arrow-fx / arrow.fx / Schedule

Schedule

sealed class ~~Schedule~~<F, Input, Output> : ScheduleOf<F, Input, Output> Deprecated: The IO datatype and it’s related type classes will disappear in Arrow 0.13.0. All useful operations are offered directly over suspend functions by Arrow Fx Coroutines. https://arrow-kt.io/docs/fx/async/

Retrying and repeating effects

A common demand when working with effects is to retry or repeat them when certain circumstances happen. Usually, the retrial or repetition does not happen right away; rather, it is done based on a policy. For instance, when fetching content from a network request, we may want to retry it when it fails, using an exponential backoff algorithm, for a maximum of 15 seconds or 5 attempts, whatever happens first.

Schedule allows you to define and compose powerful yet simple policies, which can be used to either repeat or retry computation.

Schedule has been derived from scalaz zio’s Schedule datatype and has been adapted to kotlin.

The two core methods of running a schedule are:

• retry: The effect is executed once, and if it fails, it will be reattempted based on the scheduling policy passed as an argument. It will stop if the effect ever succeeds, or the policy determines it should not be reattempted again.
• repeat: The effect is executed once, and if it succeeds, it will be executed again based on the scheduling policy passed as an argument. It will stop if the effect ever fails, or the policy determines it should not be executed again. It will return the last internal state of the scheduling policy, or the error that happened running the effect.

Constructing a policy:

Because schedules are polymorphic over any F that is also a Monad, constructing a Schedule can sometimes mean having to explicitly write the type-parameters. This can be avoided using Schedule.withMonad which partially applies the chosen Monad.

Constructing a simple schedule which recurs 10 times until it succeeds:

import arrow.fx.ForIO
import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.monad.monad

fun <A> recurTenTimes() = Schedule.recurs<ForIO, A>(IO.monad(), 10)

A more complex schedule is best put together using the withMonad constructor:

import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.monad.monad
import arrow.fx.extensions.io.monadDefer.monadDefer
import arrow.fx.typeclasses.milliseconds
import arrow.fx.typeclasses.seconds

fun <A> complexPolicy() =
  Schedule.withMonad(IO.monad()) {
    exponential<A>(10.milliseconds).whileOutput { it.nanoseconds < 60.seconds.nanoseconds }
      .andThen(spaced<A>(60.seconds) and recurs<A>(100)).jittered(IO.monadDefer())
      .zipRight(identity<A>().collect())
  }

This policy will recur with exponential backoff as long as the delay is less than 60 seconds and then continue with a spaced delay of 60 seconds. The delay is also randomized slightly to avoid coordinated backoff from multiple services. Finally we also collect every input to the schedule and return it. When used with retry this will return a list of exceptions that occured on failed attempts.

Common use cases

Common use cases Once we have building blocks and ways to combine them, let’s see how we can use them to solve some use cases.

Repeating an effect and dealing with its result

When we repeat an effect, we do it as long as it keeps providing successful results and the scheduling policy tells us to keep recursing. But then, there is a question on what to do with the results provided by each iteration of the repetition.

There are at least 3 possible things we would like to do:

• Discard all results; i.e., return Unit.
• Discard all intermediate results and just keep the last produced result.
• Keep all intermediate results.

Assuming we have an effect in IO, and we want to repeat it 3 times after its first successful execution, we can do:

import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.concurrent.concurrent
import arrow.fx.extensions.io.monad.monad
import arrow.fx.fix
import arrow.fx.repeat

fun main() {
  var counter = 0
  val io = IO { println("Run: ${counter++}") }
  //sampleStart
  val res = io.repeat(IO.concurrent(), Schedule.recurs(IO.monad(), 3))
  //sampleEnd
  println(res.fix().unsafeRunSync())
}

However, when running this new effect, its output will be the number of iterations it has performed, as stated in the documentation of the function. Also notice that we did not handle the error case, there are overloads repeatOrElse and repeatOrElseEither which offer that capability, repeat will just rethrow any error encountered.

If we want to discard the values provided by the repetition of the effect, we can combine our policy with Schedule.unit, using the zipLeft or zipRight combinators, which will keep just the output of one of the policies:

import arrow.fx.ForIO
import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.concurrent.concurrent
import arrow.fx.extensions.io.monad.monad
import arrow.fx.fix
import arrow.fx.repeat

fun main() {
  var counter = 0
  val io = IO { println("Run: ${counter++}") }
  //sampleStart
  val res = io.repeat(IO.concurrent(), Schedule.unit<ForIO, Unit>(IO.monad()).zipLeft(Schedule.recurs(IO.monad(), 3)))

  // equal to
  val res2 = io.repeat(IO.concurrent(), Schedule.recurs<ForIO, Unit>(IO.monad(), 3).zipRight(Schedule.unit(IO.monad())))

  //sampleEnd
  println(res.fix().unsafeRunSync())
  println(res2.fix().unsafeRunSync())
}

Following the same strategy, we can zip it with the Schedule.identity policy to keep only the last provided result by the effect.

import arrow.fx.ForIO
import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.concurrent.concurrent
import arrow.fx.extensions.io.monad.monad
import arrow.fx.fix
import arrow.fx.repeat

fun main() {
  var counter = 0
  val io = IO { println("Run: ${counter++}"); counter }
  //sampleStart
  val res = io.repeat(IO.concurrent(), Schedule.identity<ForIO, Int>(IO.monad()).zipLeft(Schedule.recurs(IO.monad(), 3)))

  // equal to
  val res2 = io.repeat(IO.concurrent(), Schedule.recurs<ForIO, Int>(IO.monad(), 3).zipRight(Schedule.identity<ForIO, Int>(IO.monad())))

  //sampleEnd
  println(res.fix().unsafeRunSync())
  println(res2.fix().unsafeRunSync())
}

Finally, if we want to keep all intermediate results, we can zip the policy with Schedule.collect:

import arrow.fx.ForIO
import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.concurrent.concurrent
import arrow.fx.extensions.io.monad.monad
import arrow.fx.fix
import arrow.fx.repeat

fun main() {
  var counter = 0
  val io = IO { println("Run: ${counter++}"); counter }
  //sampleStart
  val res = io.repeat(IO.concurrent(), Schedule.collect<ForIO, Int>(IO.monad()).zipLeft(Schedule.recurs(IO.monad(), 3)))

  // equal to
  val res2 = io.repeat(IO.concurrent(), Schedule.recurs<ForIO, Int>(IO.monad(), 3).zipRight(Schedule.collect<ForIO, Int>(IO.monad())))

  //sampleEnd
  println(res.fix().unsafeRunSync())
  println(res2.fix().unsafeRunSync())
}

Repeating an effect until/while it produces a certain value

We can make use of the policies doWhile or doUntil to repeat an effect while or until its produced result matches a given predicate.

import arrow.fx.ForIO
import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.concurrent.concurrent
import arrow.fx.extensions.io.monad.monad
import arrow.fx.fix
import arrow.fx.repeat

fun main() {
  var counter = 0
  val io = IO { println("Run: ${counter++}"); counter }
  //sampleStart
  val res = io.repeat(IO.concurrent(), Schedule.doWhile<ForIO, Int>(IO.monad()) { it <= 3 })
  //sampleEnd
  println(res.fix().unsafeRunSync())
}

Exponential backoff retries

A common algorithm to retry effectful operations, as network requests, is the exponential backoff algorithm. There is a scheduling policy that implements this algorithm and can be used as:

import arrow.fx.ForIO
import arrow.fx.IO
import arrow.fx.Schedule
import arrow.fx.extensions.io.monad.monad
import arrow.fx.typeclasses.milliseconds

val exponential = Schedule.exponential<ForIO, Unit>(IO.monad(), 250.milliseconds)

Types

Functions

Companion Object Functions

Extension Functions

Companion Object Extension Functions