How to properly call a single server from multiple actors / web handlers using Akka HTTP?

问题

I have a service (let's call it Service A) which uses Akka Server HTTP to handle incoming requests. Also I have 3rd party application (Service B) which provides several web services. The purpose of service A is to transform client requests, call one or multiple web services of service B, merge/transform results and serve it back to a client.

I am using Actors for some parts, and just Future for other. To make a call to Service B, I use Akka HTTP client.

Http.get(actorSystem).singleRequest(HttpRequest.create()
        .withUri("http://127.0.0.1:8082/test"), materializer)
        .onComplete(...)

The issue is, a new flow is created per each Service A request, and if there are multiple concurrent connections, it results in akka.stream.OverflowStrategy$Fail$BufferOverflowException: Exceeded configured max-open-requests value of [32] error

I already asked this question and got a suggestion to use a single Flow How to properly call Akka HTTP client for multiple (10k - 100k) requests?

While it works for a batch of requests coming from a single place, I don't know how to use a single Flow from all my concurrent request handlers.

What is the correct "Akka-way" to do it?

回答1:

I think you could use Source.queue to buffer your requests. The code below assume that you need to get the answer from 3rd party service, so having a Future[HttpResponse] is very welcomed. This way you could also provide an overflow strategy to prevent resource starvation.

import akka.actor.ActorSystem
import akka.http.scaladsl.Http
import akka.http.scaladsl.model.{HttpRequest, HttpResponse}
import akka.stream.scaladsl.{Keep, Sink, Source}
import akka.stream.{ActorMaterializer, OverflowStrategy}

import scala.concurrent.duration._
import scala.concurrent.{Await, Future, Promise}
import scala.util.{Failure, Success}

import scala.concurrent.ExecutionContext.Implicits.global

implicit val system = ActorSystem("main")
implicit val materializer = ActorMaterializer()
val pool = Http().cachedHostConnectionPool[Promise[HttpResponse]](host = "google.com", port = 80)
val queue = Source.queue[(HttpRequest, Promise[HttpResponse])](10, OverflowStrategy.dropNew)
  .via(pool)
  .toMat(Sink.foreach({
    case ((Success(resp), p)) => p.success(resp)
    case ((Failure(e), p)) => p.failure(e)
  }))(Keep.left)
  .run
val promise = Promise[HttpResponse]
val request = HttpRequest(uri = "/") -> promise

val response = queue.offer(request).flatMap(buffered => {
  if (buffered) promise.future
  else Future.failed(new RuntimeException())
})

Await.ready(response, 3 seconds)

(code copied from my blog post)

回答2:

Here is Java version of the accepted answer

final Flow<
    Pair<HttpRequest, Promise<HttpResponse>>,
    Pair<Try<HttpResponse>, Promise<HttpResponse>>,
    NotUsed> flow =
    Http.get(actorSystem).superPool(materializer);

final SourceQueue<Pair<HttpRequest, Promise<HttpResponse>>> queue = Source.<Pair<HttpRequest, Promise<HttpResponse>>>
    queue(BUFFER_SIZE, OverflowStrategy.dropNew())
    .via(flow)
        .toMat(Sink.foreach(p -> p.second().complete(p.first())), Keep.left())
        .run(materializer);

...

public CompletionStage<HttpResponse> request(HttpRequest request) {
    log.debug("Making request {}", request);

    Promise<HttpResponse> promise = Futures.promise();
    return queue.offer(Pair.create(request, promise))
        .thenCompose(buffered -> {
            if (buffered instanceof QueueOfferResult.Enqueued$) {
                return FutureConverters.toJava(promise.future())
                    .thenApply(resp -> {
                        if (log.isDebugEnabled()) {
                            log.debug("Got response {} {}", resp.status(), resp.getHeaders());
                        }
                        return resp;
                    });
            } else {
                log.error("Could not buffer request {}", request);
                return CompletableFuture.completedFuture(HttpResponse.create().withStatus(StatusCodes.SERVICE_UNAVAILABLE));
            }
        });
}

回答3:

All you need to do is setup a HostConnectionPool to Service B within your Service A code. This will give you a Flow that can be added to your Service A streams to dispatch requests from A to B using a connection pool instead of a new connection per stream. From the documentation:

As opposed to the Connection-Level Client-Side API the host-level API relieves you from manually managing individual HTTP connections. It autonomously manages a configurable pool of connections to one particular target endpoint (i.e. host/port combination).

Each materialization of this Flow, in different streams, will draw from this underlying pool of connections:

The best way to get a hold of a connection pool to a given target endpoint is the Http.get(system).cachedHostConnectionPool(...) method, which returns a Flow that can be "baked" into an application-level stream setup. This flow is also called a "pool client flow".

来源：https://stackoverflow.com/questions/34808752/how-to-properly-call-a-single-server-from-multiple-actors-web-handlers-using-a