Scala Cats Effects - IO Async Shift - ¿Cómo funciona?

Aquí hay un código de gatos Scala usando elIO Monad:

import java.util.concurrent.{ExecutorService, Executors}

import cats.effect.IO

import scala.concurrent.{ExecutionContext, ExecutionContextExecutor}
import scala.util.control.NonFatal

object Program extends App {

  type CallbackType = (Either[Throwable, Unit]) => Unit

  // IO.async[Unit] is like a Future that returns Unit on completion.
  // Unlike a regular Future, it doesn't start to run until unsafeRunSync is called.
  def forkAsync(toRun: () => Unit)(executor: ExecutorService): IO[Unit] = IO.async[Unit] { callback: CallbackType =>
    // "callback" is a function that either takes a throwable (Left) or whatever toRun returns (Right).
    println("LalalaAsync: " + Thread.currentThread().getName)
    executor.execute(new Runnable {
      def run(): Unit = {
        val nothing: Unit = toRun() // Note: This line executes the body and returns nothing, which is of type Unit.
        try {
          callback(Right(nothing)) // On success, the callback returns nothing
        } catch {
          case NonFatal(t) => callback(Left(t)) // On failure, it returns an exception
        }
      }
    })
  }

  def forkSync(toRun: () => Unit)(executor: ExecutorService): IO[Unit] = IO.apply {
    println("LalalaSync: " + Thread.currentThread().getName)
    executor.execute(new Runnable {
      def run(): Unit = {
        toRun()
      }
    })
  }

  val treadPool: ExecutorService = Executors.newSingleThreadExecutor()
  val mainThread: Thread = Thread.currentThread()

  val Global: ExecutionContextExecutor = ExecutionContext.global

  /*
  Output:
    1 Hello World printed synchronously from Main.main
    LalalaSync: scala-execution-context-global-12
    Hello World printed synchronously from thread pool.pool-1-thread-1
    LalalaAsync: scala-execution-context-global-12
    Hello World printed asynchronously from thread pool.pool-1-thread-1
    2 Hello World printed synchronously from Global .scala-execution-context-global-12
   */
  val program = for {
    _ <- IO {
      println("1 Hello World printed synchronously from Main." + Thread.currentThread().getName) // "main" thread
    }
    _ <- IO.shift(Global) // Shift to Global Execution Context
    _ <- forkSync { () =>
      println("Hello World printed synchronously from thread pool." + Thread.currentThread().getName) // "pool-1-thread-1" thread
    }(treadPool)
    _ <- forkAsync { () =>
      println("Hello World printed asynchronously from thread pool." + Thread.currentThread().getName) // "pool-1-thread-1" thread
    }(treadPool)
    _ <- IO.shift(Global) // Shift to Global Execution Context
    _ <- IO {
      println("2 Hello World printed synchronously from Global ." + Thread.currentThread().getName) // "scala-execution-context-global-13" thread
    }
  } yield ()

  program.unsafeRunSync()
}

Para ejecutarlo, deberá agregar:

libraryDependencies ++= Seq(
  "org.typelevel" %% "cats" % "0.9.0",
  "org.typelevel" %% "cats-effect" % "0.3"
),

A su archivo build.sbt.

Nota la salida:

  /*
  Output:
    1 Hello World printed synchronously from Main.main
    LalalaSync: scala-execution-context-global-12
    Hello World printed synchronously from thread pool.pool-1-thread-1
    LalalaAsync: scala-execution-context-global-12
    Hello World printed asynchronously from thread pool.pool-1-thread-1
    2 Hello World printed synchronously from Global .scala-execution-context-global-12
 */

ásicamente, no entiendo cómo funciona IO.shift (Global) o cómo funciona IO.async.

Por ejemplo, ¿por qué después de llamar a "forkAsync", si no llamo a "IO.shift (Global)", los siguientes objetos IO síncronos se ejecutan en "pool-1-thread-1". Además, ¿cuál es la diferencia entre forkAsync y forkSync en este ejemplo? Ambos comienzan en ExecutionContext.global y luego ejecutan un Runnable en "pool.pool-1-thread-1".

Al igual que forkAsync y forkSync están haciendo exactamente lo mismo o forkAsync está haciendo algo diferente? Si están haciendo lo mismo, ¿cuál es el punto de envolver el código en IO.async? Si no están haciendo lo mismo, ¿en qué se diferencian?