I'm experimenting with fs2.Stream concurrent features and got some misunderstanding about how it works. I would like to send stream content through some sink in parallel. Here is what I tried:
object TestParallelStream extends App {
val secondsOnStart = TimeUnit.MILLISECONDS.toSeconds(System.currentTimeMillis())
val stream = fs2.Stream.emits(List(1, 2, 3, 4, 5, 6, 7, 8, 9)).covary[IO]
val sink: fs2.Sink[IO, Int] = _.evalMap(i => IO {
println(s"[${TimeUnit.MILLISECONDS.toSeconds(System.currentTimeMillis()) - secondsOnStart} second]: $i")
Thread.sleep(5000)
})
val executor = Executors.newFixedThreadPool(4)
implicit val cs: ContextShift[IO] = IO.contextShift(ExecutionContext.fromExecutor(executor))
stream.observeAsync(3)(sink).compile.drain.unsafeRunSync() //1
executor.shutdown()
}
The //1 prints the following content:
[1 second]: 1
[6 second]: 2
[11 second]: 3
[16 second]: 4
[21 second]: 5
[26 second]: 6
[31 second]: 7
[36 second]: 8
[41 second]: 9
As can be seen from the output, each element is sent through the sink sequentially.
But if I modify the sink as follows:
// 5 limit and parEvalMap
val sink: fs2.Sink[IO, Int] = _.parEvalMap(5)(i => IO {
println(s"[${TimeUnit.MILLISECONDS.toSeconds(System.currentTimeMillis()) - secondsOnStart} second]: $i")
Thread.sleep(5000)
})
The output is:
[1 second]: 3
[1 second]: 2
[1 second]: 4
[1 second]: 1
[6 second]: 5
[6 second]: 6
[6 second]: 7
[6 second]: 8
[11 second]: 9
Now we have 4 elements are being sent through the sink in parallel at a time (in spite of setting 3 as a limit of observerAsync).
Even if I replace observerAsync with just observe I got the same parallelization effect.
Can you please clarify how sinks actually work?
