Question

我想实现一个Flow来处理分页结果（例如，底层服务返回一些结果，但也表示通过发出另一个请求，传入更多结果，例如光标）。

到目前为止我做过的事情：

我已经实现了以下流程和测试，但是流程没有完成。

object AdditionalRequestsFlow {

  private def keepRequest[Request, Response](flow: Flow[Request, Response, NotUsed]): Flow[Request, (Request, Response), NotUsed] = {
    Flow.fromGraph(GraphDSL.create() { implicit builder: GraphDSL.Builder[NotUsed] =>
      import GraphDSL.Implicits._
      val in = builder.add(Flow[Request])

      val bcast = builder.add(Broadcast[Request](2))
      val merge = builder.add(Zip[Request, Response]())

      in ~> bcast         ~> merge.in0
            bcast ~> flow ~> merge.in1

      FlowShape(in.in, merge.out)
    })
  }

  def flow[Request, Response, Output](
    inputFlow: Flow[Request, Response, NotUsed],
    anotherRequest: (Request, Response) => Option[Request],
    extractOutput: Response => Output,
    mergeOutput: (Output, Output) => Output
  ): Flow[Request, Output, NotUsed] = {
    Flow.fromGraph(GraphDSL.create() { implicit b =>
      import GraphDSL.Implicits._

      val start = b.add(Flow[Request])
      val merge = b.add(Merge[Request](2))
      val underlying = b.add(keepRequest(inputFlow))
      val unOption = b.add(Flow[Option[Request]].mapConcat(_.toList))
      val unzip = b.add(UnzipWith[(Request, Response), Response, Option[Request]] { case (req, res) =>
        (res, anotherRequest(req, res))
      })
      val finish = b.add(Flow[Response].map(extractOutput)) // this is wrong as we don't keep to 1 Request -> 1 Output, but first let's get the flow to work

      start ~> merge ~> underlying ~> unzip.in
                                      unzip.out0            ~>  finish
               merge <~ unOption   <~ unzip.out1

      FlowShape(start.in, finish.out)
    })
  }       
}

测试：

    import akka.NotUsed
    import akka.actor.ActorSystem
    import akka.stream.ActorMaterializer
    import akka.stream.scaladsl.{Flow, Sink, Source}
    import org.scalatest.FlatSpec
    import org.scalatest.Matchers._
    import cats.syntax.option._
    import org.scalatest.concurrent.ScalaFutures.whenReady

    class AdditionalRequestsFlowSpec extends FlatSpec {
      implicit val system = ActorSystem()
      implicit val materializer = ActorMaterializer()

      case class Request(max: Int, batchSize: Int, offset: Option[Int] = None)
      case class Response(values: List[Int], nextOffset: Option[Int])

      private val flow: Flow[Request, Response, NotUsed] = {
        Flow[Request]
          .map { request =>
            val start = request.offset.getOrElse(0)
            val end = Math.min(request.max, start + request.batchSize)
            val nextOffset = if (end == request.max) None else Some(end)
            val result = Response((start until end).toList, nextOffset)
            result
          }
      }

      "AdditionalRequestsFlow" should "collect additional responses" in {
        def anotherRequest(request: Request, response: Response): Option[Request] = {
          response.nextOffset.map { nextOffset => request.copy(offset = nextOffset.some) }
        }

        def extract(x: Response): List[Int] = x.values
        def merge(a: List[Int], b: List[Int]): List[Int] = a ::: b

        val requests =
          Request(max = 35, batchSize = 10) ::
          Request(max = 5, batchSize = 10) ::
          Request(max = 100, batchSize = 1) ::
          Nil

        val expected = requests.map { x =>
          (0 until x.max).toList
        }

        val future = Source(requests)
          .via(AdditionalRequestsFlow.flow(flow, anotherRequest, extract, merge))
          .runWith(Sink.seq)

        whenReady(future) { x =>
          x shouldEqual expected
        }
      }
    }

以可怕的阻塞方式实现相同的流程，以说明我想要实现的目标：

   def uglyHackFlow[Request, Response, Output](
    inputFlow: Flow[Request, Response, NotUsed],
    anotherRequest: (Request, Response) => Option[Request],
    extractOutput: Response => Output,
    mergeOutput: (Output, Output) => Output
  ): Flow[Request, Output, NotUsed] = {
    implicit val system = ActorSystem()
    implicit val materializer = ActorMaterializer()

    Flow[Request]
      .map { x =>
        def grab(request: Request): Output = {
          val response = Await.result(Source.single(request).via(inputFlow).runWith(Sink.head), 10.seconds) // :(
          val another = anotherRequest(request, response)
          val output = extractOutput(response)
          another.map { another =>
            mergeOutput(output, grab(another))
          } getOrElse output
        }

        grab(x)
      }
  }

这有效（但我们不应该在此时实现任何/ Await）。

已审核http://doc.akka.io/docs/akka/2.4/scala/stream/stream-graphs.html#Graph_cycles__liveness_and_deadlocks我认为其中包含答案，但我似乎无法在那里找到答案。在我的情况下，我希望循环在大多数时候应该包含一个元素，所以既不会发生缓冲区溢出也不会发生完全饥饿 - 但显然确实如此。
尝试使用.withAttributes(Attributes(LogLevels(...)))调试流但是，尽管看似正确配置的记录器，但它不会产生任何输出。

我正在寻找提示如何修复flow方法，保持相同的签名和语义（测试会通过）。

或许我在这里做了一些完全偏离基础的事情（例如，akka-stream-contrib中有一个现有的功能可以解决这个问题吗？

Answer 1

我认为使用Source.unfold比创建自定义图表更安全。以下是我通常所做的事情（根据API的不同而略有不同）。

  override def getArticles(lastTokenOpt: Option[String], filterIds: (Seq[Id]) => Seq[Id]): Source[Either[String, ImpArticle], NotUsed] = {

    val maxRows = 1000

    def getUri(cursor: String, count: Int) = s"/works?rows=$count&filter=type:journal-article&order=asc&sort=deposited&cursor=${URLEncoder.encode(cursor, "UTF-8")}"

    Source.unfoldAsync(lastTokenOpt.getOrElse("*")) { cursor =>

      println(s"Getting ${getUri(cursor, maxRows)}")
      if (cursor.nonEmpty) {
        sendGetRequest[CrossRefResponse[CrossRefList[JsValue]]](getUri(cursor, maxRows)).map {
          case Some(response) =>
            response.message match {
              case Left(list) if response.status == "ok" =>

                println(s"Got ${list.items.length} items")
                val items = list.items.flatMap { js =>
                  try {
                    parseArticle(js)
                  } catch {
                    case ex: Throwable =>
                      logger.error(s"Error on parsing: ${js.compactPrint}")
                      throw ex
                  }
                }

                list.`next-cursor` match {
                  case Some(nextCursor) =>
                    Some(nextCursor -> (items.map(Right.apply).toList ::: List(Left(nextCursor))))
                  case None =>
                    logger.error(s"`next-cursor` is missing when fetching from CrossRef [status ${response.status}][${getUri(cursor, maxRows)}]")
                    Some("" -> items.map(Right.apply).toList)
                }
              case Left(jsvalue) if response.status != "ok" =>
                logger.error(s"API error on fetching data from CrossRef [status ${response.status}][${getUri(cursor, maxRows)}]")
                None
              case Right(someError) =>
                val cause = someError.fold(errors => errors.map(_.message).mkString(", "), ex => ex.message)
                logger.error(s"API error on fetching data from CrossRef [status $cause}][${getUri(cursor, maxRows)}]")
                None
            }

          case None =>
            logger.error(s"Got error on fetching ${getUri(cursor, maxRows)} from CrossRef")
            None
        }
      } else
        Future.successful(None)
    }.mapConcat(identity)
  }

在您的情况下，您可能甚至不需要将光标推送到流。我之所以这样做是因为我将最后成功的光标存储在数据库中以便以后能够在发生故障时恢复。

Answer 2

感觉像video涵盖了你想要做的事情的要点。他们创建一个自定义Graphstage来维护状态并将其发送回服务器，响应流取决于发回的状态，他们还有一个事件来表示完成（在你的情况下，它将是你检查的地方） p>

处理分页结果的Akka Streams流程未完成

2 个答案: