And you accomplish this with the low level or DSL api.
On Saturday, April 8, 2017 at 12:56:26 AM UTC-5, rrodseth wrote:
>
> I use per-request actors. The only ask() is in the route, and is sent to a
> RequestHandler actor. The RequestHandler creates a per-request actor with
> the sender as a Props/Constructor parameter ("requestor"), and sends it a
> "Run" message.
>
> The per-request actor can coordinate with as many actors as it needs to.
> It can use "self" as a "replyTo" property of messages it sends out (tell).
> It sends the final result back to its "requestor" and stops itself.
>
> With Spray I didn't need the outermost ask(), but I think it's fine having
> one at the outermost layer.
>
> On Fri, Apr 7, 2017 at 9:45 PM, kraythe <kra...@gmail.com <javascript:>>
> wrote:
>
>> A very interesting read and I appreciate you showing examples. What I do
>> wonder about there is what the situation is when the server has to do more
>> than simply return information from a database or other similar web rest-ui
>> simple examples. For example, I have processes that require aggregating
>> data from several actors to complete the process. Now I could do these in a
>> series of ASKs but then that would feel like a complete anti-pattern to me.
>> I would rather model the system as an entirely one way calling pattern
>> where the only future is in the final completion of the route. So here is
>> the challenge:
>>
>> You have a route called /customerInfo which finds the search parameters
>> out of the JSON body that is passed to the api. Then it has to invoke calls
>> to two ProductManagerActor and InventoryManagerActor. These actors will
>> send back responses which might result in the task being done but will be
>> much more likely to require calling actor ShippingStatusActor several
>> times to ask for different pieces of data. Once you have collected all of
>> the data then finally you can assemble the JSON response. Yes. I could
>> model this imperatively with futures and using the ask pattern inside the
>> route but then I am massively constrained resources and I have never found
>> that to be scalable under load. Indeed I try to avoid futures as much as I
>> can because of these difficulties. I would prefer instead to have an actor
>> handle the request, recieving messages from the three actors and then when
>> it has assembled all of the data it would generate a response or if there
>> is a timeout or exception it would generate other responses.
>>
>> Now using the route DSL I wonder how you would manage this process within
>> your route files. At some point you have to jump off the DSL and offload
>> the task to an actor. For example you can decode the parameters and the
>> body and so on using the route but then you have to start the process of
>> collecting this massive amount of data to send back and then you are off
>> the route and your options are now limited as to how you can respond. You
>> cant simply say inside the DataCollectionActor to invoke the bad request
>> response, through the DSL. you now have to use the low level API to signal
>> a response.
>>
>> I would be curious how you could handle this use case with the DSL.
>>
>> -- Robert
>>
>>
>> On Tuesday, April 4, 2017 at 2:38:52 AM UTC-5, Daniel Stoner wrote:
>>>
>>> Hi Kraythe,
>>>
>>> Perhaps it helps to see a real world example that we've been working on
>>> - with a good number of routes involved.
>>>
>>> This is from our AkkaHttpServer class. It's job is to inject all the
>>> routes (ordersV2, searchv3, searchTerms, persistence) which consist of
>>> around 6 actual endpoints per injected class - into the right point in the
>>> hierarchy (Below the oAuth2 authenticator and any request/response loggers
>>> and whatnot that you may need).
>>>
>>> We define index.html and healthcheck route in this class since they are
>>> one liners that live above oAuth2 security otherwise we would also inject
>>> them independently.
>>>
>>> Route indexRoute = get(() -> route(pathSingleSlash(() ->
>>> getFromResource("web/index.html"))));
>>> Route healthCheck = get(() -> path(PATH_HEALTH_CHECK, () ->
>>> extractRequestContext(healthCheckHandler::handle)));
>>>
>>> Route apis = route(
>>> indexRoute,
>>> healthCheck,
>>> oauth2Authentication(
>>> accessTokenVerifier,
>>> route(
>>> ordersV2,
>>> searchV3,
>>> searchTerms,
>>> persistence
>>> )
>>> )
>>> );
>>>
>>> return logRequestResult(
>>> this::requestMethodAsInfo,
>>> this::rejectionsAsInfo,
>>> () -> handleExceptions(
>>> exceptionHandlerLogAndReturnInternalError(),
>>> () -> handleRejections(
>>> rejectionHandlerLogAndReturnNotFound(),
>>> () -> apis
>>> )
>>> )
>>> );
>>>
>>> Note the *handleExceptions *and *handleRejections* methods. Basically
>>> if your in the depths of a route (the bit supposed to handle the request)
>>> you have 3 options.
>>>
>>> 1) Handle the request and reply with a HttpResponse.
>>> return HttpResponse.create().withStatus(StatusCodes.CREATED).addHeader(
>>> RawHeader.create(HttpHeaders.LOCATION, location)
>>> )
>>> 2) Reject the request with an explicit Rejection
>>> return reject(Rejections.authorizationFailed());
>>> 3) Throw an exception directly
>>> throw new MyCustomException("Something went dodgy here!")
>>>
>>> Now how you transpose those rejections or exceptions into a HttpResponse
>>> is up to your generic rejection or exception handler right at the very top
>>> of your routes. I've included our 'ErrorHandlingDirectives' class which we
>>> simply extend (Instead of extending AllDirectives) wherever we need this:
>>>
>>> public class ErrorHandlingDirectives extends AllDirectives {
>>>
>>> private static final Logger LOG =
>>> LoggerFactory.getLogger(ErrorHandlingDirectives.class);
>>>
>>> public LogEntry requestMethodAsInfo(HttpRequest request,
>>> HttpResponse response) {
>>> String headers = toCensoredHeaderJson(request.getHeaders());
>>>
>>> return LogEntry.create(
>>> "Server has received a request\n"
>>> + request.method().name() + " " +
>>> request.getUri().toString() + "\n"
>>> + headers + "\n"
>>> + "Server responded with a response\n"
>>> + response.status() + "\n"
>>> + "Content-Type: " +
>>> response.entity().getContentType().toString() + "\n"
>>> + "Content-Length: " +
>>> response.entity().getContentLengthOption().orElse(-1),
>>> InfoLevel());
>>> }
>>>
>>> public LogEntry rejectionsAsInfo(HttpRequest request,
>>> List<Rejection> rejections) {
>>> String headers = toCensoredHeaderJson(request.getHeaders());
>>>
>>> return LogEntry.create(
>>> "Server has received a request\n"
>>> + request.method().name() + " " +
>>> request.getUri().toString() + "\n"
>>> + headers + "\n"
>>> + "Server responded with a rejection\n"
>>> +
>>> rejections.stream().map(Rejection::toString).collect(Collectors.joining("\n")),
>>> InfoLevel());
>>> }
>>>
>>> public ExceptionHandler exceptionHandlerLogAndReturnInternalError() {
>>> return ExceptionHandler
>>> .newBuilder()
>>> .matchAny(throwable -> extractRequest(request -> {
>>> LOG.warn("Error on route: " +
>>> request.method().value() + " " + request.getUri().toString() + " " +
>>> throwable.getMessage(), throwable);
>>> return complete(StatusCodes.INTERNAL_SERVER_ERROR);
>>> })
>>> ).build();
>>> }
>>>
>>> public String toCensoredHeaderJson(Iterable<HttpHeader> headers) {
>>> return StreamSupport
>>> .stream(headers.spliterator(), false)
>>> .map(header -> {
>>> if (header instanceof Authorization) {
>>> return header.name() + ": CENSORED";
>>> }
>>> return header.name() + ": " + header.value();
>>> })
>>> .collect(Collectors.joining("\n"));
>>> }
>>>
>>> }
>>>
>>>
>>> So - yes you 'can' write 1 big file with a bazillion routes in it - or
>>> you can do what most developers do eventually once things are working and
>>> split it down into lots of individual classes with their own hierarchy.
>>> Have some general rules for your approach (such as if someone throws a
>>> BeanVerifiyException you return a particular type of HTTP status code) and
>>> you'll soon be enjoying things again :)
>>>
>>> As a general rule - whenever we reach the part of our route where we
>>> actually plan to handle a request we generally spawn an actor to do the
>>> job. For me it's good that the Actor speaks in terms of messages and failed
>>> futures - and doesn't have to worry about 'Oh no I better not throw an
>>> exception because I really need to return a HttpResponse of
>>> INTERNAL_SERVER_ERROR'. This job can be done in your generic handlers :)
>>>
>>> On Monday, 3 April 2017 02:09:32 UTC+1, kraythe wrote:
>>>>
>>>> I was really excited about akka-http as I would be able to unburden my
>>>> code from the baggage of play and handle my server side as a pure akka
>>>> actors app but unless I am much mistaken something is dreadfully amiss
>>>> with
>>>> the implementation.
>>>>
>>>> One of the main core features is the actor paradigm and the integration
>>>> of a rich actor system. However, the preferred approach to akka-http seems
>>>> to be a throwback to one file programming. The main reason it seems this
>>>> way is the DSL. Take this example from a tutorial:
>>>>
>>>> path("bank" / IntNumber) { id =>
>>>> get {
>>>> complete {
>>>> getById(id).map{result =>
>>>> if(result.isDefined)
>>>> HttpResponse(entity =write(result.get))
>>>> else
>>>> HttpResponse(entity ="This bank does not exist")
>>>> }
>>>>
>>>>
>>>> }
>>>> }
>>>> }~
>>>> path("bank" / "update") {
>>>> post {
>>>> entity(as[String]) { bankJson =>
>>>> complete {
>>>> val bank =parse(bankJson).extract[Bank]
>>>> update(bank).map{result => HttpResponse(entity ="Bank
>>>> has been updated successfully")}
>>>> }
>>>> }
>>>> }
>>>> }
>>>> }
>>>> }
>>>>
>>>>
>>>> Simple enough right? Too me I see the start of an anti-pattern but lets
>>>> look further. It gets worse though, quickly, as shown in the akka-http
>>>> documentation here
>>>> <http://doc.akka.io/docs/akka-http/10.0.0/scala/http/routing-dsl/index.html#longer-example>.
>>>>
>>>> Still not bothered? The problem is that these examples are shallow and not
>>>> rooted in the real world. In the bank application above would be hundreds
>>>> of endpoints and each endpoint would have to validate data send by the
>>>> user, check to see whether that data correct against the database and a
>>>> dozen other things that would alter the nature of the return type to a
>>>> bad
>>>> request or internal error. The banking app would also have to log the
>>>> problems so forensics can be done on malicious users. Just taking one
>>>> route
>>>> "deposit" would be several hundred lines of code INSIDE the route.
>>>> However,
>>>> it seems that there is no way to break off the route, offload it to
>>>> another
>>>> component (such as a Per Request Actor) and then continue the DSL where
>>>> you
>>>> left off. I had the chance to see for an app in another company that was
>>>> asking my advice and their route is 12k lines long and at one point nested
>>>> 30 levels deep.
>>>>
>>>> Now I know what you might say, "But Robert, you can break up the route
>>>> into multiple files" which is true but something has to manually
>>>> concatenate all of those routes together and they cant be done off of the
>>>> main route. once you are in the routes DSL you are stuck there. Sure, you
>>>> can call an actor with a future to do a completion but that actor itself
>>>> might return data that requires a different kind of completion based upon
>>>> certain criteria such as whether the user has had their account suspended.
>>>> So if your completions are diverse, how do you break up the route?
>>>>
>>>> Now if someone has answers to these issues I would love to hear them
>>>> but after researching I found that basically PRA's are deprecated in favor
>>>> of a "convenient" DSL that entraps the user. For my purposes I opted to go
>>>> with the low level API and factor off the route dispatching to a routing
>>>> actor (yes, I know this is what the materializer does) and then just pull
>>>> out route data the old fashioned way. My router, does path checking and
>>>> then dispatches to another actor to handle that specific request and then
>>>> sends the HttpResponse entity back to the sender which completes the ask
>>>> and the route. My startup looks like this:
>>>>
>>>> val serverSource: Source[Http.IncomingConnection, Future[Http.
>>>> ServerBinding]] =
>>>> Http().bind(interface = "localhost", port = 8080)
>>>> log.info("Server online at http://localhost:8080";)
>>>> val bindingFuture: Future[Http.ServerBinding] =
>>>> serverSource.to(Sink.foreach { connection => // foreach
>>>> materializes the source
>>>> import akka.pattern.ask
>>>> println("Accepted new connection from " + connection.
>>>> remoteAddress)
>>>> connection.handleWithAsyncHandler(request => (httpRouter ?
>>>> request).mapTo[HttpResponse], parallelism = 4)
>>>> }).run()
>>>>
>>>>
>>>> A snippet of the router looks like this.
>>>>
>>>> class HttpRequestRouter extends Actor {
>>>> protected val log = Logging(context.system, this)
>>>>
>>>>
>>>> override def receive: Receive = {
>>>> case request: HttpRequest =>
>>>> val requestId = UUID.randomUUID()
>>>> request match {
>>>> case HttpRequest(GET, Uri.Path("/"), _, _, _) =>
>>>> notFound(requestId, request) // todo Implement this
>>>> case HttpRequest(POST, Uri.Path("/hello"), _, _, _) =>
>>>> invokeActor(classOf[HelloActor], requestId, request)
>>>> case HttpRequest(GET, Uri.Path("/users"), _, _, _) =>
>>>> invokeActor(classOf[ListUsersActor], requestId, request)
>>>> case HttpRequest(GET, Uri.Path("/addUser"), _, _, _) =>
>>>> invokeActor(classOf[AddUserActor], requestId, request)
>>>> case uri =>
>>>> notFound(requestId, request)
>>>> }
>>>> case msg => log.warning("Received unknown message: {}", msg)
>>>> }
>>>>
>>>>
>>>> private def invokeActor(actorType: Class[_], requestId: UUID, request
>>>> : HttpRequest) = {
>>>> context.actorOf(Props(actorType, sender(), requestId, request),
>>>> requestId.toString)
>>>> }}
>>>>
>>>> This allows me to fork off PRAs as needed but it kind of stinks in one
>>>> way because there are a lot of tools in the DSL for unpacking entities and
>>>> so on that I cant use, or rather if there is a way I and neither I nor
>>>> anyone within the reach of google has figured it out.
>>>>
>>>> So what am I missing? Do people really love this monstrous DSL even
>>>> though in a 100 endpoint system the thing will be gargantuan? Is there a
>>>> means to fork off at any point in the DSL and then "reboot the stream"? It
>>>> would be nice if some of the DSL tools could be invoked arbitrarily inside
>>>> the PRAs on the request object like.
>>>> class OrderPRA(replyTo: ActorRef, requestId: UUID, request: HttpRequest
>>>> ) {
>>>>
>>>> // ... code
>>>> sender.tell(withRequest(request) {
>>>> entity(as[Order]) { order =>
>>>> complete {
>>>> // ... write order to DB
>>>> "Order received"
>>>> }
>>>> }
>>>> }), self)
>>>> }
>>>>
>>>>
>>>> Opinions? Thoughts?
>>>>
>>>>
>>>>
>>>>
>>>> --
>> >>>>>>>>>> Read the docs: http://akka.io/docs/
>> >>>>>>>>>> Check the FAQ:
>> http://doc.akka.io/docs/akka/current/additional/faq.html
>> >>>>>>>>>> Search the archives: https://groups.google.com/group/akka-user
>> ---
>> You received this message because you are subscribed to the Google Groups
>> "Akka User List" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to akka-user+...@googlegroups.com <javascript:>.
>> To post to this group, send email to akka...@googlegroups.com
>> <javascript:>.
>> Visit this group at https://groups.google.com/group/akka-user.
>> For more options, visit https://groups.google.com/d/optout.
>>
>
>
--
>>>>>>>>>> Read the docs: http://akka.io/docs/
>>>>>>>>>> Check the FAQ:
>>>>>>>>>> http://doc.akka.io/docs/akka/current/additional/faq.html
>>>>>>>>>> Search the archives: https://groups.google.com/group/akka-user
---
You received this message because you are subscribed to the Google Groups "Akka
User List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to akka-user+unsubscr...@googlegroups.com.
To post to this group, send email to akka-user@googlegroups.com.
Visit this group at https://groups.google.com/group/akka-user.
For more options, visit https://groups.google.com/d/optout.
