I'm writing scala <-> java interop wrappers for Futures and I don't know the Right Way to implement scala.concurrent.Future.onComplete (http://www.scala-lang.org/api/current/index.html#scala.concurrent.Future). This probably works:
def onComplete[U](func: Try[T] => U)(implicit executor: ExecutionContext): Unit = {
executor.execute(new Runnable {
#tailrec
def run = value match {
case Some(t) => func(t)
case None => { Thread.sleep(100); run }
}
})
}
but Asynchronous IO in Scala with futures suggests that when I have to block I should pass the relevant part of the code to scala.concurrent.blocking to let the ExecutionContext know what's up. The problem is that when I surround the value match{...} with blocking {} it's no longer a tail call.
What's the proverbial right way to do this?
Edit: for completeness here is the entire wrapping class:
class JavaFutureWrapper[T](val jf: java.util.concurrent.Future[T]) extends scala.concurrent.Future[T] {
def isCompleted = jf.isDone
def result(atMost: Duration)(implicit permit: CanAwait): T =
atMost match { case Duration(timeout, units) => jf.get(timeout, units) }
def onComplete[U](func: Try[T] => U)(implicit executor: ExecutionContext): Unit = {
executor.execute(new Runnable {
#tailrec
def run = value match {
case Some(t) => func(t)
case None => { Thread.sleep(100); run }
}
})
}
def ready(atMost: Duration)(implicit permit: CanAwait): this.type = atMost match {
case Duration(timeout, units) => {
jf.get(timeout, units)
this
}
}
def value: Option[Try[T]] = (jf.isCancelled, jf.isDone) match {
case (true, _) => Some(Failure(new Exception("Execution was cancelled!")))
case (_, true) => Some(Success(jf.get))
case _ => None
}
}
I would just wait for the Java future to complete:
import scala.util.{Try, Success, Failure}
import scala.concurrent._
import java.util.concurrent.TimeUnit
class JavaFutureWrapper[T](val jf: java.util.concurrent.Future[T])
extends scala.concurrent.Future[T] {
...
def onComplete[U](func: Try[T] => U)(implicit executor: ExecutionContext): Unit =
executor.execute(new Runnable {
def run: Unit = {
val result = Try(blocking(jf.get(Long.MaxValue, TimeUnit.MILLISECONDS)))
func(result)
}
})
...
}
Hmm, my edit to 0__ 's answer didn't get approved, so for the sake of future readers, here's the solution I'm going with (which is simplified from 0__'s)
def onComplete[U](func: Try[T] => U)(implicit executor: ExecutionContext): Unit = {
executor.execute(new Runnable {
def run = func(Try( blocking { jf.get } ))
})
}
Related
I'm trying to implement an application that controls a camera. Camera commands are represented as a stream of CameraAction objects:
sealed trait CameraMessage
case object Record(recordId: String) extends CameraMessage
case object Stop extends CameraMessage
...
val s = Stream[F, CameraMessage]
Let's say I have a test stream that emits "Record" and emits "Stop" 20 seconds later, after another 20 seconds another "Record" message is emitted and so on, the input stream is infinite.
Then the app consumes "Record" it should create an instance of GStreamer pipeline (i.e. it is an effect) and "run" it, on "Stop" it 'stops' the pipeline and closes it. Then on subsequent "Record" the pattern is repeated with new GStreamer pipeline.
The problem is that I need to pass an instance of impure, mutable object between handles of stream events.
FS2 documentation suggest to use chunks to make a stream stateful, so I tried
def record(gStreamerPipeline: String, fileName: String)
(implicit sync: Sync[F]): F[Pipeline] =
{
... create and open pipeline ...
}
def stopRecording(pipe: Pipeline)(implicit sync: Sync[F]): F[Unit] = {
... stop pipeline, release resources ...
}
def effectPipe(pipelineDef: String)(implicit L: Logger[F]):
Pipe[F, CameraMessage, F[Unit]] = {
type CameraSessionHandle = Pipeline
type CameraStream = Stream[F, CameraSessionHandle]
s: Stream[F, CameraMessage] =>
s.scanChunks(Stream[F, CameraSessionHandle]()) {
case (s: CameraStream, c: Chunk[CameraMessage]) =>
c.last match {
case Some(Record(fileName)) =>
(Stream.bracket(record(pipelineDef, fileName))(p => stopRecording(p)), Chunk.empty)
case Some(StopRecording) =>
(Stream.empty, Chunk(s.compile.drain))
case _ =>
(s, Chunk.empty)
}
}
}
The problem with this code that actual recording does not happen on 'Record' event but rather then the effect of the whole chunk is evaluated, i.e. when 'StopRecording' message arrives the camera is turned on and then immediately turned off again.
How can I pass a "state" without chunking? Or is there any other way to achieve the result I need?
This may be similar to
FS2 Stream with StateT[IO, _, _], periodically dumping state
but the difference is that the state in my case is not a pure data structure but a resource.
I eventually was able so solve it using Mutable Reference pattern as described in https://typelevel.org/blog/2018/06/07/shared-state-in-fp.html
Here is the code:
import cats.effect._
import cats.syntax.all._
import fs2.Stream
import scala.concurrent.{ExecutionContext, ExecutionContextExecutor}
import scala.language.higherKinds
class FRef[F[_], T](implicit sync: Sync[F]) {
private var state: T = _
def set(n: T): F[Unit] = sync.delay(this.state = n)
def get: F[T] = sync.pure(state)
}
object FRef {
def apply[F[_], T](implicit sync: Sync[F]): F[FRef[F, T]] = sync.delay { new FRef() }
}
class CameraImpl(id: String) extends Camera {
override def record(): Unit = {
println(s"Recording $id")
}
override def stop(): Unit = {
println(s"Stopping $id")
}
override def free(): Unit = {
Thread.sleep(500)
println(s"Freeing $id")
}
}
object Camera {
def apply(id: String) = new CameraImpl(id)
}
trait Camera {
def record(): Unit
def stop(): Unit
def free(): Unit
}
sealed trait CameraMessage
case class Record(recordId: String) extends CameraMessage
case object StopRecording extends CameraMessage
class Streamer[F[_]](implicit sync: Sync[F]) {
def record(id: String): F[Camera] = {
sync.delay {
val r = Camera(id)
r.record()
r
}
}
def stopRecording(pipe: Camera): F[Unit] = {
sync.delay {
pipe.stop()
pipe.free()
}
}
def effectPipe(state: FRef[F, Option[Camera]])(
implicit sync: Sync[F]): Stream[F, CameraMessage] => Stream[F, Unit] = {
type CameraStream = Stream[F, Camera]
s: Stream[F, CameraMessage] =>
s.evalMap {
case Record(fileName) =>
for {
r <- record(fileName)
_ <- state.set(Some(r))
} yield ()
case StopRecording =>
for {
s <- state.get
_ <- stopRecording(s.get)
_ <- state.set(None)
} yield ()
}
}
}
object FS2Problem extends IOApp {
import scala.concurrent.duration._
override def run(args: List[String]): IO[ExitCode] = {
implicit val ec: ExecutionContextExecutor = ExecutionContext.global
val streamer = new Streamer[IO]
val s = Stream.awakeEvery[IO](5.seconds).take(10).zipWithIndex.map {
case (_, idx) =>
idx % 2 match {
case 0 =>
Record(s"Record $idx")
case _ =>
StopRecording
}
}
val ss = for {
streamerState <- Stream.eval(FRef[IO, Option[Camera]])
s <- s.through(streamer.effectPipe(streamerState))
} yield ()
ss.compile.drain.map(_ => ExitCode.Success)
}
}
I have two methods:
def getNextJob: Future[Option[Job]]
def process(job: Job): Future[Unit]
I would like to process all Jobs until there are no jobs remaining.
I can do this with Await e.g.
private def process()(implicit ctx: ExecutionContext): Future[Unit] = {
var job: Option[Job] = Await.result(service.getNextJob, FiniteDuration(2, TimeUnit.SECONDS))
while(job.isDefined) {
Await.result(process(job.get), FiniteDuration(2, TimeUnit.SECONDS))
job = Await.result(service.getNextJob, FiniteDuration(2, TimeUnit.SECONDS))
}
Future.successful()
}
But this is ugly and doesn't use Futures properly. Is there a way I could chain the futures somehow to replace this?
def go()(implicit ctx: ExecutionContext): Future[Unit] =
getNextJob.flatMap { maybeJob ⇒
if(maybeJob.isDefined) process(maybeJob.get).flatMap(_ ⇒ go())
else Future.unit
}
Note: It is not tail recursive.
def processAll()(implicit ec: ExecutionContext): Future[Unit] =
getNextJob.flatMap {
case Some(job) => process(job).flatMap(_ => processAll())
case None => Future.unit
}
To process them all possibly concurrently:
def processAll()(implicit ec: ExecutionContext): Future[Unit] =
getNextJob.flatMap {
case Some(job) => process(job).zipWith(processAll())((_,_) => ())
case None => Future.unit
}
In my Play application, I service my requests usings cats-effect's IO, instead of Future in the controller, like this (super-simplified):
def handleServiceResult(serviceResult: ServiceResult): Result = ...
def serviceMyRequest(request: Request): IO[ServiceResult] = ...
def myAction = Action { request =>
handleServiceResult(
serviceMyRequest(request).unsafeRunSync()
)
}
Requests are then processed (asynchronously) on Play's default thread pool. Now, I want to implement multiple thread pools to handle different sorts of requests. Were I using Futures, I could do this:
val myCustomExecutionContext: ExecutionContext = ...
def serviceMyRequest(request: Request): Future[ServiceResult] = ...
def myAction = Action.async { request =>
Future(serviceMyRequest(request))(myCustomExecutionContext)
.map(handleServiceResult)(defaultExecutionContext)
}
But I'm not using Futures, I'm using IO, and I'm not sure about the right way to go about implementing it. This looks promising, but seems a bit clunky:
def serviceMyRequest(request: Request): IO[ServiceResult] = ...
def myAction = Action { request =>
val ioServiceResult = for {
_ <- IO.shift(myCustomExecutionContext)
serviceResult <- serviceMyRequest(request)
_ <- IO.shift(defaultExecutionContext)
} yield {
serviceResult
}
handleServiceResult(ioServiceResult.unsafeRunSync())
}
Is this the right way to implement it? Is there a best practice here? Am I screwing up badly? Thanks.
Ok, so since this doesn't seem to be well-trodden ground, this is what I ended up implementing:
trait PlayIO { self: BaseControllerHelpers =>
implicit class IOActionBuilder[A](actionBuilder: ActionBuilder[Request, A]) {
def io(block: Request[A] => IO[Result]): Action[A] = {
actionBuilder.apply(block.andThen(_.unsafeRunSync()))
}
def io(executionContext: ExecutionContext)(block: Request[A] => IO[Result]): Action[A] = {
val shiftedBlock = block.andThen(IO.shift(executionContext) *> _ <* IO.shift(defaultExecutionContext))
actionBuilder.apply(shiftedBlock.andThen(_.unsafeRunSync()))
}
}
}
Then (using the framework from the question) if I mix PlayIO into the controller, I can do this,
val myCustomExecutionContext: ExecutionContext = ...
def handleServiceResult(serviceResult: ServiceResult): Result = ...
def serviceMyRequest(request: Request): IO[ServiceResult] = ...
def myAction = Action.io(myCustomExecutionContext) { request =>
serviceMyRequest(request).map(handleServiceResult)
}
such that I execute the action's code block on myCustomExecutionContext and then, once complete, thread-shift back to Play's default execution context.
Update:
This is a bit more flexible:
trait PlayIO { self: BaseControllerHelpers =>
implicit class IOActionBuilder[R[_], A](actionBuilder: ActionBuilder[R, A]) {
def io(block: R[A] => IO[Result]): Action[A] = {
actionBuilder.apply(block.andThen(_.unsafeRunSync()))
}
def io(executionContext: ExecutionContext)(block: R[A] => IO[Result]): Action[A] = {
if (executionContext == defaultExecutionContext) io(block) else {
val shiftedBlock = block.andThen(IO.shift(executionContext) *> _ <* IO.shift(defaultExecutionContext))
io(shiftedBlock)
}
}
}
}
Update2:
Per the comment above, this will ensure we always shift back to the default thread pool:
trait PlayIO { self: BaseControllerHelpers =>
implicit class IOActionBuilder[R[_], A](actionBuilder: ActionBuilder[R, A]) {
def io(block: R[A] => IO[Result]): Action[A] = {
actionBuilder.apply(block.andThen(_.unsafeRunSync()))
}
def io(executionContext: ExecutionContext)(block: R[A] => IO[Result]): Action[A] = {
if (executionContext == defaultExecutionContext) io(block) else {
val shiftedBlock = block.andThen { ioResult =>
IO.shift(executionContext).bracket(_ => ioResult)(_ => IO.shift(defaultExecutionContext))
}
io(shiftedBlock)
}
}
}
}
My old code looks something like below, where all db calls blocking.
I need help converting this over to using Futures.
def getUserPoints(username: String): Option[Long]
db.getUserPoints(username) match {
case Some(userPoints) => Some(userPoints.total)
case None => {
if (db.getSomething("abc").isEmpty) {
db.somethingElse("asdf") match {
case Some(pointId) => {
db.setPoints(pointId, username)
db.findPointsForUser(username)
}
case _ => None
}
} else {
db.findPointsForUser(username)
}
}
}
}
My new API is below where I am returning Futures.
db.getUserPoints(username: String): Future[Option[UserPoints]]
db.getSomething(s: String): Future[Option[Long]]
db.setPoints(pointId, username): Future[Unit]
db.findPointsForUser(username): Future[Option[Long]]
How can I go about converting the above to use my new API that uses futures.
I tried using a for-compr but started to get wierd errors like Future[Nothing].
var userPointsFut: Future[Long] = for {
userPointsOpt <- db.getUserPoints(username)
userPoints <- userPointsOpt
} yield userPoints.total
But it gets a bit tricky with all the branching and if clauses and trying to convert it over to futures.
I would argue that the first issue with this design is that the port of the blocking call to a Future should not wrap the Option type:
The blocking call:
def giveMeSomethingBlocking(for:Id): Option[T]
Should become:
def giveMeSomethingBlocking(for:Id): Future[T]
And not:
def giveMeSomethingBlocking(for:Id): Future[Option[T]]
The blocking call give either a value Some(value) or None, the non-blocking Future version gives either a Success(value) or Failure(exception) which fully preserves the Option semantics in a non-blocking fashion.
With that in mind, we can model the process in question using combinators on Future. Let's see how:
First, lets refactor the API to something we can work with:
type UserPoints = Long
object db {
def getUserPoints(username: String): Future[UserPoints] = ???
def getSomething(s: String): Future[UserPoints] = ???
def setPoints(pointId:UserPoints, username: String): Future[Unit] = ???
def findPointsForUser(username: String): Future[UserPoints] = ???
}
class PointsNotFound extends Exception("bonk")
class StuffNotFound extends Exception("sthing not found")
Then, the process would look like:
def getUserPoints(username:String): Future[UserPoints] = {
db.getUserPoints(username)
.map(userPoints => userPoints /*.total*/)
.recoverWith{
case ex:PointsNotFound =>
(for {
sthingElse <- db.getSomething("abc")
_ <- db.setPoints(sthingElse, username)
points <- db.findPointsForUser(username)
} yield (points))
.recoverWith{
case ex: StuffNotFound => db.findPointsForUser(username)
}
}
}
Which type-checks correctly.
Edit
Given that the API is set in stone, a way to deal with nested monadic types is to define a MonadTransformer. In simple words, let's make Future[Option[T]] a new monad, let's call it FutureO that can be composed with other of its kind. [1]
case class FutureO[+A](future: Future[Option[A]]) {
def flatMap[B](f: A => FutureO[B])(implicit ec: ExecutionContext): FutureO[B] = {
val newFuture = future.flatMap{
case Some(a) => f(a).future
case None => Future.successful(None)
}
FutureO(newFuture)
}
def map[B](f: A => B)(implicit ec: ExecutionContext): FutureO[B] = {
FutureO(future.map(option => option map f))
}
def recoverWith[U >: A](pf: PartialFunction[Throwable, FutureO[U]])(implicit executor: ExecutionContext): FutureO[U] = {
val futOtoFut: FutureO[U] => Future[Option[U]] = _.future
FutureO(future.recoverWith(pf andThen futOtoFut))
}
def orElse[U >: A](other: => FutureO[U])(implicit executor: ExecutionContext): FutureO[U] = {
FutureO(future.flatMap{
case None => other.future
case _ => this.future
})
}
}
And now we can re-write our process preserving the same structure as the future-based composition.
type UserPoints = Long
object db {
def getUserPoints(username: String): Future[Option[UserPoints]] = ???
def getSomething(s: String): Future[Option[Long]] = ???
def setPoints(pointId: UserPoints, username:String): Future[Unit] = ???
def findPointsForUser(username: String): Future[Option[Long]] = ???
}
class PointsNotFound extends Exception("bonk")
class StuffNotFound extends Exception("sthing not found")
def getUserPoints2(username:String): Future[Option[UserPoints]] = {
val futureOpt = FutureO(db.getUserPoints(username))
.map(userPoints => userPoints /*.total*/)
.orElse{
(for {
sthingElse <- FutureO(db.getSomething("abc"))
_ <- FutureO(db.setPoints(sthingElse, username).map(_ => Some(())))
points <- FutureO(db.findPointsForUser(username))
} yield (points))
.orElse{
FutureO(db.findPointsForUser(username))
}
}
futureOpt.future
}
[1] with acknowledgements to http://loicdescotte.github.io/posts/scala-compose-option-future/
I want to add an after(d: FiniteDuration)(callback: => Unit) util to Scala Futures that would enable me to do this:
val f = Future(someTask)
f.after(30.seconds) {
println("f has not completed in 30 seconds!")
}
f.after(60.seconds) {
println("f has not completed in 60 seconds!")
}
How can I do this?
Usually I use a thread pool executor and promises:
import scala.concurrent.duration._
import java.util.concurrent.{Executors, ScheduledThreadPoolExecutor}
import scala.concurrent.{Future, Promise}
val f: Future[Int] = ???
val executor = new ScheduledThreadPoolExecutor(2, Executors.defaultThreadFactory(), AbortPolicy)
def withDelay[T](operation: ⇒ T)(by: FiniteDuration): Future[T] = {
val promise = Promise[T]()
executor.schedule(new Runnable {
override def run() = {
promise.complete(Try(operation))
}
}, by.length, by.unit)
promise.future
}
Future.firstCompletedOf(Seq(f, withDelay(println("still going"))(30 seconds)))
Future.firstCompletedOf(Seq(f, withDelay(println("still still going"))(60 seconds)))
One way is to use Future.firstCompletedOf (see this blogpost):
val timeoutFuture = Future { Thread.sleep(500); throw new TimeoutException }
val f = Future.firstCompletedOf(List(f, timeoutFuture))
f.map { case e: TimeoutException => println("f has not completed in 0.5 seconds!") }
where TimeoutException is some exception or type.
Use import akka.pattern.after. If you want to implement it without akka here is the source code. The other (java) example is TimeoutFuture in com.google.common.util.concurrent.
Something like this, perhaps:
object PimpMyFuture {
implicit class PimpedFuture[T](val f: Future[T]) extends AnyVal {
def after(delay: FiniteDuration)(callback: => Unit): Future[T] = {
Future {
blocking { Await.ready(f, delay) }
} recover { case _: TimeoutException => callback }
f
}
}
}
import PimpMyFuture._
Future { Thread.sleep(10000); println ("Done") }
.after(5.seconds) { println("Still going") }
This implementation is simple, but it basically doubles the number of threads you need - each active future effectively occupies two threads - which is a bit wasteful. Alternatively, you could use scheduled tasks to make your waits non-blocking. I don't know of a "standard" scheduler in scala (each lib has their own), but for a simple task like this you can use java's TimerTask directly:
object PimpMyFutureNonBlocking {
val timer = new java.util.Timer
implicit class PimpedFuture[T](val f: Future[T]) extends AnyVal {
def after(delay: FiniteDuration)(callback: => Unit): Future[T] = {
val task = new java.util.TimerTask {
def run() { if(!f.isCompleted) callback }
}
timer.schedule(task, delay.toMillis)
f.onComplete { _ => task.cancel }
f
}
}
}