Skip to content

Commit

Permalink
Merge pull request #856 from akka/wip-2.1-throttler
Browse files Browse the repository at this point in the history
2.1: contribution module throttle
  • Loading branch information
rkuhn committed Nov 6, 2012
2 parents 165d8d5 + d102868 commit 0dcd59f
Show file tree
Hide file tree
Showing 5 changed files with 562 additions and 0 deletions.
1 change: 1 addition & 0 deletions akka-contrib/docs/index.rst
Expand Up @@ -29,6 +29,7 @@ The Current List of Modules
.. toctree::

reliable-proxy
throttle

Suggested Way of Using these Contributions
------------------------------------------
Expand Down
60 changes: 60 additions & 0 deletions akka-contrib/docs/throttle.rst
@@ -0,0 +1,60 @@
Throttling Actor Messages
=========================

Introduction
------------

Suppose you are writing an application that makes HTTP requests to an external
web service and that this web service has a restriction in place: you may not
make more than 10 requests in 1 minute. You will get blocked or need to pay if
you don’t stay under this limit. In such a scenario you will want to employ
a *message throttler*.

This extension module provides a simple implementation of a throttling actor,
the :class:`TimerBasedThrottler`.


How to use it
-------------

You can use a :class:`TimerBasedThrottler` as follows:

.. includecode:: @contribSrc@/src/test/scala/akka/contrib/throttle/TimerBasedThrottlerSpec.scala#demo-code

Please refer to the ScalaDoc documentation for the details.


The guarantees
--------------

:class:`TimerBasedThrottler` uses a timer internally. When the throttler’s rate is 3 msg/s,
for example, the throttler will start a timer that triggers
every second and each time will give the throttler exactly three "vouchers";
each voucher gives the throttler a right to deliver a message. In this way,
at most 3 messages will be sent out by the throttler in each interval.

It should be noted that such timer-based throttlers provide relatively **weak guarantees**:

* Only *start times* are taken into account. This may be a problem if, for example, the
throttler is used to throttle requests to an external web service. If a web request
takes very long on the server then the rate *observed on the server* may be higher.
* A timer-based throttler only makes guarantees for the intervals of its own timer. In
our example, no more than 3 messages are delivered within such intervals. Other
intervals on the timeline, however, may contain more calls.

The two cases are illustrated in the two figures below, each showing a timeline and three
intervals of the timer. The message delivery times chosen by the throttler are indicated
by dots, and as you can see, each interval contains at most 3 point, so the throttler
works correctly. Still, there is in each example an interval (the red one) that is
problematic. In the first scenario, this is because the delivery times are merely the
start times of longer requests (indicated by the four bars above the timeline that start
at the dots), so that the server observes four requests during the red interval. In the
second scenario, the messages are centered around one of the points in time where the
timer triggers, causing the red interval to contain too many messages.

.. image:: throttler.png

For some application scenarios, the guarantees provided by a timer-based throttler might
be too weak. Charles Cordingley’s `blog post <http://www.cordinc.com/blog/2010/04/java-multichannel-asynchronous.html>`_
discusses a throttler with stronger guarantees (it solves problem 2 from above).
Future versions of this module may feature throttlers with better guarantees.
Binary file added akka-contrib/docs/throttler.png
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@@ -0,0 +1,296 @@
/**
* Copyright (C) 2009-2012 Typesafe Inc. <http://www.typesafe.com>
*/

package akka.contrib.throttle

import scala.concurrent.duration.{ Duration, FiniteDuration }
import scala.util.control.NonFatal
import scala.collection.immutable.{ Queue Q }
import akka.actor.{ ActorRef, Actor, FSM }
import Throttler._
import TimerBasedThrottler._
import java.util.concurrent.TimeUnit
import akka.AkkaException

/**
* Marker trait for throttlers.
*
* == Throttling ==
* A <em>throttler</em> is an actor that is defined through a <em>target actor</em> and a <em>rate</em>
* (of type [[akka.contrib.throttle.Throttler.Rate]]). You set or change the target and rate at any time through the `SetTarget(target)`
* and `SetRate(rate)` messages, respectively. When you send the throttler any other message `msg`, it will
* put the message `msg` into an internal queue and eventually send all queued messages to the target, at
* a speed that respects the given rate. If no target is currently defined then the messages will be queued
* and will be delivered as soon as a target gets set.
*
* A [[akka.contrib.throttle.Throttler]] understands actor messages of type
* [[akka.contrib.throttle.Throttler.SetTarget]], [[akka.contrib.throttle.Throttler.SetRate]], in
* addition to any other messages, which the throttler will consider as messages to be sent to
* the target.
*
* == Transparency ==
* Notice that the throttler `forward`s messages, i.e., the target will see the original message sender (and not the throttler) as the sender of the message.
*
* == Persistence ==
* Throttlers usually use an internal queue to keep the messages that need to be sent to the target.
* You therefore cannot rely on the throttler's inbox size in order to learn how much messages are
* outstanding.
*
* It is left to the implementation whether the internal queue is persisted over application restarts or
* actor failure.
*
* == Processing messages ==
* The target should process messages as fast as possible. If the target requires substantial time to
* process messages, it should distribute its work to other actors (using for example something like
* a `BalancingDispatcher`), otherwise the resulting system will always work <em>below</em>
* the threshold rate.
*
* <em>Example:</em> Suppose the throttler has a rate of 3msg/s and the target requires 1s to process a message.
* This system will only process messages at a rate of 1msg/s: the target will receive messages at at most 3msg/s
* but as it handles them synchronously and each of them takes 1s, its inbox will grow and grow. In such
* a situation, the target should <em>distribute</em> its messages to a set of worker actors so that individual messages
* can be handled in parallel.
*
* @see [[akka.contrib.throttle.TimerBasedThrottler]]
*/
trait Throttler { this: Actor }

/**
* Message types understood by [[akka.contrib.throttle.Throttler]]'s.
*
* @see [[akka.contrib.throttle.Throttler]]
* @see [[akka.contrib.throttle.Throttler.Rate]]
*/
object Throttler {
/**
* A rate used for throttling.
*
* There are some shorthands available to construct rates:
* {{{
* import java.util.concurrent.TimeUnit._
* import scala.concurrent.duration.{ Duration, FiniteDuration }
*
* val rate1 = 1 msgsPer (1, SECONDS)
* val rate2 = 1 msgsPer Duration(1, SECONDS)
* val rate3 = 1 msgsPer (1 seconds)
* val rate4 = 1 msgsPerSecond
* val rate5 = 1 msgsPerMinute
* val rate6 = 1 msgsPerHour
* }}}
*
* @param numberOfCalls the number of calls that may take place in a period
* @param duration the length of the period
* @see [[akka.contrib.throttle.Throttler]]
*/
case class Rate(val numberOfCalls: Int, val duration: FiniteDuration) {
/**
* The duration in milliseconds.
*/
def durationInMillis(): Long = duration.toMillis
}

/**
* Set the target of a [[akka.contrib.throttle.Throttler]].
*
* You may change a throttler's target at any time.
*
* Notice that the messages sent by the throttler to the target will have the original sender (and
* not the throttler) as the sender. (In Akka terms, the throttler `forward`s the message.)
*
* @param target if `target` is `None`, the throttler will stop delivering messages and the messages already received
* but not yet delivered, as well as any messages received in the future will be queued
* and eventually be delivered when a new target is set. If `target` is not `None`, the currently queued messages
* as well as any messages received in the the future will be delivered to the new target at a rate not exceeding the current throttler's rate.
*/
case class SetTarget(target: Option[ActorRef])

/**
* Set the rate of a [[akka.contrib.throttle.Throttler]].
*
* You may change a throttler's rate at any time.
*
* @param rate the rate at which messages will be delivered to the target of the throttler
*/
case class SetRate(rate: Rate)

import language.implicitConversions

/**
* Helper for some syntactic sugar.
*
* @see [[akka.contrib.throttle.Throttler.Rate]]
*/
implicit class RateInt(val numberOfCalls: Int) extends AnyVal {
def msgsPer(duration: Int, timeUnit: TimeUnit) = Rate(numberOfCalls, Duration(duration, timeUnit))
def msgsPer(duration: FiniteDuration) = Rate(numberOfCalls, duration)
def msgsPerSecond = Rate(numberOfCalls, Duration(1, TimeUnit.SECONDS))
def msgsPerMinute = Rate(numberOfCalls, Duration(1, TimeUnit.MINUTES))
def msgsPerHour = Rate(numberOfCalls, Duration(1, TimeUnit.HOURS))
}

}

/**
* Implementation-specific internals.
*/
object TimerBasedThrottler {
private[throttle] case object Tick

// States of the FSM: A `TimerBasedThrottler` is in state `Active` iff the timer is running.
private[throttle] sealed trait State
private[throttle] case object Idle extends State
private[throttle] case object Active extends State

// Messages, as we queue them to be sent later
private[throttle] case class Message(message: Any, sender: ActorRef)

// The data of the FSM
private[throttle] sealed case class Data(target: Option[ActorRef],
callsLeftInThisPeriod: Int,
queue: Q[Message])
}

/**
* A [[akka.contrib.throttle.Throttler]] that uses a timer to control the message delivery rate.
*
* ==Example==
* For example, if you set a rate like "3 messages in 1 second", the throttler
* will send the first three messages immediately to the target actor but will need to impose a delay before
* sending out further messages:
* {{{
* // A simple actor that prints whatever it receives
* val printer = system.actorOf(Props(new Actor {
* def receive = {
* case x => println(x)
* }
* }))
* // The throttler for this example, setting the rate
* val throttler = system.actorOf(Props(new TimerBasedThrottler(3 msgsPer (1.second))))
* // Set the target
* throttler ! SetTarget(Some(printer))
* // These three messages will be sent to the printer immediately
* throttler ! "1"
* throttler ! "2"
* throttler ! "3"
* // These two will wait at least until 1 second has passed
* throttler ! "4"
* throttler ! "5"
* }}}
*
* ==Implementation notes==
* This throttler implementation internally installs a timer that repeats every `rate.durationInMillis` and enables `rate.numberOfCalls`
* additional calls to take place. A `TimerBasedThrottler` uses very few system resources, provided the rate's duration is not too
* fine-grained (which would cause a lot of timer invocations); for example, it does not store the calling history
* as other throttlers may need to do.
*
* However, a `TimerBasedThrottler` only provides ''weak guarantees'' on the rate (see also
* <a href='http://letitcrash.com/post/28901663062/throttling-messages-in-akka-2'>this blog post</a>):
*
* - Only ''delivery'' times are taken into account: if, for example, the throttler is used to throttle
* requests to an external web service then only the start times of the web requests are considered.
* If a web request takes very long on the server then more than `rate.numberOfCalls`-many requests
* may be observed on the server in an interval of duration `rate.durationInMillis()`.
* - There may be intervals of duration `rate.durationInMillis()` that contain more than `rate.numberOfCalls`
* message deliveries: a `TimerBasedThrottler` only makes guarantees for the intervals
* of its ''own'' timer, namely that no more than `rate.numberOfCalls`-many messages are delivered within such intervals. Other intervals on the
* timeline may contain more calls.
*
* For some applications, these guarantees may not be sufficient.
*
* ==Known issues==
*
* - If you change the rate using `SetRate(rate)`, the actual rate may in fact be higher for the
* overlapping period (i.e., `durationInMillis()`) of the new and old rate. Therefore,
* changing the rate frequently is not recommended with the current implementation.
* - The queue of messages to be delivered is not persisted in any way; actor or system failure will
* cause the queued messages to be lost.
*
* @see [[akka.contrib.throttle.Throttler]]
*/
class TimerBasedThrottler(var rate: Rate) extends Actor with Throttler with FSM[State, Data] {
startWith(Idle, Data(None, rate.numberOfCalls, Q()))

// Idle: no messages, or target not set
when(Idle) {
// Set the rate
case Event(SetRate(rate), d)
this.rate = rate
stay using d.copy(callsLeftInThisPeriod = rate.numberOfCalls)

// Set the target
case Event(SetTarget(t @ Some(_)), d) if !d.queue.isEmpty
goto(Active) using deliverMessages(d.copy(target = t))
case Event(SetTarget(t), d)
stay using d.copy(target = t)

// Queuing
case Event(msg, d @ Data(None, _, queue))
stay using d.copy(queue = queue.enqueue(Message(msg, context.sender)))
case Event(msg, d @ Data(Some(_), _, Seq()))
goto(Active) using deliverMessages(d.copy(queue = Q(Message(msg, context.sender))))
// Note: The case Event(msg, t @ Data(Some(_), _, _, Seq(_*))) should never happen here.
}

when(Active) {
// Set the rate
case Event(SetRate(rate), d)
this.rate = rate
// Note: this should be improved (see "Known issues" in class comments)
stopTimer()
startTimer(rate)
stay using d.copy(callsLeftInThisPeriod = rate.numberOfCalls)

// Set the target (when the new target is None)
case Event(SetTarget(None), d)
// Note: We do not yet switch to state `Inactive` because we need the timer to tick once more before
stay using d.copy(target = None)

// Set the target (when the new target is not None)
case Event(SetTarget(t @ Some(_)), d)
stay using d.copy(target = t)

// Tick after a `SetTarget(None)`: take the additional permits and go to `Idle`
case Event(Tick, d @ Data(None, _, _))
goto(Idle) using d.copy(callsLeftInThisPeriod = rate.numberOfCalls)

// Period ends and we have no more messages: take the additional permits and go to `Idle`
case Event(Tick, d @ Data(_, _, Seq()))
goto(Idle) using d.copy(callsLeftInThisPeriod = rate.numberOfCalls)

// Period ends and we get more occasions to send messages
case Event(Tick, d @ Data(_, _, _))
stay using deliverMessages(d.copy(callsLeftInThisPeriod = rate.numberOfCalls))

// Queue a message (when we cannot send messages in the current period anymore)
case Event(msg, d @ Data(_, 0, queue))
stay using d.copy(queue = queue.enqueue(Message(msg, context.sender)))

// Queue a message (when we can send some more messages in the current period)
case Event(msg, d @ Data(_, _, queue))
stay using deliverMessages(d.copy(queue = queue.enqueue(Message(msg, context.sender))))
}

onTransition {
case Idle -> Active startTimer(rate)
case Active -> Idle stopTimer()
}

initialize

private def startTimer(rate: Rate) = setTimer("morePermits", Tick, rate.duration, true)
private def stopTimer() = cancelTimer("morePermits")

/**
* Send as many messages as we can (while respecting the rate) to the target and
* return the state data (with the queue containing the remaining ones).
*/
private def deliverMessages(data: Data): Data = {
val queue = data.queue
val nrOfMsgToSend = scala.math.min(queue.length, data.callsLeftInThisPeriod)

queue.take(nrOfMsgToSend).foreach(x data.target.get.tell(x.message, x.sender))

data.copy(queue = queue.drop(nrOfMsgToSend), callsLeftInThisPeriod = data.callsLeftInThisPeriod - nrOfMsgToSend)
}
}

0 comments on commit 0dcd59f

Please sign in to comment.