Funtional Concurrency in Scala 101

Piotr Gawryś | twitter.com/p_gawrys

Slides: https://slides.com/avasil/funtional-concurrency-in-scala-101#/

Plan

Brief Overview
Introduction to Functional Programming
Introduction to Concurrency (with FP!)
Exercises

Schedule


9.00 - 10:30	workshop
10:30 - 11:00	break
11:00 - 12:30	workshop
12:30 - 14:00	lunch break
14:00 - 15:30	workshop
15:30 - 16:00	break
16:00 - 17:00	workshop

What's ahead

Couple of concurrent puzzles for a warm up
Using newly acquired skills in real use case

Battle City (Tank 1990)

Before we begin...

git clone git@github.com:Avasil/fp-concurrency-101.git
or https://github.com/Avasil/fp-concurrency-101.git

git checkout exercises

sbt exercises/compile
sbt client/fastOptJS
sbt server/compile

Functional Programming

Programming with pure, referentially transparent functions
- Replacing an expression by its bound value doesn't alter the behavior of your program
Lawful, algebraic structures

Benefits

Consistent, known behavior
Compositionality
Easy refactoring and optimization rules

What prevents composition?

Connected sequence
Leaky abstraction
Side effects!

Side effects

Source of example:

"Rúnar Óli Bjarnason - Composing Programs"

class Cafe {
    def buyCoffee(cc: CreditCard): Coffee = {
        val cup = new Coffee()
        cc.charge(cup.price)
        cup
    }
}

Side effects

Source of example:

"Rúnar Óli Bjarnason - Composing Programs"

class Cafe {
    def buyCoffee(cc: CreditCard): (Coffee, Charge) = {
        val cup = new Coffee()
        (cup, new Charge(cc, cup.price))
    }
}

So how to write pure functions?

val a: Int = 10
val b: Int = a + a // 20

val c: Int = 10 + 10 // 20

val r = scala.util.Random

val a: Int = r.nextInt // 7
val b: Int = a + a     // 7 + 7

val c: Int = r.nextInt + r.nextInt // 7 + 5

So how to write pure functions?

import monix.eval.Task
import cats.implicits._

val r = scala.util.Random
// when run, generates 7
val a: Task[Int] = Task(r.nextInt) 
a.runSyncUnsafe // 7

val b: Task[Int] = (a, a).mapN(_ + _) 
b.runSyncUnsafe // 7 + 5

val c: Task[Int] = (Task(r.nextInt), Task(r.nextInt)).mapN(_ + _) 
c.runSyncUnsafe // 7 + 5

val d: Task[Int] = a.map(x => x + x)
d.runSyncUnsafe // 7 + 7

Wait, but it still does side effect so what's the point?

Referential Transparency - Benefits

Local reasoning
Representing the program as value

Check out great explanations there:

https://www.reddit.com/r/scala/comments/8ygjcq/can_someone_explain_to_me_the_benefits_of_io/e2jfp9b/

Concurrency

Doing things interleaved

Parallelism

Doing things at the same time to finish faster

Credit: https://twitter.com/impurepics

Concurrency on JVM

Threads

Map 1:1 to OS native threads which means we can execute up to N threads in parallel (N = number of cores)

Context Switch

Before a thread can start doing work, the OS needs to store state of earlier task, restore the new one etc.

Concurrency on JVM

Thread Pools

Takes care of creating and reusing threads, distributing work

Blocking Threads

Nothing else can be run on blocked thread, resources are wasted.

Thread Scheduling

Preemptive multitasking

Tasks are interrupted to allow other tasks to run. Guarantees that each task will get its own "slice" of time.

Cooperative multitasking

Tasks voluntarily yield control to the scheduler.

Monix

Scala / Scala.js library for composing asynchronous programs
Lots of cool data types such as Task, Observable, ConcurrentQueue and many more
Purely functional API
Cats-Effect integration

Monix Task

Lazily evaluated
Lots of concurrency related features
Error Handling
Cancelation
Parallelism
Resource Safety
... and it composes like a charm!

Basic Usage

import monix.eval.Task

val task = Task { println("hello!") }

// nothing happens yet, it's just a description 
// of evaluation that will produce `Unit`, end
// with error or never complete
val program: Task[Unit] =
  for {
     _ <- task 
     _ <- task 
  } yield ()

// prints "hello!" twice
program.runSyncUnsafe()

import monix.eval.Task
import cats.syntax.flatmap._

val program: Task[Unit] = 
  taskA.flatMap(_ => taskB)

val program: Task[Unit] = 
  taskA >> taskB

import monix.eval.Task
import scala.concurrent.duration._

// waits 100 millis (without blocking any Thread!) 
// then prints "woke up"
Task.sleep(100.millis) >> Task(println("woke up"))


// never finishes
Task.never >> Task(println("woke up"))

Error Handling

import monix.eval.Task

val taskA = Task(println("A"))
val taskB = Task.raiseError(new Exception("B"))
val taskC = Task(println("C"))

val program: Task[Int] = 
  for {
    _ <- taskA 
    _ <- taskB // short circuits Task with error
    _ <- taskC // taskC will never happen!
  } yield 42

// A
// Exception in thread "main" java.lang.Exception: B
program.runSyncUnsafe

// A
// Process finished with exit code 0
val safeProgram: Task[Either[Throwable, Int]] = 
  program.attempt

safeProgram.runSyncUnsafe

// will print A, B and C
val program: Task[Int] =
  for {
    _ <- taskA
    // in case of error executes the Task in argument
    _ <- taskB.handleErrorWith(_ => Task(println("B")))
    _ <- taskC
  } yield 42

Error Handling

def retryBackoff[A](source: Task[A])
                   (maxRetries: Int, delay: FiniteDuration): Task[A] = {
  source.onErrorHandleWith { ex =>
    if (maxRetries > 0)
      retryBackoff(source)(maxRetries - 1, delay * 2)
        .delayExecution(delay)
    else
      Task.raiseError(ex)
  }
}

Parallelism

import monix.eval.Task
import cats.implicits._

val taskA = Task(println("A"))
val taskB = Task(println("B"))

val listOfTasks: List[Task[Unit]] = List(taskA, taskB)

val taskOfList: Task[List[Unit]] = Task.gather(listOfTasks)

// using Parallel type class instance from Cats
val parTask: Task[Unit] = 
  (taskA, taskB).parTupled

val taskInParallel: Task[Unit] =
  for {
    // run in "background"
    fibA <- taskA.start 
    fibB <- taskB.start
    // wait for ioa result
    _ <- fibA.join 
    _ <- fibB.join
  } yield ()


// run both in parallel 
// and cancel the loser
val raceTask: Task[Either[Unit, Unit]] = 
  Task.race(taskA, taskB)

Scheduler

Thread Pool on steroids
Configurable Execution Model
Tracing Scheduler (for TaskLocal support)
Test Scheduler that can simulate passing time

Scheduler

  import monix.execution.schedulers.TestScheduler
  import scala.concurrent.duration._

  def retryBackoff[A](source: Task[A])
                     (maxRetries: Int, delay: FiniteDuration): Task[A] = {
    source.onErrorHandleWith { ex =>
      if (maxRetries > 0)
        retryBackoff(source)(maxRetries - 1, delay * 2)
          .delayExecution(delay)
      else
        Task.raiseError(ex)
    }
  }

  val sc = TestScheduler()

  val failedTask: Task[Int] = Task.raiseError[Int](new Exception("boom"))
  val task: Task[Int] = retryBackoff(failedTask)(5, 2.hours)

  val f: CancelableFuture[Int] = task.runToFuture(sc)
  
  println(f.value) // None
  sc.tick(10.hours)
  println(f.value) // None
  sc.tick(1000.days)
  println(f.value) // Some(Failure(java.lang.Exception: boom))

Observable

Inspired by RxJava / ReactiveX
Push-based streaming with backpressure
(mostly) Purely functional API
Choose your adventure

See Alex Nedelcu (author of Monix) presentation for origins: https://monix.io/presentations/2018-tale-two-monix-streams.html

Observable

trait Observable[+A] {
  def subscribe(o: Observer[A]): Cancelable
}

trait Observer[-T] {
  def onNext(elem: T): Future[Ack]

  def onError(ex: Throwable): Unit

  def onComplete(): Unit
}

Observable

val list: Observable[Long] = 
    Observable.range(0, 1000)
      .take(100)
      .map(_ * 2)
       
val task: Task[Long] =
    list.foldLeftL(0L)(_ + _)

val task: Task[List[Int]] = 
  for {
    queue     <- ConcurrentQueue.unbounded[Task, Int]()
    // feeds queue in the background
    _         <- runProducer(queue).start 
    list      <- Observable
                   .repeatEvalF(queue.poll())
                   .throttleLast(150.millis)
                   .takeWhile(_ > 100)
                   .toListL
  } yield list

Observable

import monix.eval.Task
import monix.execution.schedulers.TestScheduler
import monix.reactive.Observable

import scala.concurrent.duration._

// using `TestScheduler` to manipulate time
implicit val sc = TestScheduler()

val stream: Task[Unit] = {
  Observable
    .intervalWithFixedDelay(2.second)
    .mapEval(l => Task.sleep(2.second).map(_ => l))
    .foreachL(println)
}

stream.runToFuture(sc)

sc.tick(2.second) // prints 0
sc.tick(4.second) // prints 1
sc.tick(4.second) // prints 2
sc.tick(4.second) // prints 3

Concurrent Data Structures

cats-effect and monix.catnap provide several purely functional data structures for sharing state and synchronization

https://typelevel.org/cats-effect/concurrency/

cats.effect.concurrent.Ref

Must watch: https://vimeo.com/294736344

Purely functional wrapper for AtomicRef. Allows for safe concurrent access and mutation but preserves referential transparency.

abstract class Ref[F[_], A] {
  def get: F[A]
  def set(a: A): F[Unit]
  def modify[B](f: A => (A, B)): F[B]
  // ... and more
}

cats.effect.concurrent.Ref

import monix.eval.Task
import cats.effect.concurrent.Ref

val stringsRef: Task[Ref[Task, List[String]]] =
  Ref.of[IOTaskList[String]](List())

def take(ref: Ref[Task, List[String]]): Task[Option[String]] =
  ref.modify(list => (list.drop(1), list.headOption))

def put(ref: Ref[Task, List[String]], s: String): Task[Unit] =
  ref.modify(list => (s :: list, ()))

val program: Task[Unit] =
  for {
    _ <- stringsRef.flatMap(put(_, "Scala"))
    element <- stringsRef.flatMap(take)
  } yield println(element)

program.runSyncUnsafe() // prints None

cats.effect.concurrent.Ref

val program: Task[Unit] =
  for {
    ref <- stringsRef
    _ <- put(ref, "Scala")
    element <- take(ref)
  } yield println(element)

program.runSyncUnsafe() // prints "Some(Scala)"

The only way to share a state is to pass it as a parameter. :)

cats.effect.concurrent.Deferred

abstract class Deferred[F[_], A] {
  def get: F[A]
  def complete(a: A): F[Unit]
}

val program =
  for {
    stopRunning <- Deferred[Task, Unit]
    _           <- runImportantService(healthcheck).start
    _           <- Task.race(stopRunning.get, runOtherService.loopForever)
  } yield ()

// somewhere in ImportantService
stopRunning.complete(())

Looking for help?

Funtional Concurrency in Scala 101

By Piotr Gawryś

Funtional Concurrency in Scala 101

1,113

Piotr Gawryś

p_gawrys

Funtional Concurrency in Scala 101

Plan

Schedule

What's ahead

Battle City (Tank 1990)

Before we begin...

Functional Programming

Functional Programming

Benefits

What prevents composition?

Side effects

Side effects

So how to write pure functions?

So how to write pure functions?

So how to write pure functions?

Wait, but it still does side effect so what's the point?

Referential Transparency - Benefits

Concurrency

Parallelism

Concurrency on JVM

Threads

Context Switch

Concurrency on JVM

Thread Pools

Blocking Threads

Thread Scheduling

Preemptive multitasking

Cooperative multitasking

Monix

Monix Task

Basic Usage

Error Handling

Error Handling

Parallelism

Scheduler

Scheduler

Observable

Observable

Observable

Observable

Concurrent Data Structures

cats.effect.concurrent.Ref

cats.effect.concurrent.Ref

cats.effect.concurrent.Ref

cats.effect.concurrent.Deferred

Looking for help?

Funtional Concurrency in Scala 101

More from Piotr Gawryś