How to bind akka http with akka streams?

Question

I'm trying to use streams instead of pure actors to handle http requests and I came with the following code:

trait ImagesRoute {

  val log = LoggerFactory.getLogger(this.getClass)

  implicit def actorRefFactory: ActorRefFactory
  implicit def materializer: ActorMaterializer

  val source =
    Source
      .actorRef[Image](Int.MaxValue, OverflowStrategy.fail)
      .via(Flow[Image].mapAsync(1)(ImageRepository.add))
      .toMat(Sink.asPublisher(true))(Keep.both)

  val route = {
    pathPrefix("images") {
      pathEnd {
        post {
          entity(as[Image]) { image =>

            val (ref, publisher) = source.run()

            val addFuture = Source.fromPublisher(publisher)

            val future = addFuture.runWith(Sink.head[Option[Image]])

            ref ! image

            onComplete(future.mapTo[Option[Image]]) {
              case Success(img) =>
                complete(Created, img)

              case Failure(e) =>
                log.error("Error adding image resource", e)
                complete(InternalServerError, e.getMessage)
            }
          }
        }
      }
    }
  }
}

I'm not sure if this is the correct way to do that, or even if this is a good approach or if I should use an actor to interact with the route, using the ask pattern and then inside the actor, stream everything.

Any ideas?

Answer 1

If you're only expecting 1 image from the entity then you don't need to create a Source from an ActorRef and you don't need Sink.asPublisher, you can simply use Source.single:

def imageToComplete(img : Option[Image]) : StandardRoute = 
  img.map(i => complete(Created, i))
     .getOrElse {
       log error ("Error adding image resource", e)
       complete(InternalServerError, e.getMessage
     }

...

entity(as[Image]) { image =>

  val future : Future[StandardRoute] = 
    Source.single(image)
          .via(Flow[Image].mapAsync(1)(ImageRepository.add))
          .runWith(Sink.head[Option[Image]])
          .map(imageToComplete)

  onComplete(future)
}

Simplifying your code further, the fact that you are only processing 1 image means that Streams are unnecessary since there is no need for backpressure with just 1 element:

val future : Future[StandardRoute] = ImageRepository.add(image)
                                                    .map(imageToComplete)

onComplete(future)

In the comments you indicated

"this is just a simple example, but the stream pipeline should be bigger doing a lot of things like contacting external resources and eventually back pressure things"

This would only apply if your entity was a stream of images. If you're only ever processing 1 image per HttpRequest then backpressure never applies, and any stream you create will be a slower version of a Future.

If your entity is in fact a stream of Images, then you could use it as part of stream:

val byteStrToImage : Flow[ByteString, Image, _] = ???

val imageToByteStr : Flow[Image, Source[ByteString], _] = ???

def imageOptToSource(img : Option[Image]) : Source[Image,_] =
  Source fromIterator img.toIterator

val route = path("images") {
  post {
    extractRequestEntity { reqEntity =>

      val stream = reqEntity.via(byteStrToImage)
                            .via(Flow[Image].mapAsync(1)(ImageRepository.add))
                            .via(Flow.flatMapConcat(imageOptToSource))
                            .via(Flow.flatMapConcat(imageToByteStr))

      complete(HttpResponse(status=Created,entity = stream))
    }
  }
}