concurrency semaphore

A semaphore has a non-negative number of permits available. Acquiring a permit decrements the current number of permits and releasing a permit increases the current number of permits. An acquire that occurs when there are no permits available results in semantic blocking until a permit becomes available.

abstract class Semaphore[F[_]] {
  def available: F[Long]
  def acquire: F[Unit]
  def release: F[Unit]
  // ... and more

Semantic Blocking and Cancellation

Semaphore does what we call “semantic” blocking, meaning that no actual threads are being blocked while waiting to acquire a permit.

Behavior on cancellation:

  • Blocking acquires are cancelable if the semaphore is created with Semaphore.apply (and hence, with a Concurrent[F] instance).
  • Blocking acquires are non-cancelable if the semaphore is created with Semaphore.uncancelable (and hence, with an Async[F] instance).

Shared Resource

When multiple processes try to access a precious resource you might want to constraint the number of accesses. Here is where Semaphore[F] is useful.

Three processes are trying to access a shared resource at the same time but only one at a time will be granted access and the next process have to wait until the resource gets available again (availability is one as indicated by the semaphore counter).

R1, R2 & R3 will request access of the precious resource concurrently so this could be one possible outcome:

R1 >> Availability: 1
R2 >> Availability: 1
R2 >> Started | Availability: 0
R3 >> Availability: 0
R1 >> Started | Availability: 0
R2 >> Done | Availability: 0
R3 >> Started | Availability: 0
R1 >> Done | Availability: 0
R3 >> Done | Availability: 1

This means when R1 and R2 requested the availability it was one and R2 was faster in getting access to the resource so it started processing. R3 was the slowest and saw that there was no availability from the beginning.

Once R2 was done R1 started processing immediately showing no availability. Once R1 was done R3 started processing immediately showing no availability. Finally, R3 was done showing an availability of one once again.

import cats.Parallel
import cats.effect.{Concurrent, IO, Timer}
import cats.effect.concurrent.Semaphore
import cats.implicits._

import scala.concurrent.ExecutionContext
import scala.concurrent.duration._

// Needed for getting a Concurrent[IO] instance
implicit val ctx = IO.contextShift(
// Needed for `sleep`
implicit val timer = IO.timer(

class PreciousResource[F[_]](name: String, s: Semaphore[F])(implicit F: Concurrent[F], timer: Timer[F]) {
  def use: F[Unit] =
    for {
      x <- s.available
      _ <- F.delay(println(s"$name >> Availability: $x"))
      _ <- s.acquire
      y <- s.available
      _ <- F.delay(println(s"$name >> Started | Availability: $y"))
      _ <- timer.sleep(3.seconds)
      _ <- s.release
      z <- s.available
      _ <- F.delay(println(s"$name >> Done | Availability: $z"))
    } yield ()

implicit val par: Parallel[IO, IO] = Parallel[IO, IO.Par].asInstanceOf[Parallel[IO, IO]]

val program: IO[Unit] =
  for {
    s  <- Semaphore[IO](1)
    r1 = new PreciousResource[IO]("R1", s)
    r2 = new PreciousResource[IO]("R2", s)
    r3 = new PreciousResource[IO]("R3", s)
    _  <- List(r1.use, r2.use, r3.use).parSequence.void
  } yield ()