正文
Scalaz（51）－ scalaz-stream: 资源使用安全－Resource Safety
提示：扫一扫查出行【扫一扫了解最新限行尾号】
复制提示
scalaz-stream是一个数据流处理工具库，对资源使用，包括：开启文件、连接网络、连接数据库等这些公共资源使用方面都必须确定使用过程的安全：要保证在作业终止时能进行事后处理程序（finalizer）来释放相关的文件、网络链接、数据库连接等。所谓作业终止包括正常的作业完成（End)、人工强行终止（Kill）及出现异常中断（Exception）。scalaz-stream并且保证了无论在数据产生的上游Source或者消费数据的下游Process都能在作业终止时运行上游Source的finalizer。scalaz-stream是按照下面的两种情况要求来设计它的finalizer启动程序的：
1、在数据产生源头环节可能开始占用资源，那么在这个环节的终止状态中必须保证运行事后处理程序
2、在消费数据的下游环节终止时必须能够运行由上游Process定义的事后处理程序
我们用一些例子来示范以上场景：
//数据产生源
val src = Process.emitAll(Seq("a","b","c")).toSource //> p : scalaz.stream.Process[scalaz.concurrent.Task,String] = Emit(List(a, b, c))
//指定事后处理程序
val p1 = src.onComplete{Process.suspend{println("---RUN CLEANUP---");Process.halt}}
//> p1 : scalaz.stream.Process[[x]scalaz.concurrent.Task[x],String] = Append(Em
//正常终止 //| it(List(a, b, c)),Vector(<function1>))
p1.runLog.run //> ---RUN CLEANUP---
//提前强制终止 //| res0: Vector[String] = Vector(a, b, c)
p1.take().runLog.run //> ---RUN CLEANUP---
//异常终止 //| res1: Vector[String] = Vector(a, b)
p1.map{_.toDouble}.runLog.run //> ---RUN CLEANUP---
//| java.lang.NumberFormatException: For input string: "a"
在scalaz-stream里我们用onComplete来指定一个Source的事后处理程序（finalizer）。我们可以从上面的例子里看到Source状态在正常终止、提前终止、异常终止时都运行指定给Source自身的finalizer。Process.onComplete是这样定义的：
/**
* Run `p2` after this `Process` completes normally, or in the event of an error.
* This behaves almost identically to `append`, except that `p1 append p2` will
* not run `p2` if `p1` halts with an `Error` or is killed. Any errors raised by
* `this` are reraised after `p2` completes.
*
* Note that `p2` is made into a finalizer using `asFinalizer`, so we
* can be assured it is run even when this `Process` is being killed
* by a downstream consumer.
*/
final def onComplete[F2[x] >: F[x], O2 >: O](p2: => Process[F2, O2]): Process[F2, O2] =
this.onHalt { cause => p2.asFinalizer.causedBy(cause) } /**
* When this `Process` halts, call `f` to produce the next state.
* Note that this function may be used to swallow or handle errors.
*/
final def onHalt[F2[x] >: F[x], O2 >: O](f: Cause => Process[F2, O2]): Process[F2, O2] = {
val next = (t: Cause) => Trampoline.delay(Try(f(t)))
this match {
case (append: Append[F2, O2] @unchecked) => Append(append.head, append.stack :+ next)
case emt@Emit(_) => Append(emt, Vector(next))
case awt@Await(_, _, _) => Append(awt, Vector(next))
case hlt@Halt(rsn) => Append(hlt, Vector(next))
}
} /**
* Mostly internal use function. Ensures this `Process` is run even
* when being `kill`-ed. Used to ensure resource safety in various
* combinators.
*/
final def asFinalizer: Process[F, O] = {
def mkAwait[F[_], A, O](req: F[A], cln: A => Trampoline[Process[F,Nothing]])(rcv: EarlyCause \/ A => Trampoline[Process[F, O]]) = Await(req, rcv,cln)
step match {
case Step(e@Emit(_), cont) => e onHalt {
case Kill => (halt +: cont).asFinalizer.causedBy(Kill)
case cause => (Halt(cause) +: cont).asFinalizer
}
case Step(Await(req, rcv, cln), cont) => mkAwait(req, cln) {
case -\/(Kill) => Trampoline.delay(Await(req, rcv, cln).asFinalizer.causedBy(Kill))
case x => rcv(x).map(p => (p +: cont).asFinalizer)
}
case hlt@Halt(_) => hlt
}
}
我们看到onComplete的作用是在当前Process进入终止状态时（正常或非正常）运行一个finalizer（p2.asFinalizer）。onHalt则将finalizer附加在当前状态后面。这样在当前状态为Halt时就会运行finalizer。asFinalizer保证即使是强行终止情况也会运行finalizer。那么如果下游的Process提前终止，是否会运行finalizer呢？
//下游正常终止
(p1 |> process1.filter(_ == true) |> process1.take()).runLog.run
//> ---RUN CLEANUP---
//| res3: Vector[String] = Vector()
//下游提前终止
(p1 |> process1.take()).runLog.run //> ---RUN CLEANUP---
//| res4: Vector[String] = Vector(a, b)
//隔层下游提前终止
(p1 |> process1.id.map{_.toUpperCase} |> process1.take()).runLog.run
//> ---RUN CLEANUP---
//| res5: Vector[String] = Vector(A, B)
//下游异常终止
(p1 |> process1.id.map{_.toDouble}).runLog.run //> ---RUN CLEANUP---
//| java.lang.NumberFormatException: For input string: "a"
事实证明下游在任何终止情况下都会运行上游定义的finalizer。那么scalaz-stream是怎么做到从下游运行上游定义的finalizer呢？我想答案一定会跟这个|>符号的pipe函数有关：
/**
* Feed the output of this `Process` as input of `p1`. The implementation
* will fuse the two processes, so this process will only generate
* values as they are demanded by `p1`. If `p1` signals termination, `this`
* is killed with same reason giving it an opportunity to cleanup.
*/
final def pipe[O2](p1: Process1[O, O2]): Process[F, O2] =
p1.suspendStep.flatMap({ s1 =>
s1 match {
case s@Step(awt1@Await1(rcv1), cont1) =>
val nextP1 = s.toProcess
this.step match {
case Step(awt@Await(_, _, _), cont) => awt.extend(p => (p +: cont) pipe nextP1)
case Step(Emit(os), cont) => cont.continue pipe process1.feed(os)(nextP1)
case hlt@Halt(End) => hlt pipe nextP1.disconnect(Kill).swallowKill
case hlt@Halt(rsn: EarlyCause) => hlt pipe nextP1.disconnect(rsn)
} case Step(emt@Emit(os), cont) =>
// When the pipe is killed from the outside it is killed at the beginning or after emit.
// This ensures that Kill from the outside is not swallowed.
emt onHalt {
case End => this.pipe(cont.continue)
case early => this.pipe(Halt(early) +: cont).causedBy(early)
} case Halt(rsn) => this.kill onHalt { _ => Halt(rsn) }
}
}) /** Operator alias for `pipe`. */
final def |>[O2](p2: Process1[O, O2]): Process[F, O2] = pipe(p2)
pipe函数的输入参数p1就是下游Process。当下游的p1状态是Halt(rsn)时，表示p1终结（提前或者正常），this.kill会将上游强制终结并运行上游onHalt函数。我们在上面的分析里已经知道Source的finalizer是在它的onHalt函数里运行的。这样就明确解释了为何在任何情况下都能保证finalizer的运行。
scalaz-stream在io对象里提供了一个linesR函数。我们可以用这个函数来读取文件系统里的文件：
val fileLines = io.linesR(s"/Users/TraverseUsage.scala")
//> fileLines : scalaz.stream.Process[scalaz.concurrent.Task,String] = Await(scalaz.concurrent.Task@6279cee3,<function1>,<function1>)
val lns = fileLines.onComplete(Process.eval[Task,String]{Task.delay{println("--FILE CLOSED--");""}})
//> lns : scalaz.stream.Process[[x]scalaz.concurrent.Task[x],String] = Append(Await(scalaz.concurrent.Task@6279cee3,<function1>,<function1>),Vector(<function1>)) lns.take().runLog.run //> --FILE CLOSED--
//| res6: Vector[String] = Vector(package scalaz.example, "", object TraverseUsage extends App {)
lns.map {_.toDouble}.runLog.run //> --FILE CLOSED--
//| java.lang.NumberFormatException: empty String caused by: java.lang.NumberFormatException: For input string: "package scalaz.example"
我们看到这个文件的使用是安全的，因为在任何终结情况下都会自动关闭打开的文件。实际上linesR打开文件后已经指定了释放文件的方式，我们看看下面的源码：
/**
* Creates a `Process[Task,String]` from the lines of a file, using
* the `iteratorR` combinator to ensure the file is closed
* when processing the stream of lines is finished.
*/
def linesR(filename: String)(implicit codec: Codec): Process[Task,String] =
linesR(Source.fromFile(filename)(codec)) /**
* Creates a `Process[Task,String]` from the lines of the `InputStream`,
* using the `iteratorR` combinator to ensure the `InputStream` is closed
* when processing the stream of lines is finished.
*/
def linesR(in: => InputStream)(implicit codec: Codec): Process[Task,String] =
linesR(Source.fromInputStream(in)(codec)) /**
* Creates a `Process[Task,String]` from the lines of the `Source`,
* using the `iteratorR` combinator to ensure the `Source` is closed
* when processing the stream of lines is finished.
*/
def linesR(src: => Source): Process[Task,String] = {
iteratorR(Task.delay(src))(src => Task.delay(src.close()))(r => Task.delay(r.getLines()))
}
这个iteratorR就已经指定了finalizer：src=>Task.delay(src.close())：
/**
* Create a Process from an iterator that is tied to some resource,
* `R` (like a file handle) that we want to ensure is released.
* See `linesR` for an example use.
* @param req acquires the resource
* @param release releases the resource
* @param mkIterator creates the iterator from the resource
* @tparam R is the resource
* @tparam O is a value in the iterator
* @return
*/
def iteratorR[R, O](req: Task[R])(
release: R => Task[Unit])(
mkIterator: R => Task[Iterator[O]]): Process[Task, O] = {
bracket[Task, R, O](req)(r => Process.eval_(release(r)))(r => iterator(mkIterator(r)) )
}
iteratorR提供了req,mkIterator,release三个输入参数，分别是开启文件，读取数据及释放文件的方法。我们也可以直接用iteratorR来示范上面的文件数据读取例子：
val iterLines =
io.iteratorR(Task.delay{Source.fromFile(s"/Users/TraverseUsage.scala")})(
src => Task.delay{src.close()})(
r => Task.delay{r.getLines()}) //> iterLines : scalaz.stream.Process[scalaz.concurrent.Task,String] = Await(scalaz.concurrent.Task@1a0dcaa,<function1>,<function1>)
iterLines.take().runLog.run //> res7: Vector[String] = Vector(package scalaz.example, "", object TraverseUsage extends App {, " import scalaz._", "")
这样来说将来我们可以用iteratorR来使用数据库，因为我们可以在这里指定数据库的连接、读写及关闭释放的具体方法。
实际运行finalizer的是这个bracket函数：
/**
* Resource and preemption safe `await` constructor.
*
* Use this combinator, when acquiring resources. This build a process that when run
* evaluates `req`, and then runs `rcv`. Once `rcv` is completed, fails, or is interrupted, it will run `release`
*
* When the acquisition (`req`) is interrupted, neither `release` or `rcv` is run, however when the req was interrupted after
* resource in `req` was acquired then, the `release` is run.
*
* If,the acquisition fails, use `bracket(req)(onPreempt)(rcv).onFailure(err => ???)` code to recover from the
* failure eventually.
*
*/
def bracket[F[_], A, O](req: F[A])(release: A => Process[F, Nothing])(rcv: A => Process[F, O]): Process[F, O] = {
Await(req,
{ (r: EarlyCause \/ A) => Trampoline.delay(Try(r.fold(Halt.apply, a => rcv(a) onComplete release(a) ))) },
{ a: A => Trampoline.delay(release(a)) })
}
bracket是个对数据进行逐行读写操作的函数。我们看到无论req的运算结果是成功a或失败r，release(a)都得以运行。
