How to handle multiple Promises in an (akka) Actor? - scala

I have an Akka actor responsible of handling http calls. I use scala dispatch to send multiple HTTP requests over an API:
urls.foreach { u
val service = url(u)
val promise = Http(service OK as.String).either
for(p <- promise)
{
p match
{
case Left(error) =>
faultHandler(error)
case Right(result) =>
resultHandler(result)
}
}
In the resultHandlerfunction, I increment an instance variable nbOfResults and compare to the number of calls I have done.
def resultHandler(result:String)
{
this.nbOfResults++
...
if(nbOfResults == nbOfCalls)
// Do something
}
Is it safe ? May the nbOfResultsvaraible be accessed at the same time if two calls return their results simultaneously ?
For now, I believed that the actor is more or less equivalent to a thread and therefore the callback functions are not executed concurrently. Is it correct ?

Here is a variant of Alexey Romanov response using only dispatch :
//Promises will be of type Array[Promise[Either[Throwable, String]]]
val promises = urls.map { u =>
val service = url(u)
Http(service OK as.String).either
}
//Http.promise.all transform an Iterable[Promise[A]] into Promise[Iterable[A]]
//So listPromise is now of type Promise[Array[Either[Throwable, String]]]
val listPromise = Http.promise.all(promises)
for (results <- listPromise) {
//Here results is of type Array[Either[Throwable, String]]
results foreach { result =>
result match {
Left(error) => //Handle error
Right(response) => //Handle response
}
}
}

There is a far better way:
val promises = urls.map {u =>
val service = url(u)
val promise = Http(service OK as.String).either
}
val listPromise = Future.sequence(promises)
listPromise.onComplete { whatever }

I agree with Alexey Romanov on his answer. Whatever way you choose to synchronize your http requests beware of the way your are processing the promises completion. Your intuition is correct in that concurrent access may appear on the state of the actor. The better way to handle this would be to do something like this:
def resultHandler(result: String) {
//on completion we are sending the result to the actor who triggered the call
//as a message
self ! HttpComplete(result)
}
and in the actor's receive function:
def receive = {
//PROCESS OTHER MESSAGES HERE
case HttpComplete(result) => //do something with the result
}
This way, you make sure that processing the http results won't violate the actor's state from the exterior, but from the actor's receive loop which is the proper way to do it

val nbOfResults = new java.util.concurrent.atomic.AtomicInteger(nbOfCalls)
// After particular call was ended
if (nbOfResults.decrementAndGet <= 0) {
// Do something
}
[EDIT] Removed old answer with AtomicReference CAS - while(true), compareAndSet, etc

Related

Avoid syncronous calls in Akka Actors due of Await result

Because I do some "complex" operations, I think my Actor became asyncronous. The main problem I think is that I use Await.result inside the method which return responses.
actor:
def process(subscribers: Set[ActorRef]): Receive = {
case Join(ref) => context become process(subscribers + ref)
case Leave(ref) => context become process(subscribers - ref)
case Push(request) =>
val filteredSubscribers = (subscribers - sender())
.filter(s => exists(s, request)) // just some actor filters
filteredSubscribers.foreach { subscriber =>
// here I have a Map with each actor requests
val actorOptions = getActorOptions(subscriber)
subscriber ? getResponse(actorOptions, request)
}
}
The problem is inside getResponse (I think).
getResponse(actorOptions: JsValue, request: SocketRequest): JsValue = {
(actorOptions \ "dashboardId").asOpt[Int] match {
case Some(id) => {
val response = widgetsService.getByDashboadId(id) map { widgets =>
val widgetsResponse: List[Future[String]] = widgets.map(w => {
widgetsService.getDataById(w.id) map {
data => s"""{ "widgetId": ${w.id}, "data": $data }"""
}
})
var responses: List[String] = List.empty
widgetsResponse.foreach(f => {
f.onComplete {
case Success(value) => responses = value :: responses
case Failure(e) => println(s"Something happened: ${e.getMessage}")
}
})
// first time when I use Await.result
// used to populate the responses list with data from all futures
Await.result(Future.sequence(widgetsResponse), Duration.Inf)
Json.parse(s"""{
"dashboardId": $id,
"widgets": [${response.mkString(", ")}]
}""".stripMargin)
}
// second time when I use Await.result
// used to return a JsValue instead of a Future[JsValue]
Await.result(response, Duration.Inf)
}
case None => buildDefaultJson // return default json value, unimportant for this example
}
}
Due of that, In frontend, if I have 2 sockets clients, the response for the second will be send only after first.
I found that I can obtain a "fake" increase of performance if I embrance the getResponse in a future inside of my Actor.
filteredSubscribers.foreach { subscriber =>
val actorOptions = getActorOptions(subscriber)
Future(subscriber ? getResponse(actorOptions, request))
}
So, for both subscribers the action will be started in same time, but when the first will reach the Await.result, the second will be locked until first is done.
I need to avoid using Await.result there, but I don't know how to get the results of a list of futures, without using for-comprehension (because is a dynamically list) for first time where I use it.
Because Akka ask operator (?) return a Future[Any], I tried that my getResponse method to return directly a JsValue to be mapped then in Future[JsValue]. If I remove the second Await.result and my method will return Future[JsValue], then the actor will return a Future[Future[JsValue]] which I don't think is too right.
After some more researches and solutions found on so, my code become:
Future.sequence(widgetsResponse) map { responses =>
Json.parse(
s"""
|{
|"dashboardId": $id,
|"tableSourceId": $tableSourceId,
|"widgets": [ ${responses.mkString(", ")}]
|}""".stripMargin
)
}
getResponse returns a Future[JsValue] now, removing both Await.result cases, and actor case become:
filteredSubscribers.foreach { subscriber =>
val actorOptions = getActorOptions(subscriber)
getResponse(actorOptions, request) map { data =>
subscriber ? data
}
}
I don't know why, still have a synchronous behavior. Damn, this can be due of my subscribers type: Set[ActorRef]? I tried to use parallel foreach and this looks like solving my problem:
filteredSubscribers.par.foreach { subscriber =>
val actorOptions = getActorOptions(subscriber)
getResponse(actorOptions, request) map { data =>
subscriber ? data
}
}

How should I wait for a scala dispatch (0.11.0) http get request to be completed?

I'm using the scala dispatch (0.11.0) library to send an HTTP GET request to a remote server. I want to wait for the response before executing the code which follows the request.
My request is of the form :
val req = :/("myurl.com") <:< myheaders OK as.Response(identity)
If I write :
val future = http(req)
future()
var res: String = null
future onComplete {
case Success(r) => res = r.getResponseBody
case _ => println("KO")
}
println(res)
I get null.
This is also the case if I write :
val future = http(req)
var res: String = null
while (!future.isCompleted) {
Thread.sleep(1000)
}
future onComplete {
case Success(r) => res = r.getResponseBody
case _ => println("KO")
}
println(res)
But with the following code :
val future = http(req)
var res: String = null
future onComplete {
case Success(r) => res = r.getResponseBody
case _ => println("KO")
}
while (!future.isCompleted) {
Thread.sleep(1000)
}
println(res)
I get the expected response.
Does someone understand this ?
It seems to me that calling Thread.sleep is not a good thing, could someone give me a hint on how I should handle this problem correctly ?
EDIT: #Randal Schulz thank you for your help, but as you posted in comments, I can't validate your answer.
As my problem was to wait (and do nothing else) until I get a valid response to an HTTP GET request, I think a satisfying way to do it is using Await.result. I removed the side-effect from my code. I used the option method to deal with Future failures (as I was interested only in successes), and I dealt with time out exception the classic way.
I think I could do it as wheaties mentionned, staying in Future, but I need more practice...
TL;DR
The best advice I can give you for working in an asynchronous work flow is that what goes into a Future stays in a Future.
Answer
The issue is that you have no idea when the Future will complete so if you want to use an asynchronous process, you're going to have to write in an asynchronous manner. The code as you have written never stops or blocks on the Future you create so the minute it creates the Future and hand it off to another thread, the current thread is free to then evaluate the res variable.
Hence, place most of what you're doing in a flow like follows:
myFuture map (func1) map (func2) map (func3) onComplete{
case Success(value) => println(value.getResponseBody)
case _ => println('KO')
}
Don't attempt to access something via side-effect like you are.
If you're really clever and you have several Future you can compose them:
val f1 = myFuture map(func1)
val f2 = myOtherFuture map(func2) map (func3)
val f3 = for{
v1 <- f1
v2 <- f2
} yield functionTakingBoth(v1, v2)
f3 onComplete{
//and do stuff here
}
I finally managed to write what I wanted using futures :
def loop(): Future[String] = {
val future = http(req).option
future flatMap ((x: Option[Response]) => x match {
case Some(rep) => rep.getResponseBody
case None => loop()
}
}
Now I can use the result of this function without explicitely waiting for the response to come.

Akka: Send a future message to an Actor

I have the following code inside an Actor
def receive = {
case All() => {
val collection: BSONCollection = db("ping")
val future:Future[List[Ping]] = collection.find(BSONDocument()).cursor[Ping].toList()
val zender = sender
future onComplete {
case Success(list) => zender ! list
case Failure(throwable) => zender ! List()
}
}
}
I don't like how I have to use the onComplete function to send the result back to the sender actor. I'd like to know if it is possible to convert it into something like this:
def receive = {
case All() => {
val collection: BSONCollection = db("ping")
val future:Future[List[Ping]] = collection.find(BSONDocument()).cursor[Ping].toList()
"sender ! future" // one option
"future.map( list => sender ! list)" //Another option. I know it's not map, but maybe another function
}
}
I feel that this flows better with future chaining.
You can use the pipe pattern for that. Just import akka.pattern.pipe and then you'll be able to pipe messages from futures to actors with future pipeTo actor.
If you want to have an empty list when failure happens, you probably want to have chained calls of "recover" and "pipeTo".

Are promises flawed?

Like the author of this question I'm trying to understand the reasoning for user-visible promises in Scala 2.10's futures and promises.
Particularly, going again to the example from the SIP, isn't it completely flawed:
import scala.concurrent.{ future, promise }
val p = promise[T]
val f = p.future
val producer = future {
val r = produceSomething()
p success r
continueDoingSomethingUnrelated()
}
val consumer = future {
startDoingSomething()
f onSuccess {
case r => doSomethingWithResult()
}
}
I am imagining the case where the call to produceSomething results in a runtime exception. Because promise and producer-future are completely detached, this means the system hangs and the consumer will never complete with either success or failure.
So the only safe way to use promises requires something like
val producer = future {
try {
val r.produceSomething()
p success r
} catch {
case e: Throwable =>
p failure e
throw e // ouch
}
continueDoingSomethingUnrelated()
}
Which obviously error-prone and verbose.
The only case I can see for a visible promise type—where future {} is insufficient—is the one of the callback hook in M. A. D.'s answer. But the example of the SIP doesn't make sense to me.
This is why you rarely use success and failure unless you already know something is bulletproof. If you want bulletproof, this is what Try is for:
val producer = future {
p complete Try( produceSomething )
continueDoingSomethingUnrelated()
}
It doesn't seem necessary to throw the error again; you've already dealt with it by packing it into the answer to the promise, no? (Also, note that if produceSomething itself returns a future, you can use completeWith instead.)
Combinators
You can use Promise to build additional Future combinators that aren't already in the library.
"Select" off the first future to be satisfied. Return as a result, with the remainder of the Futures as a sequence: https://gist.github.com/viktorklang/4488970.
An after method that returns a Future that is completed after a certain period of time, to "time out" a Future: https://gist.github.com/3804710.
You need Promises to be able to create other combinators like this.
Adapt Callbacks
Use Promise to adapt callback-based APIs to Future-based APIs. For example:
def retrieveThing(key: String): Future[Thing] = {
val p = Promise[Thing]()
val callback = new Callback() {
def receive(message: ThingMessage) {
message.getPayload match {
case t: Thing =>
p success t
case err: SystemErrorPayload =>
p failure new Exception(err.getMessage)
}
}
}
thingLoader.load(key, callback, timeout)
p.future
}
Synchronizers
Build synchronizers using Promise. For example, return a cached value for an expensive operation, or compute it, but don't compute twice for the same key:
private val cache = new ConcurrentHashMap[String, Promise[T]]
def getEntry(key: String): Future[T] = {
val newPromise = Promise[T]()
val foundPromise = cache putIfAbsent (key, newPromise)
if (foundPromise == null) {
newPromise completeWith getExpensive(key)
newPromise.future
} else {
foundPromise.future
}
}
Promise has a completeWith(f: Future) method that would solve this problem by automatically handling the success/failure scenarios.
promise.completeWith( future {
r.produceSomething
})

Akka avoiding wrapping future when responding to non-Actor code

I'm making a small caching actor with Akka 2 and to make the actor not block I perform all calculations inside futures. However a problem is that this actor also need to interact with with code that is not itself in an actor, so that I need to use the "ask" pattern to get a value.
My question is, how do I avoid wrapping the Future of my calculation inside another Future when using the ask pattern?
For example
val f = myCache ? GetOrCalc("myKey", myCalculation) // this will be a Future[Future[...]] but I would like a Future[...]
// meanwhile, inside the actor
def receive = {
case GetOrCalc(key, calculation) =>
if (keyNotExists) sender ! Future { calculation() } // calculation() is long-running
else sender ! cacheMap(key)
}
Ideally I could use the Future.pipeTo function but I'm afraid this doesn't get counted as a "response" for non-actor code
This is the solution:
val f = myCache ? GetOrCalc("myKey", myCalculation)
def receive = {
case GetOrCalc(key, calculation) =>
if (keyNotExists) Future { calculation() } pipeTo sender
else sender ! cacheMap(key)
}
Send-And-Receive-Future">http://doc.akka.io/docs/akka/2.0.3/scala/actors.html#Ask_Send-And-Receive-Future
Add onComplete to the calculation future.
def receive = {
case GetOrCalc(key, calculation) =>
if (keyNotExists) // calculation() is long-running
Future { calculation() } onComplete {
case Right(result) => result match {
case Some(value) => sender ! value
case None => sender ! Status.Failure(new Exception("Cannot find the value"))
}
case Left(ex) =>
sender ! Status.Failure(ex)
}
else sender ! cacheMap(key)
}
And there is an article about using Akka to build a cache system. http://letitcrash.com/post/30509298968/case-study-an-auto-updating-cache-using-actors