I am currently playing around with akka streams and tried the following example.
Get the first element from kafka when requesting a certain HTTP endpoint. This is the code I wrote and its working.
get {
path("ticket" / IntNumber) { ticketNr =>
val future = Consumer.plainSource(consumerSettings, Subscriptions.topics("tickets"))
.take(1)
.completionTimeout(5 seconds)
.runWith(Sink.head)
onComplete(future) {
case Success(record) => complete(HttpEntity(ContentTypes.`text/html(UTF-8)`, record.value()))
case _ => complete(HttpResponse(StatusCodes.NotFound))
}
}
}
I am just wondering if this is the ideomatic way of working with (akka) streams. So is there a more "direct" way of connecting the kafka stream to the HTTP response stream?
For example, when POSTing I do this:
val kafkaTicketsSink = Flow[String]
.map(new ProducerRecord[Array[Byte], String]("tickets", _))
.toMat(Producer.plainSink(producerSettings))(Keep.right)
post {
path("ticket") {
(entity(as[Ticket]) & extractMaterializer) { (ticket, mat) => {
val f = Source.single(ticket).map(t => t.description).runWith(kafkaTicketsSink)(mat)
onComplete(f) { _ =>
val locationHeader = headers.Location(s"/ticket/${ticket.id}")
complete(HttpResponse(StatusCodes.Created, headers = List(locationHeader)))
}
}
}
}
}
Maybe this can also be improved??
You could keep a single, backpressured stream alive using Sink.queue
. You can pull an element from the materialized queue every time a request is received. This should give you back one element if available, and backpressure otherwise.
Something along the lines of:
val queue = Consumer.plainSource(consumerSettings, Subscriptions.topics("tickets"))
.runWith(Sink.queue())
get {
path("ticket" / IntNumber) { ticketNr =>
val future: Future[Option[ConsumerRecord[String, String]]] = queue.pull()
onComplete(future) {
case Success(Some(record)) => complete(HttpEntity(ContentTypes.`text/html(UTF-8)`, record.value()))
case _ => complete(HttpResponse(StatusCodes.NotFound))
}
}
}
More info on Sink.queue
can be found in the docs.