Over the past few days I have been trying to figure out the best way to download a HTTP resource to a file using Akka Streams and HTTP.
Initially I started with the Future-Based Variant and that looked something like this:
def downloadViaFutures(uri: Uri, file: File): Future[Long] = {
val request = Get(uri)
val responseFuture = Http().singleRequest(request)
responseFuture.flatMap { response =>
val source = response.entity.dataBytes
source.runWith(FileIO.toFile(file))
}
}
That was kind of okay but once I learnt more about pure Akka Streams I wanted to try and use the Flow-Based Variant to create a stream starting from a Source[HttpRequest]
. At first this completely stumped me until I stumbled upon the flatMapConcat
flow transformation. This ended up a little more verbose:
def responseOrFail[T](in: (Try[HttpResponse], T)): (HttpResponse, T) = in match {
case (responseTry, context) => (responseTry.get, context)
}
def responseToByteSource[T](in: (HttpResponse, T)): Source[ByteString, Any] = in match {
case (response, _) => response.entity.dataBytes
}
def downloadViaFlow(uri: Uri, file: File): Future[Long] = {
val request = Get(uri)
val source = Source.single((request, ()))
val requestResponseFlow = Http().superPool[Unit]()
source.
via(requestResponseFlow).
map(responseOrFail).
flatMapConcat(responseToByteSource).
runWith(FileIO.toFile(file))
}
Then I wanted to get a little tricky and use the Content-Disposition
header.
Going back to the Future-Based Variant:
def destinationFile(downloadDir: File, response: HttpResponse): File = {
val fileName = response.header[ContentDisposition].get.value
val file = new File(downloadDir, fileName)
file.createNewFile()
file
}
def downloadViaFutures2(uri: Uri, downloadDir: File): Future[Long] = {
val request = Get(uri)
val responseFuture = Http().singleRequest(request)
responseFuture.flatMap { response =>
val file = destinationFile(downloadDir, response)
val source = response.entity.dataBytes
source.runWith(FileIO.toFile(file))
}
}
But now I have no idea how to do this with the Future-Based Variant. This is as far as I got:
def responseToByteSourceWithDest[T](in: (HttpResponse, T), downloadDir: File): Source[(ByteString, File), Any] = in match {
case (response, _) =>
val source = responseToByteSource(in)
val file = destinationFile(downloadDir, response)
source.map((_, file))
}
def downloadViaFlow2(uri: Uri, downloadDir: File): Future[Long] = {
val request = Get(uri)
val source = Source.single((request, ()))
val requestResponseFlow = Http().superPool[Unit]()
val sourceWithDest: Source[(ByteString, File), Unit] = source.
via(requestResponseFlow).
map(responseOrFail).
flatMapConcat(responseToByteSourceWithDest(_, downloadDir))
sourceWithDest.runWith(???)
}
So now I have a Source
that will emit one or more (ByteString, File)
elements for each File
(I say each File
since there is no reason the original Source
has to be a single HttpRequest
).
Is there anyway to take these and route them to a dynamic Sink
?
I'm thinking something like flatMapConcat
, such as:
def runWithMap[T, Mat2](f: T => Graph[SinkShape[Out], Mat2])(implicit materializer: Materializer): Mat2 = ???
So that I could complete downloadViaFlow2
with:
def destToSink(destination: File): Sink[(ByteString, File), Future[Long]] = {
val sink = FileIO.toFile(destination, true)
Flow[(ByteString, File)].map(_._1).toMat(sink)(Keep.right)
}
sourceWithDest.runWithMap {
case (_, file) => destToSink(file)
}
The solution does not require a flatMapConcat. If you don't need any return values from the file writing then you can use Sink.foreach
:
def writeFile(downloadDir : File)(httpResponse : HttpResponse) : Future[Long] = {
val file = destinationFile(downloadDir, httpResponse)
httpResponse.entity.dataBytes.runWith(FileIO.toFile(file))
}
def downloadViaFlow2(uri: Uri, downloadDir: File) : Future[Unit] = {
val request = HttpRequest(uri=uri)
val source = Source.single((request, ()))
val requestResponseFlow = Http().superPool[Unit]()
source.via(requestResponseFlow)
.map(responseOrFail)
.map(_._1)
.runWith(Sink.foreach(writeFile(downloadDir)))
}
Note that the Sink.foreach
creates Futures
from the writeFile
function. Therefore there's not much back-pressure involved. The writeFile could be slowed down by the hard drive but the stream would keep generating Futures. To control this you can use Flow.mapAsyncUnordered
(or Flow.mapAsync
) :
val parallelism = 10
source.via(requestResponseFlow)
.map(responseOrFail)
.map(_._1)
.mapAsyncUnordered(parallelism)(writeFile(downloadDir))
.runWith(Sink.ignore)
If you want to accumulate the Long values for a total count you need to combine with a Sink.fold
:
source.via(requestResponseFlow)
.map(responseOrFail)
.map(_._1)
.mapAsyncUnordered(parallelism)(writeFile(downloadDir))
.runWith(Sink.fold(0L)(_ + _))
The fold will keep a running sum and emit the final value when the source of requests has dried up.