的几种形式进行:

1、产生副作用的运算是同步运行的。这种情况最容易处理,因为直接可以获取结果

2、产生副作用的运算是异步的:通过调用一次callback函数来提供运算结果

3、产生副作用的运算是异步的,但结果必须通过多次调用callback函数来分批提供

下面我们就一种一种情况来分析:

1、同步运算最容易处理:我们只需要把运算包嵌在Stream.eval里就行了:

def destroyUniverse: Unit = println("BOOOOM!!!")  //> destroyUniverse: => Unit
val s = Stream.eval_(Task.delay(destroyUniverse)) ++ Stream("...move on")
    //> s  : fs2.Stream[fs2.Task,String] = append(attemptEval(Task).flatMap(<function1>).flatMap(<function1>), Segment(Emit(Chunk(()))).flatMap(<function1>))
s.runLog.unsafeRun                        //> BOOOOM!!!
                                          //| res8: Vector[String] = Vector(...move on)


2、第二种情况:fs2里的Async trait有个async是用来登记callback函数的:

trait Async[F[_]] extends Effect[F] { self =>
/**
   Create an `F[A]` from an asynchronous computation, which takes the form
   of a function with which we can register a callback. This can be used
   to translate from a callback-based API to a straightforward monadic
   version.
   */
  def async[A](register: (Either[Throwable,A] => Unit) => F[Unit]): F[A] =
    bind(ref[A]) { ref =>
    bind(register { e => runSet(ref)(e) }) { _ => get(ref) }}
...

我们用一个实际的例子来做示范,假设我们有一个callback函数readBytes:

trait Connection {
  def readBytes(onSuccess: Array[Byte] => Unit, onFailure: Throwable => Unit): Unit

这个Connection就是一个交互界面(interface)。假设它是这样实现实例化的:

val conn = new Connection {
  def readBytes(onSuccess: Array[Byte] => Unit, onFailure: Throwable => Unit): Unit = {
    Thread.sleep(1000)
    onSuccess(Array(1,2,3,4,5))
  }
}  //> conn  : demo.ws.fs2Concurrent.connection = demo.ws.fs2Concurrent$$anonfun$main$1$$anon$1@4c40b76e

我们可以用async登记(register)这个callback函数,把它变成纯代码可组合的(monadic)组件Task[Array[Byte]]:

val bytes = T.async[Array[Byte]] { (cb: Either[Throwable,Array[Byte]] => Unit) => {
   Task.delay { conn.readBytes (
     ready => cb(Right(ready)),
     fail => cb(Left(fail))
   ) }
}}             //> bytes  : fs2.Task[Array[Byte]] = Task

这样我们才能用Stream.eval来运算bytes:

Stream.eval(bytes).map(_.toList).runLog.unsafeRun //> res9: Vector[List[Byte]] = Vector(List(1, 2, 3, 4, 5))


这种只调用一次callback函数的情况也比较容易处理:当我们来不及处理数据时停止读取就是了。如果需要多次调用callback,比如外部程序也是一个Stream API:一旦数据准备好就调用一次callback进行传送。这种情况下可能出现我们的程序来不及处理收到的数据的状况。我们可以用fs2.async包提供的queue来解决这个问题:

import fs2.async
  import fs2.util.Async

  type Row = List[String]
  // defined type alias Row

  trait CSVHandle {
    def withRows(cb: Either[Throwable,Row] => Unit): Unit
  }
  // defined trait CSVHandle

  def rows[F[_]](h: CSVHandle)(implicit F: Async[F]): Stream[F,Row] =
    for {
      q <- Stream.eval(async.unboundedQueue[F,Either[Throwable,Row]])
      _ <- Stream.suspend { h.withRows { e => F.unsafeRunAsync(q.enqueue1(e))(_ => ()) }; Stream.emit(()) }
      row <- q.dequeue through pipe.rethrow
    } yield row
  // rows: [F[_]](h: CSVHandle)(implicit F: fs2.util.Async[F])fs2.Stream[F,Row]


enqueue1和dequeue在Queue trait里是这样定义的:

/**
 * Asynchronous queue interface. Operations are all nonblocking in their
 * implementations, but may be 'semantically' blocking. For instance,
 * a queue may have a bound on its size, in which case enqueuing may
 * block until there is an offsetting dequeue.
 */
trait Queue[F[_],A] {
/**
   * Enqueues one element in this `Queue`.
   * If the queue is `full` this waits until queue is empty.
   *
   * This completes after `a`  has been successfully enqueued to this `Queue`
   */
  def enqueue1(a: A): F[Unit]

/** Repeatedly call `dequeue1` forever. */
  def dequeue: Stream[F, A] = Stream.repeatEval(dequeue1)

  /** Dequeue one `A` from this queue. Completes once one is ready. */
  def dequeue1: F[A]
...


我们用enqueue1把一次callback调用存入queue。dequeue的运算结果是Stream[F,Row],所以我们用dequeue运算存在queue里的任务取出数据。

fs2提供了signal,queue,semaphore等数据类型。下面是一些使用示范:async.signal

Stream.eval(async.signalOf[Task,Int](0)).flatMap {s =>
    val monitor: Stream[Task,Nothing] =
      s.discrete.through(log("s updated>")).drain
    val data: Stream[Task,Int] =
      Stream.range(10,16).through(randomDelay(1.second))
    val writer: Stream[Task,Unit] =
      data.evalMap {d => s.set(d)}
    monitor merge writer
  }.run.unsafeRun                                 //> s updated>0
                                                  //| s updated>10
                                                  //| s updated>11
                                                  //| s updated>12
                                                  //| s updated>13
                                                  //| s updated>14
                                                  //| s updated>15


async.queue使用示范:

Stream.eval(async.boundedQueue[Task,Int](5)).flatMap {q =>
    val monitor: Stream[Task,Nothing] =
      q.dequeue.through(log("dequeued>")).drain
    val data: Stream[Task,Int] =
      Stream.range(10,16).through(randomDelay(1.second))
    val writer: Stream[Task,Unit] =
      data.to(q.enqueue)
    monitor mergeHaltBoth  writer

  }.run.unsafeRun                                 //> dequeued>10
                                                  //| dequeued>11
                                                  //| dequeued>12
                                                  //| dequeued>13
                                                  //| dequeued>14
                                                  //| dequeued>15


fs2还在time包里提供了一些定时自动产生数据的函数和类型。我们用一些代码来示范它们的用法:

 time.awakeEvery[Task](1.second)
   .through(log("time:"))
   .take(5).run.unsafeRun                         //> time:1002983266 nanoseconds
                                                  //| time:2005972864 nanoseconds
                                                  //| time:3004831159 nanoseconds
                                                  //| time:4002104307 nanoseconds
                                                  //| time:5005091850 nanoseconds


awakeEvery产生的是一个无穷数据流,所以我们用take(5)来取前5个元素。我们也可以让它运算5秒钟:

 val tick = time.awakeEvery[Task](1.second).through(log("time:"))
    //> tick  : fs2.Stream[fs2.Task,scala.concurrent.duration.FiniteDuration] = Segment(Emit(Chunk(()))).flatMap(<function1>).flatMap(<function1>).flatMap(<function1>)
 tick.run.unsafeRunFor(5.seconds)                 //> time:1005685270 nanoseconds
                                                  //| time:2004331473 nanoseconds
                                                  //| time:3005046945 nanoseconds
                                                  //| time:4002795227 nanoseconds
                                                  //| time:5002807816 nanoseconds
                                                  //| java.util.concurrent.TimeoutException


如果我们希望避免TimeoutException,可以用Task.schedule:

val tick = time.awakeEvery[Task](1.second).through(log("time:"))
   //> tick  : fs2.Stream[fs2.Task,scala.concurrent.duration.FiniteDuration] = Seg
ment(Emit(Chunk(()))).flatMap(<function1>).flatMap(<function1>).flatMap(<function1>)
 tick.interruptWhen(Stream.eval(Task.schedule(true,5.seconds)))
      .run.unsafeRun                              //> time:1004963839 nanoseconds
                                                  //| time:2005325025 nanoseconds
                                                  //| time:3005238921 nanoseconds
                                                  //| time:4004240985 nanoseconds
                                                  //| time:5001334732 nanoseconds
                                                  //| time:6003586673 nanoseconds
                                                  //| time:7004728267 nanoseconds
                                                  //| time:8004333608 nanoseconds
                                                  //| time:9003907670 nanoseconds
                                                  //| time:10002624970 nanoseconds


最直接的方法是用fs2的tim.sleep:

 (time.sleep[Task](5.seconds) ++ Stream.emit(true)).runLog.unsafeRun
                                                  //> res14: Vector[Boolean] = Vector(true)
 tick.interruptWhen(time.sleep[Task](5.seconds) ++ Stream.emit(true))
    .run.unsafeRun                                //> time:1002078506 nanoseconds
                                                  //| time:2005144318 nanoseconds
                                                  //| time:3004049135 nanoseconds
                                                  //| time:4002963861 nanoseconds
                                                  //| time:5000088103 nanoseconds



























评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值