Scheduler.forkJoin stops executing tasks after reaching its maximum thread count

[ghik]

Try running this code:

implicit val scheduler: Scheduler = Scheduler.forkJoin(4, 8)

// make the pool reach its maximum thread count
Task.parTraverseUnordered(0 until 1000) { _ =>
   Task(scala.concurrent.blocking(Thread.sleep(10)))
}.runSyncUnsafe()

println("slept")

// this task never executes
Task(println("foo")).executeAsync.runSyncUnsafe()

println("done")

After the pool reaches its maximum size caused by blocking operations, it does not execute any tasks even despite the fact that blocking operations have already finished and threads are idle.

The culprit seems to be monix.execution.internal.forkJoin.DynamicWorkerThreadFactory - the code above works fine if the Scheduler is defined like this:

implicit val scheduler: Scheduler = 
  Scheduler(new ForkJoinPool(4, ForkJoinPool.defaultForkJoinWorkerThreadFactory, null, true, 0, 8, 1, null, 60, TimeUnit.SECONDS))

Environment: Monix 3.4.1, Scala 2.13.8, JDK 11

[ghik]

OK, now I understand why this was done the way it was - ForkJoinPool changed between JDK8 and JDK11 (it didn't have maxThreads parameter in JDK8). Unfortunately, JDK8 is also prone to this bug so I'm not sure what would a JDK-independent fix look like.