ZIO Schedule: Conquering Flakiness & Recurrence with Pure Functional Programming

ZIO Schedule John A. De Goes — @jdegoes - http://degoes.net Scala in the City
ZIO Schedule
Scala in the City, October 2018 - London, UK
John A. De Goes
@jdegoes - http://degoes.net

What is ZIO? Retries Repeats Schedule Composition

ZIO
ZIO lets you build high-performance,
type-safe, concurrent, asynchronous
applications that don’t leak resources,
and are easy to reason about
compositionally, test, and refactor.

github.com/scalaz/scalaz-zio
ZIO

def main: Unit = {
println("Hello, what is your name?")
val name = readLine()
println(s"Good morning, $name!")
}
Second-Class Effects
✗ Pass to functions
✗ Return from functions
✗ Store in data structures
✗ Async/sync/concurrency
✗ Async/sync resource-safety

def main: IO[IOException, Unit] = for {
_ <- putStrLn("Hello, what is your name?")
name <- getStrLn
_ <- putStrLn(s"Good morning, $name!")
} yield ()
First-Class Effects
✓ Pass to functions
✓ Return from functions
✓ Store in data structures
✓ Async/sync/concurrency
✓ Async/sync resource-safety

IO[E, A]
IO[E, A] is an immutable value that
describes an effectful program, which
may fail with a value of type E, or return
a value of type A.
ZIO IO

IO[Nothing, A]
Never fails
ZIO IO

IO[E, Nothing]
Never returns
ZIO IO

IO[Nothing, Nothing]
Never fails or returns
ZIO IO

for {
queue <- Queue.unbounded[String]
worker = queue.take.flatMap(putStrLn).forever
workers10k = List.fill(10000)(worker)
_ <- IO.forkAll(workers10k)
_ <- queue.offer("Chocolate!").forever.fork
} yield ()
Lightweight “Threads”
✓ Massive scalability over threads
✓ Non-blocking
✓ Safe concurrency
✓ Automatically interruptible
✓ Always resource-safe
✓ Powerful composition

for {
resp <- api.getUser(userId)
_ <- db.updateUser(userId, resp.profile)
_ <- promise.complete(())
} yield ()
✓ Non-blocking
Asynchronicity

requests.parTraverse { request =>
for {
product <- findProduct(request.id)
update <- applyDiscount(product.id, discount)
_ <- db.updateProduct(product.id, update)
} yield product.id
}
✓ Non-blocking
Concurrency

def fib(n: Int): IO[Nothing, Int] =
if (n <= 1) IO.now(n)
else fib(n-1).parWith(fib(n-2))(_ + _)
fib(Int.MaxValue).fork(_.interrupt)
✓ Non-blocking
Interruption

openFile.bracket(closeFile(_)) { handle =>
def loop(ref: Ref[Chunk[Byte]]) =
for {
chunk <- handle.readChunk
<- ref.update(_ ++ chunk)
all <- if (handle.hasMore) loop(ref)
else ref.get
} yield all
Ref(Chunk.empty).flatMap(loop)
}
✓ Non-blocking
Bracket

def webCrawler[E: Monoid, A: Monoid](seeds: Set[URL],
router: URL => Set[URL],
processor: (URL, String) => IO[E, A]):
IO[Nothing, Crawl[E, A]] = {
def loop(seeds: Set[URL], ref: Ref[CrawlState[E, A]]): IO[Nothing, Unit] =
ref.update(_ |+| CrawlState.visited(seeds)) *> IO.parTraverse(seeds)(
seed =>
getURL(seed).redeem(_ => IO.unit,
html => for {
visited <- ref.get.map(_.visited)
seeds <- IO.now(extractURLs(seed, html).toSet
.flatMap(router) -- visited)
crawl <- processor(seed, html).redeemPure(Crawl(_, mzero[A]),
Crawl(mzero[E], _))
_ <- ref.update(_ |+| CrawlState.crawled(crawl))
_ <- loop(seeds, ref)
} yield ())).void
Ref(mzero[CrawlState[E, A]]).flatMap(ref => loop(seeds, ref).map(_ =>
ref.get.map(_.crawl)))
}
✓ Non-blocking
Composable

Retries

App Web API

App Web API
408

App Web API
408429

App Web API
408429
423

App Web API
408429
423
500

App Web API
408429
423
500 503

App Web API
408429
423
500 503
504

AppDatabase
Cache
Local
Storage
Cloud
Storage
Streaming Logs
Analytics
Web APIs

def networkRequest(): A = ???
def networkRequestWithRetries(max: Int, millis: Long): A = {
var i = 0
while (i < max) {
try {
return networkRequest()
}
catch {
case _ : Exception =>
i = i + 1
Thread.sleep(millis)
}
}
}
Retrying Synchronously

Repeats

App
Generate
Report

App
Generate
Report
Every
user

App
Generate
Report
Every
user
Every
day

App
Generate
Report
Every
user
Every
day
Every
account

App
Generate
Report
Every
user
Every
day
Every
account
Every
week

AppService
Heartbeat
Data
Retrieval
Remote
Uploading
Report
Generation
Log Rotation File Cleanup
User
Reminders
Garbage
Collection

def reportGen(): A = ???
def repeatedReportGen(max: Int, millis: Long): List[A] = {
var list = List.empty[A]
var i = 0
while (i < max) {
list = reportGen() :: list
Thread.sleep(millis)
i = i + 1
}
list
}
Repeating Synchronously

Schedule

ZIO SCHEDULE
Schedule[A, B]
Schedule[A, B] is an immutable
value that describes an effectful
schedule, which after consuming an A,
produces a B and decides to halt or
continue after some delay d.

ZIO SCHEDULE
Schedule[A, B]
A
B
B
Continue?
Continue after delay d
Halt immediately

Should I repeat a
program that failed
with an E?
Should I repeat a
program that
returned an A?
Retries Repeats

Retries Repeats
Schedule[E, B] Schedule[A, B]
Retry policy
Consumes errors of type E
Emits values of type B
Repeat policy
Consumes return values of type
A
Emits values of type B

Schedule.never : Schedule[Any, Nothing]
Zero Recurrences
Accepts any input
Never recurs, never emitting a value

Schedule.once : Schedule[Any, Unit]
One Recurrence
Accepts any input
Recurs once, emitting unit

Schedule.forever : Schedule[Any, Int]
Infinite Recurrences
Consumes any input
Recurs forever without delay, emitting the number of recurrences so far

Schedule.recurs(n: Int) : Schedule[Any, Int]
Fixed Recurrences
Accepts any input
Recurs the specified number of times, emitting number of recurrences so far

durationtask task
Schedule.spaced(d: Duration) : Schedule[Any, Int]
Spaced Recurrences
Accepts any input
Recurs forever with delay, emitting the number of recurrences so far

duration
task task
Schedule.fixed(d: Duration) : Schedule[Any, Int]
Fixed Recurrences
Accepts any input
Recurs forever with a delay, emitting number of recurrences so far
duration

Schedule.exponential(d: Duration, f: Double = 2.0) : Schedule[Any, Duration]
Exponential Recurrences
Accepts any input
Recurs forever with delay, emitting the current delay between steps
Scaling factor
Starting duration

Schedule.doWhile[A](f: A => Boolean): Schedule[A, A]
Conditional Recurrences
Accepts an A
Recurs while the predicate is true without delay, emitting the same A
Recurrence condition

Schedule.doUntil[A](f: A => Boolean): Schedule[A, A]
Conditional Recurrences
Accepts an A
Recurs until the predicate is true without delay, emitting the same A
Recurrence condition

Schedule.collect: Schedule[A, List[A]]
Collecting Recurrences
Recurs forever without delay, emitting a list of all consumed A’s
Consumes an A

Schedule.identity[A]: Schedule[A, A]
Identity Recurrences
Recurs forever without delay, emitting the same A
Consumes an A

Schedule.unfold[A](a: => A)(f: A => A): Schedule[Any, A]
Unfold Recurrences
Recurs forever without delay, emitting an A by unfolding the seed through repeated application
Consumes any value

Schedule.point[A](a: => A): Schedule[Any, A]
Constant Output
Recurs forever without delay, emitting a constant
Consumes any value

Schedule.lift[A, B](f: A => B): Schedule[A, B]
Function Output
Recurs forever without delay, emitting the function applied to the input
Consumes an A

Retries Repeats
val action: IO[E, A] = ???
val policy: Schedule[E, B] = ???
val retried: IO[E, A] =
action retry policy
val policy: Schedule[A, B] = ???
val repeated: IO[E, B] =
action repeat policy

Retries Repeats
val orElse: (E, B) => IO[E2, A] = ???
val retried: IO[E2, A] =
action retryOrElse (policy, orElse)
val orElse: (E, Option[B]) => IO[E2, B] = ???
val repeated: IO[E2, B] =
action repeatOrElse (policy, onError)

Retries Repeats
val orElse: (E, B) => IO[E2, B] = ???
val retried: IO[E2, Either[B, A]] =
action retryOrElse0 (policy, orElse)
val orElse: (E, Option[B]) => IO[E2, C] = ???
val repeated: IO[E2, Either[C, B]] =
action repeatOrElse0 (policy, onError)

Composition

def networkRequestWithRetries(factor: Float = 1.5f, init: Int = 1, cur: Int = 0)
(implicit as: ActorSystem): Future[String] = {
networkRequest().recoverWith {
case NetworkException =>
val next: Int =
if (cur == 0) init
else Math.ceil(cur * factor).toInt
after(next.milliseconds, as.scheduler, global, Future.successful(1)).flatMap { _ =>
networkRequestWithRetries(factor, init, next)
}
case t: Throwable => throw t
}
}
Future Retries*
* https://hackernoon.com/exponential-back-off-with-scala-futures-7426340d0069

def retry[F[_]: Timer: RaiseThrowable, O](
fo : F[O],
delay : FiniteDuration,
nextDelay : FiniteDuration => FiniteDuration,
maxAttempts : Int,
retriable : Throwable => Boolean): Stream[F, O] = ???
FS2 Retry

Monix Retry & Repeat
Retries Repeats
val task : Task[A] = ???
val retries: Int = ???
val retried1: Task[A] =
task onErrorRestart (retries)
val pred: Throwable => Boolean = ???
val retried2: Task[A] =
task onErrorRestartIf (pred)
// onErrorRestartLoop
val task: Task[A] = ???
val pred: A => Boolean = ???
val repeated: Task[A] =
task restartUntil (pred)

ZIO SCHEDULE
Monoid
append
zero

ZIO SCHEDULE
Monoid
append
zero
Applicative
map
point
ap

ZIO SCHEDULE
Monoid
append
zero
Applicative
map
point
ap
Category
id
compose

ZIO SCHEDULE
Monoid
append
zero
Applicative
map
point
ap
Category
id
compose
Profunctor
lmap
rmap
dimap

ZIO SCHEDULE
Monoid
append
zero
Applicative
map
point
ap
Category
id
compose
Profunctor
lmap
rmap
dimap
Strong
first
second

ZIO SCHEDULE
Monoid
append
zero
Applicative
map
point
ap
Category
id
compose
Profunctor
lmap
rmap
dimap
Strong
first
second
Choice
left
right

(s1 : Schedule[A, B]) && (s2 : Schedule[A, C]) : Schedule[A, (B, C)]
Intersection
s1 s2 s1 && s2
Continue : Boolean b1 b2 b1 && b2
Delay : Duration d1 d2 d1.max(d2)
Emit : (A, B) a b (a, b)

(s1 : Schedule[A, B]) || (s2 : Schedule[A, C]) : Schedule[A, (B, C)]
Union
s1 s2 s1 || s2
Continue : Boolean b1 b2 b1 || b2
Delay : Duration d1 d2 d1.min(d2)
Emit : (A, B) a b (a, b)

(s1 : Schedule[A, B]) andThen (s2 : Schedule[A, C]) : Schedule[A, Either[B, C]]
Sequence
A
A
B
C
s1
s2
s1 andThen s2A Either[B, C]

(s1 : Schedule[A, B]) >>> (s2 : Schedule[B, C]) : Schedule[A, C]
Compose
A
B
B
C
s1
s2
s1 >>> s2A C

(s1 : Schedule[A, B]).jittered : Schedule[A, B]
Jittering
A AB Bs1 s1’
Delays randomly
jittered

(s1 : Schedule[A, B]).collect : Schedule[A, List[B]]
Collecting
A AB List[C]s1 s1’
Emissions collected

(s1 : Schedule[A, B]).whileInput(f: A => Boolean) : Schedule[A, B]
Filtering By Input
A AB Bs1 s1’
Continues while/until f returns true
(s1 : Schedule[A, B]).untilInput(f: A => Boolean) : Schedule[A, B]

(s1 : Schedule[A, B]).whileOutput(f: B => Boolean) : Schedule[A, B]
Filtering By Output
A AB Bs1 s1’
Continues while/until f returns true
(s1 : Schedule[A, B]).untilOutput(f: B => Boolean) : Schedule[A, B]

(s1 : Schedule[A, B]).map(f: B => C) : Schedule[A, C]
Mapping
A AB Cs1 s1’
B mapped to C

(s1 : Schedule[A, B]) *> (s2 : Schedule[A, C]) : Schedule[A, C]
(s1 && s2).map(_._2)
Left / Right Ap
(s1 : Schedule[A, B]) <* (s2 : Schedule[A, C]) : Schedule[A, B]
(s1 && s2).map(_._1)

(s1 : Schedule[A, B]).logInput(f: A => IO[Nothing, Unit]) : Schedule[A, B]
Logging Inputs
A AB Bs1 s1’
Inputs logged to f

(s1 : Schedule[A, B]).logOutput(f: B => IO[Nothing, Unit]) : Schedule[A, B]
Logging Outputs
A AB Bs1 s1’
Outputs logged to f

Crafting a ZIO Schedule
Produce a jittered schedule that first does exponential spacing (starting from 10
milliseconds), but then after the spacing reaches 60 seconds, switches over to fixed
spacing of 60 seconds between recurrences, but will only do that for up to 100
times, and emits a list of the collected inputs.

Schedule.exponential(10.milliseconds)

(Schedule.exponential(10.milliseconds)
andThen ???)

(Schedule.exponential(10.milliseconds).whileOutput(_ < 60.seconds)
andThen ???)

andThen
Schedule.fixed(60.seconds))

andThen
(Schedule.fixed(60.seconds) && Schedule.recurs(100))

andThen
(Schedule.fixed(60.seconds) && Schedule.recurs(100)).jittered

andThen
(Schedule.fixed(60.seconds) && Schedule.recurs(100)).jittered
*> Schedule.identity[A].collect

def customSchedule[A]: Schedule[A, List[A]] = {
import Schedule._
(exponential(10.millis).whileOutput(_ < 60.secs) andThen
(fixed(60.secs) && recurs(100)).jittered *> identity[A].collect
}

Github
https://github.com/scalaz/scalaz-zio
Microsite
https://scalaz.github.io/scalaz-zio/
ZIO Chat
https://gitter.im/scalaz/scalaz-zio/
Scalaz Chat
https://gitter.im/scalaz/scalaz/
That’s A Wrap!

Functional Scala by John A. De Goes
Local Remote Date
London London Time Oct 21 - 23
Paris Paris Time Oct 25 - 27
SF / SV Pacific Coast Time Nov 18 - 21
That’s A Wrap!

THANK YOU!
@jdegoes - http://degoes.net
That’s A Wrap!

ZIO Schedule: Conquering Flakiness & Recurrence with Pure Functional Programming

More Related Content

What's hot (20)

Similar to ZIO Schedule: Conquering Flakiness & Recurrence with Pure Functional Programming (20)

More from John De Goes (20)

Recently uploaded (20)

ZIO Schedule: Conquering Flakiness & Recurrence with Pure Functional Programming