Hi, I have a `taskChannel`, a `resultChannel`, a ...

# coroutines

[Esa]

Hi,

I have a taskChannel, a resultChannel, a producer and multiple consumers. The producer is lightning fast and the consumers need a bit of time. Once the producer is done, it closes the taskChannel.

launch {
tasks.forEach {
taskChannel.send(it)
}

taskChannel.close()
}

Then this is the consumer side of things.
launch {
repeat (10) {
launch {
for (task in taskChannel) {
val result = process(task)
resultChannel.send(result)
}
}
}
}

This resultChannel undergoes one final step, which is logging the results:
for (result in resultChannel) {
log(result)
}

resultChannel.close()

My question is really just if I’ve closed the channels at the correct times etc. Also, is this a correct usage of the channels, or have I created some antipatterns or some such here? Any replies appreciated. It’s my first time looking into channels, and I like the way they let me start all three processes (production, consumption and logging) at the same time

[Dominaezzz]

Have you tried running this yet? The resultChannel for loop looks suspicious. I reckon it will deadlock there. The channel as to be closed for the for loop to finish.

[Esa]

It seems to get through all contents of the channel, but it doesn’t “finish” as you said. Once all content has been consumed it still seems to expect more. Should I move the closing of this channel further up, to the consumers? If I do, how exactly? There are multiple consumers there, should I move it into the inner launch?

[Dominaezzz]

You can wrap the repeat in a coroutineScope and close the channel right after.

[Esa]

So the repeat holds a launch call, and outside of the repeat block i close the channel?

[Dominaezzz]

Yes, like so.

launch {
  coroutineScope {
    repeat (10) {
        launch {
            for (task in taskChannel) {
                val result = process(task)
                resultChannel.send(result)
            }
        }
    }
  }
  resultChannel.close()
}

[Esa]

Ah, nice. I was not aware you could scope work like that and be sure any calls after would only be invoked after all children had completed their work. Thanks!

[Esa]

This worked perfectly, thank you. Successfully completed this time instead of waiting in limbo.

[Esa]

If I need to be able to instantly pause / stop every worker, what’s the best way to achieve that?

Explore two options here, instantly and after current Job

[Dominaezzz]

What is a worker?

[Esa]

In the last snippet posted in this thread, there’s a repeat() { launch { // this is a worker } } section there.

Anything inside that launch I concider a worker

[Esa]

That launch has a for-in loop that needs to be able to stop and resume work on command, not sure how I’d go about that the best way

[Dominaezzz]

Do you want to stop them individually or as a group?

[Esa]

Whichever is the better approach. I don’t have a preference, the important bit is that on command, everything stops (suspends?) picking new tasks from the channel until a new command is given to resume picking new tasks.

[Dominaezzz]

Tricky, I'd argue you should do this via the taskChannel , don't send any tasks through, when you wish to pause.

[Esa]

Aha okay, So I could set bufferSize to 1, and when I send this command somehow, it prevents the taskChannel from receiving new tasks in a way.. Got it, thanks. Will try it