Akka Streams - jump over FlowShape - scala

Question

I have the following Graph.
At the inflateFlow stage, I check if there is already a request processed in DB. If there is already a processed message, I want to return MsgSuccess and not a RequestProcess, but the next FlowShape won't accept that, it needs a RequestProcess. Is there a way to jump from flowInflate to flowWrap without adding Either everywhere?
GraphDSL.create() { implicit builder: GraphDSL.Builder[NotUsed] =>
import GraphDSL.Implicits._
val flowInflate = builder.add(wrapFunctionInFlowShape[MsgRequest, RequestProcess](inflateFlow))
val flowProcess = builder.add(wrapFunctionInFlowShape[RequestProcess, SuccessProcess](convertFlow))
val flowWrite = builder.add(wrapFunctionInFlowShape[SuccessProcess, SuccessProcess](writeFlow))
val flowWrap = builder.add(wrapFunctionInFlowShape[SuccessProcess, MsgSuccess](wrapFlow))
flowInflate ~> flowProcess ~> flowWrite ~> flowWrap
FlowShape(flowInflate.in, flowWrap.out)
}
def wrapFunctionInFlowShape[Input, Output](f: Input => Output): Flow[Input, Output, NotUsed] = {
Flow.fromFunction { input =>
f(input)
}
}
//check for cache
def inflateFlow(msgRequest: MsgRequest): Either[RequestProcess, MsgSuccess] = {
val hash: String = hashMethod(msgRequest)
if(existisInDataBase(hash))
Right(MsgSuccess(hash))
else
Left(inflate(msgRequest))
}
def convertFlow(requestPorocess: RequestPocess): SuccessProcess = {}//process the request}
def writeFlow(successProcess: SuccessProcess): SuccessProcess = {}//write to DB}
def wrapFlow(successProcess: SuccessProcess): MsgSuccess = {}//wrap and return the message}

Answer 1

You can define alternative paths in a stream with a partition. In your case, the PartitionWith stage in the Akka Stream Contrib project could be helpful. Unlike the Partition stage in the standard Akka Streams API, PartitionWith allows the output types to be different: in your case, the output types are RequestProcess and MsgSuccess.
First, to use PartitionWith, add the following dependency to your build.sbt:
libraryDependencies += "com.typesafe.akka" %% "akka-stream-contrib" % "0.8"
Second, replace inflateFlow with the partition:
def split = PartitionWith[MsgRequest, RequestProcess, MsgSuccess] { msgRequest =>
val hash = hashMethod(msgRequest)
if (!existisInDataBase(hash))
Left(inflate(msgRequest))
else
Right(MsgSuccess(hash))
}
Then incorporate that stage into your graph:
val flow = Flow.fromGraph(GraphDSL.create() { implicit builder: GraphDSL.Builder[NotUsed] =>
import GraphDSL.Implicits._
val pw = builder.add(split)
val flowProcess = builder.add(wrapFunctionInFlowShape[RequestProcess, SuccessProcess](convertFlow))
val flowWrite = builder.add(wrapFunctionInFlowShape[SuccessProcess, SuccessProcess](writeFlow))
val flowWrap = builder.add(wrapFunctionInFlowShape[SuccessProcess, MsgSuccess](wrapFlow))
val mrg = builder.add(Merge[MsgSuccess](2))
pw.out0 ~> flowProcess ~> flowWrite ~> flowWrap ~> mrg.in(0)
pw.out1 ~> mrg.in(1)
FlowShape(pw.in, mrg.out)
})
If an incoming MsgRequest is not found in the database and is converted to a RequestProcess, then that message goes through your original flow path. If an incoming MsgRequest is in the database and resolves to a MsgSuccess, then it bypasses the intermediate steps in the flow. In both cases, the resulting MsgSuccess messages are merged from the two alternative paths into one flow outlet.