Scalaz（47）－ scalaz-stream: 深入了解-Source

scalaz-stream库的主要设计目标是实现函数式的I/O编程（functional I/O）。这样用户就能使用功能单一的基础I/O函数组合成为功能完整的I/O程序。还有一个目标就是保证资源的安全使用（resource safety）：使用scalaz-stream编写的I/O程序能确保资源的安全使用，特别是在完成一项I/O任务后自动释放所有占用的资源包括file handle、memory等等。我们在上一篇的讨论里笼统地解释了一下scalaz-stream核心类型Process的基本情况，不过大部分时间都用在了介绍Process1这个通道类型。在这篇讨论里我们会从实际应用的角度来介绍整个scalaz-stream链条的设计原理及应用目的。我们提到过Process具有Emit/Await/Halt三个状态，而Append是一个链接stream节点的重要类型。先看看这几个类型在scalaz-stream里的定义：

case class Emit[+O](seq: Seq[O]) extends HaltEmitOrAwait[Nothing, O] with EmitOrAwait[Nothing, O]

case class Await[+F[_], A, +O](
    req: F[A]
    , rcv: (EarlyCause \/ A) => Trampoline[Process[F, O]] @uncheckedVariance
    , preempt : A => Trampoline[Process[F,Nothing]] @uncheckedVariance = (_:A) => Trampoline.delay(halt:Process[F,Nothing])
    ) extends HaltEmitOrAwait[F, O] with EmitOrAwait[F, O] 

case class Halt(cause: Cause) extends HaltEmitOrAwait[Nothing, Nothing] with HaltOrStep[Nothing, Nothing]

case class Append[+F[_], +O](
    head: HaltEmitOrAwait[F, O]
    , stack: Vector[Cause => Trampoline[Process[F, O]]] @uncheckedVariance
    ) extends Process[F, O] 

我们看到Process[F,O]被包嵌在Trampoline类型里，所以Process是通过Trampoline来实现函数结构化的，可以有效解决大量stream运算堆栈溢出问题（StackOverflowError）。撇开Trampoline等复杂的语法，以上类型可以简化成以下理论结构：

 rait Process[+F[_],+O]
 case object Cause

 case class Emit[O](out: O) extends Process[Nothing, O] 

 case class Halt(cause: Cause) extends Process[Nothing,Nothing]

 case class Await[+F[_],E,+O](
   req: F[E],
   rcv: E => Process[F,O],
   preempt: E => Process[F,Nothing] = Halt) extends Process[F,O]

 case class Append[+F[_],+O](
   head: Process[F,O],
   stack: Vector[Cause => Process[F,O]]) extends Process[F,O]  

我们来说明一下：

Process[F[_],O]：从它的类型可以推断出scalaz-stream可以在输出O类型元素的过程中进行可能含副作用的F类型运算。

Emit[O]（out: O)：发送一个O类型元素；不可能进行任何附加运算

Halt(cause: Cause)：停止发送；cause是停止的原因：End-完成发送，Err-出错终止，Kill-强行终止

Await[+F[_],E,+O]：这个是运算流的核心Process状态。先进行F运算req，得出结果E后输入函数rcv转换到下一个Process状态，完成后执行preempt这个事后清理函数。这不就是个flatMap函数结构版嘛。值得注意的是E类型是个内部类型，由F运算产生后输入rcv后就不再引用了。我们可以在preepmt函数里进行资源释放。如果我们需要构建一个运算流，看来就只有使用这个Await类型了

Append[+F[_],+O]：Append是一个Process[F,O]链接类型。首先它不但担负了元素O的传送，更重要的是它还可以把上一节点的F运算传到下一个节点。这样才能在下面节点时运行对上一个节点的事后处置函数(finalizer)。Append可以把多个节点结成一个大节点：head是第一个节点，stack是一串函数，每个函数接受上一个节点完成状态后运算出下一个节点状态

一个完整的scalaz-stream由三个类型的节点组成Source(源点)/Transducer(传换点)/Sink(终点)。节点间通过Await或者Append来链接。我们再来看看Source/Transducer/Sink的类型款式：

上游：Source       >>> Process0[O]   >>> Process[F[_],O]

中游：Transduce    >>> Process1[I,O]

下游：Sink/Channel >>> Process[F[_],O => F[Unit]], Channel >>> Process[F[_],I => F[O]]

我们可以用一个文件处理流程来描述完整scalaz-stream链条的作用：

Process[F[_],O]，用F[O]方式读取文件中的O值，这时F是有副作用的

>>> Process[I,O]，I代表从文件中读取的原始数据，O代表经过筛选、处理产生的输出数据

>>> O => F[Unit]是一个不返回结果的函数，代表对输入的O类型数据进行F运算，如把O类型数据存写入一个文件

/>> I => F[O]是个返回结果的函数，对输入I进行F运算后返回O，如把一条记录写入数据库后返回写入状态

以上流程简单描述：从文件中读出数据->加工处理读出数据->写入另一个文件。虽然从描述上看起来很简单，但我们的目的是资源安全使用：无论在任何终止情况下：正常读写、中途强行停止、出错终止，scalaz-stream都会主动关闭开启的文件、停止使用的线程、释放占用的内存等其它资源。这样看来到不是那么简单了。我们先试着分析Source/Transducer/Sink这几种类型的作用：

 import Process._
 emit()                        //> res0: scalaz.stream.Process0[Int] = Emit(Vector(0))
 emitAll(Seq(,,))            //> res1: scalaz.stream.Process0[Int] = Emit(List(1, 2, 3))
 Process(,,)                 //> res2: scalaz.stream.Process0[Int] = Emit(WrappedArray(1, 2, 3))
 Process()                      //> res3: scalaz.stream.Process0[Nothing] = Emit(List())

以上都是Process0的构建方式，也算是数据源。但它们只是代表了内存中的一串值，对我们来说没什么意义，因为我们希望从外设获取这些值，比如从文件或者数据库里读取数据，也就是说需要F运算效果。Process0[O] >>> Process[Nothing,O]，而我们需要的是Process[F,O]。那么我们这样写如何呢？

 val p: Process[Task,Int] = emitAll(Seq(,,))
    //> p  : scalaz.stream.Process[scalaz.concurrent.Task,Int] = Emit(List(1, 2, 3))

 emitAll(Seq(,,)).toSource
    //> res4: scalaz.stream.Process[scalaz.concurrent.Task,Int] = Emit(List(1, 2, 3))
                                                   

类型倒是匹配了，但表达式Emit(...)里没有任何Task的影子，这个无法满足我们对Source的需要。看来只有以下这种方式了：

 await(Task.delay{})(emit)
 //> res5: scalaz.stream.Process[scalaz.concurrent.Task,Int] = Await(scalaz.concurrent.Task@57855c9a,<function1>,<function1>)
 eval(Task.delay{})
 //> res6: scalaz.stream.Process[scalaz.concurrent.Task,Int] = Await(scalaz.concurrent.Task@63e2203c,<function1>,<function1>)

现在不但类型匹配，而且表达式里还包含了Task运算。我们通过Task.delay可以进行文件读取等带有副作用的运算，这是因为Await将会运行req:F[E] >>> Task[Int]。这正是我们需要的Source。那我们能不能用这个Source来发出一串数据呢？

 def emitSeq[A](xa: Seq[A]):Process[Task,A] =
   xa match {
     case h :: t => await(Task.delay {h})(emit) ++ emitSeq(t)
     case Nil => halt
   }                                     //> emitSeq: [A](xa: Seq[A])scalaz.stream.Process[scalaz.concurrent.Task,A]
 val es1 = emitSeq(Seq(,,))           //> es1  : scalaz.stream.Process[scalaz.concurrent.Task,Int] = Append(Await(scalaz.concurrent.Task@2d6eabae,<function1>,<function1>),Vector(<function1>))
 val es2 = emitSeq(Seq("a","b","c"))     //> es2  : scalaz.stream.Process[scalaz.concurrent.Task,String] = Append(Await(
 scalaz.concurrent.Task@4e7dc304,<function1>,<function1>),Vector(<function1>))
 es1.runLog.run                          //> res7: Vector[Int] = Vector(1, 2, 3)
 es2.runLog.run                          //> res8: Vector[String] = Vector(a, b, c)

以上示范中我们用await运算了Task，然后返回了Process[Task,?]，一个可能带副作用运算的Source。实际上我们在很多情况下都需要从外部的源头用Task来获取一些数据，通常这些数据源都对数据获取进行了异步（asynchronous）运算处理，然后通过callback方式来提供这些数据。我们可以用Task.async函数来把这些callback函数转变成Task，下一步我们只需要用Process.eval或者await就可以把这个Task升格成Process[Task,?]。我们先看个简单的例子：假如我们用scala.concurrent.Future来进行异步数据读取，可以这样把Future转换成Process：

 def getData(dbName: String): Task[String] = Task.async { cb =>
    import scala.concurrent._
    import scala.concurrent.ExecutionContext.Implicits.global
    import scala.util.{Success,Failure}
    Future { s"got data from $dbName" }.onComplete {
      case Success(a) => cb(a.right)
      case Failure(e) => cb(e.left)
    }
 }                                        //> getData: (dbName: String)scalaz.concurrent.Task[String]
 val procGetData = eval(getData("MySQL")) //> procGetData  : scalaz.stream.Process[scalaz.concurrent.Task,String] = Await(scalaz.concurrent.Task@dd3b207,<function1>,<function1>)
 procGetData.runLog.run                   //> res9: Vector[String] = Vector(got data from MySQL)

我们也可以把java的callback转变成Task：

   import com.ning.http.client._
   val asyncHttpClient = new AsyncHttpClient()     //> asyncHttpClient  : com.ning.http.client.AsyncHttpClient = com.ning.http.client.AsyncHttpClient@245b4bdc
   def get(s: String): Task[Response] = Task.async[Response] { callback =>
     asyncHttpClient.prepareGet(s).execute(
       new AsyncCompletionHandler[Unit] {
         def onCompleted(r: Response): Unit = callback(r.right)

         def onError(e: Throwable): Unit = callback(e.left)
       }
     )
   }                 //> get: (s: String)scalaz.concurrent.Task[com.ning.http.client.Response]
   val prcGet = Process.eval(get("http://sina.com"))
                     //> prcGet  : scalaz.stream.Process[scalaz.concurrent.Task,com.ning.http.client.Response] = Await(scalaz.concurrent.Task@222545dc,<function1>,<function1>)
   prcGet.run.run    //> 12:25:27.852 [New I/O worker #1] DEBUG c.n.h.c.p.n.r.NettyConnectListener -Request using non cached Channel '[id: 0x23fa1307, /192.168.200.3:50569 =>sina.com/66.102.251.33:80]':

如果直接按照scalaz Task callback的类型款式 def async（callback:(Throwable \/ Unit) => Unit)：

   def read(callback: (Throwable \/ Array[Byte]) => Unit): Unit = ???
                                  //> read: (callback: scalaz.\/[Throwable,Array[Byte]] => Unit)Unit
   val t: Task[Array[Byte]] = Task.async(read)     //> t  : scalaz.concurrent.Task[Array[Byte]] = scalaz.concurrent.Task@1a677343
   val t2: Task[Array[Byte]] = for {
     bytes <- t
     moarBytes <- t
   } yield (bytes ++ moarBytes)          //> t2  : scalaz.concurrent.Task[Array[Byte]] = scalaz.concurrent.Task@15de0b3c
   val prct2 = Process.eval(t2)          //> prct2  : scalaz.stream.Process[scalaz.concurrent.Task,Array[Byte]] = Await(scalaz.concurrent.Task@15de0b3c,<function1>,<function1>)

   def asyncRead(succ: Array[Byte] => Unit, fail: Throwable => Unit): Unit = ???
                           //> asyncRead: (succ: Array[Byte] => Unit, fail: Throwable => Unit)Unit
   val t3: Task[Array[Byte]] = Task.async { callback =>
      asyncRead(b => callback(b.right), err => callback(err.left))
   }                      //> t3  : scalaz.concurrent.Task[Array[Byte]] = scalaz.concurrent.Task@489115ef
   val t4: Task[Array[Byte]] = t3.flatMap(b => Task(b))
                          //> t4  : scalaz.concurrent.Task[Array[Byte]] = scalaz.concurrent.Task@3857f613
   val prct4 = Process.eval(t4)      //> prct4  : scalaz.stream.Process[scalaz.concurrent.Task,Array[Byte]] = Await(scalaz.concurrent.Task@3857f613,<function1>,<function1>)

我们也可以用timer来产生Process[Task,A]：

   import scala.concurrent.duration._
   implicit val scheduler = java.util.concurrent.Executors.newScheduledThreadPool()
                   //> scheduler  : java.util.concurrent.ScheduledExecutorService = java.util.concurrent.ScheduledThreadPoolExecutor@516be40f[Running, pool size = 0, active threads = 0, queued tasks = 0, completed tasks = 0]
   val fizz = time.awakeEvery(.seconds).map(_ => "fizz")
                   //> fizz  : scalaz.stream.Process[scalaz.concurrent.Task,String] = Await(scalaz.concurrent.Task@5762806e,<function1>,<function1>)
   val fizz3 = fizz.take()   //> fizz3  : scalaz.stream.Process[scalaz.concurrent.Task,String] = Append(Halt(End),Vector(<function1>))
   fizz3.runLog.run           //> res9: Vector[String] = Vector(fizz, fizz, fizz)

Queue、Top和Signal都可以作为带副作用数据源的构建器。我们先看看Queue是如何产生数据源的：

   type BigStringResult = String
   val qJobResult = async.unboundedQueue[BigStringResult]
                          //> qJobResult  : scalaz.stream.async.mutable.Queue[demo.ws.blogStream.BigStringResult] = scalaz.stream.async.mutable.Queue$$anon$1@25d250c6
   def longGet(jobnum: Int): BigStringResult = {
     Thread.sleep()
     s"Some large data sets from job#${jobnum}"
   }                      //> longGet: (jobnum: Int)demo.ws.blogStream.BigStringResult

 //  multi-tasking
     val start = System.currentTimeMillis()        //> start  : Long = 1468556250797
     Task.fork(qJobResult.enqueueOne(longGet())).unsafePerformAsync{case _ => ()}
     Task.fork(qJobResult.enqueueOne(longGet())).unsafePerformAsync{case _ => ()}
     Task.fork(qJobResult.enqueueOne(longGet())).unsafePerformAsync{case _ => ()}
     val timemill = System.currentTimeMillis() - start
                                                   //> timemill  : Long = 17
     Thread.sleep()
     qJobResult.close.run
  val bigData = {
  //multi-tasking
     val j1 = qJobResult.dequeue
     val j2 = qJobResult.dequeue
     val j3 = qJobResult.dequeue
     for {
      r1 <- j1
      r2 <- j2
      r3 <- j3
     } yield r1 + ","+ r2 + "," + r3
   }                       //> bigData  : scalaz.stream.Process[[x]scalaz.concurrent.Task[x],String] = Await(scalaz.concurrent.Task@778d1062,<function1>,<function1>)

   bigData.runLog.run      //> res9: Vector[String] = Vector(Some large data sets from job#2,Some large data sets from job#3,Some large data sets from job#1)

再看看Topic示范：

 import scala.concurrent._
   import scala.concurrent.duration._
   import scalaz.stream.async.mutable._
   import scala.concurrent.ExecutionContext.Implicits.global
   val sharedData: Topic[BigStringResult] = async.topic()
        //> sharedData  : scalaz.stream.async.mutable.Topic[demo.ws.blogStream.BigStringResult] = scalaz.stream.async.package$$anon$1@797badd3
   val subscriber = sharedData.subscribe.runLog    //> subscriber  : scalaz.concurrent.Task[Vector[demo.ws.blogStream.BigStringResult]] = scalaz.concurrent.Task@226a82c4
   val otherThread = future {
     subscriber.run // Added this here - now subscriber is really attached to the topic
   }                //> otherThread  : scala.concurrent.Future[Vector[demo.ws.blogStream.BigStringResult]] = List()
   // Need to give subscriber some time to start up.
   // I doubt you'd do this in actual code.

   // topics seem more useful for hooking up things like
   // sensors that produce a continual stream of data,

   // and where individual values can be dropped on
   // floor.
   Thread.sleep()

   sharedData.publishOne(longGet()).run // don't just call publishOne; need to run the resulting task
   sharedData.close.run // Don't just call close; need to run the resulting task

   // Need to wait for the output
   val result = Await.result(otherThread, Duration.Inf)
        //> result  : Vector[demo.ws.blogStream.BigStringResult] = Vector(Some large data sets from job#1)

以上对可能带有副作用的Source的各种产生方法提供了解释和示范。scalaz-stream的其他类型节点将在下面的讨论中深入介绍。