rcardin / sus4s   0.0.3

Apache License 2.0 GitHub

A Direct-Style Scala Wrapper Around the Structured Concurrency of Project Loom

Scala versions: 3.x

sus4s

A Direct-Style Scala Wrapper Around the Structured Concurrency of Project Loom

Dependency

The library is available on Maven Central. To use it, add the following dependency to your build.sbt files:

libraryDependencies += "in.rcard.sus4s" %% "core" % "0.0.3"

The library is only available for Scala 3.

Usage

The library provides a direct-style API for Project Loom's structured concurrency. It requires JDK 21 and Scala 3. Moreover, Java preview features must be enabled in the Scala compiler.

The main entry point is the sus4s package object. The following code snippet shows how to use the library:

import in.rcard.sus4s.sus4s.*
import scala.concurrent.duration.*

val result: Int = structured {
  val job1: Job[Int] = fork {
    delay(1.second)
    42
  }
  val job2: Job[Int] = fork {
    delay(500.millis)
    43
  }
  job1.value + job2.value
}
println(result) // 85

The structured method creates a new structured concurrency scope represented by the Suspend trait. It's built on the java.util.concurrent.StructuredTaskScope class. Hence, the threads forked inside the structured block are Java Virtual Threads.

The fork method creates a new Java Virtual Thread that executes the given code block. The fork method executes functions declared with the capability of suspend:

def findUserById(id: UserId): Suspend ?=> User

Coloring a function with the Suspend capability tells the caller that the function performs a suspendable operation, aka some effect. Suspension is managed by the Loom runtime, which is responsible for scheduling the virtual threads.

A type alias is available for the Suspend capability:

type Suspended[A] = Suspend ?=> A

So, the above function can be rewritten as:

def findUserById(id: UserId): Suspended[User]

The structured function uses structured concurrency to run the suspendable tasks. In detail, it ensures that the thread executing the block waits to complete all the forked tasks. The structured blocks terminate when:

  • all the forked tasks complete successfully
  • one of the forked tasks throws an exception
  • the block throws an exception

The Job Class

Forking a suspendable function means creating a new virtual thread that executes the function. The thread is represented by the Job class. The Job class provides the value method that waits for the completion of the virtual thread and returns the result of the function:

val job1: Job[Int] = fork {
  delay(1.second)
  42
}
val meaningOfLife: Int = job1.value

If you're not interested in the result of the function, you can use the join method:

val job1: Job[Int] = fork {
  delay(1.second)
  println("The meaning of life is 42")
}
job1.join()

The structured function is entirely transparent to any exception thrown by the block or forked tasks.

Canceling a Job

Canceling a job is possible by calling the cancel method on the Job instance. The following code snippet shows how:

val queue = new ConcurrentLinkedQueue[String]()
val result = structured {
  val job1 = fork {
    val innerCancellableJob = fork {
      while (true) {
        delay(2.seconds)
        queue.add("cancellable")
      }
    }
    delay(1.second)
    innerCancellableJob.cancel()
    queue.add("job1")
  }
  val job = fork {
    delay(500.millis)
    queue.add("job2")
    43
  }
  job.value
}
queue.toArray should contain theSameElementsInOrderAs List("job2", "job1")
result shouldBe 43

Cancellation is collaborative. In the above example, the job innerCancellableJob is marked for cancellation by the call innerCancellableJob.cancel(). However, the job is not immediately canceled. The job is canceled when it reaches the first point operation that can be interrupted by the JVM. Hence, cancellation is based on the concept of interruption. In the above example, the innerCancellableJob is canceled when it reaches the delay(2.seconds) operation. The job will never be canceled if we remove the delay operation. A similar behavior is implemented by Kotlin coroutines (see Kotlin Coroutines - A Comprehensive Introduction / Cancellation for further details).

Cancelling a job follows the relationship between parent and child jobs. If a parent's job is canceled, all the children's jobs are canceled as well:

val expectedQueue = structured {
  val queue = new ConcurrentLinkedQueue[String]()
  val job1 = fork {
    val innerJob = fork {
      fork {
        delay(3.seconds)
        println("inner-inner-Job")
        queue.add("inner-inner-Job")
      }
      delay(2.seconds)
      println("innerJob")
      queue.add("innerJob")
    }
    delay(1.second)
    queue.add("job1")
  }
  val job = fork {
    delay(500.millis)
    job1.cancel()
    queue.add("job2")
    43
  }
  queue
}
expectedQueue.toArray should contain theSameElementsInOrderAs List("job2")

Trying to get the value from a canceled job will throw an InterruptedException. However, joining a canceled job will not throw any exception.

You won't pay any additional cost for canceling a job. The cancellation mechanism is based on the interruption of the virtual thread. No new structured scope is created for the cancellation mechanism.

Racing Jobs

The library provides the race method to race two jobs. The race function returns the result of the first job that completes. The other job is canceled. The following code snippet shows how to use the race method:

val results = new ConcurrentLinkedQueue[String]()
val actual: Int | String = 
  race[Int, String](
    {
      delay(1.second)
      results.add("job1")
      throw new RuntimeException("Error")
    }, {
      delay(500.millis)
      results.add("job2")
      "42"
    }
  )
actual should be("42")
results.toArray should contain theSameElementsInOrderAs List("job2")

If the first job completes with an exception, the race method waits for the second job to complete. and returns the result of the second job. If the second job completes with an exception, the race method throws the first exception it encountered.

Each job adhere to the rules of structured concurrency. The race function is optimized. Every raced block creates more than one virtual thread under the hood, which should not be a problem for the Loom runtime.

Contributing

If you want to contribute to the project, please do it! Any help is welcome.

Acknowledgments

This project is inspired by the Ox and the Unwrapped libraries.