如何在 Kotlin 的公共池下创建单线程协程上下文？

Question

Short requirement: have ability to create corotine context, which will executed in the single thread only (eg without parallelism).

Additional requirement: it is better to use existing CommonPool (eg Thread Pool) for these tasks

Actually kotlin coroutines have method newSingleThreadContext which will create separate thread and schedule all tasks into it. However, this is dedicated thread, so ~1000 such contexts will require a lot of resources.

Therefore, I'd like to have context with the following characteristics:

• Maximum one task can be executed at the same time
• This context should reuse any other (eg parent context). Eg context should hold no additional threads

Answer 1

Here's a solution:

When you say, for example withSerialContext(Dispatchers.Default) {doWork()} , it executes doWork() on a default dispatcher thread, but all of its parts will execute one at a time like they do in runBlocking{}. Note that even though it's one thread at a time, there is no guarantee that it will be the same thread for the whole operation.

suspend fun <T> withSerialContext(
        context: CoroutineDispatcher,
        block: suspend CoroutineScope.() -> T
): T = withContext(SerialContextDispatcher(context), block)

private class SerialContextDispatcher(private val target: CoroutineDispatcher) : CoroutineDispatcher() {

    private val q = ConcurrentLinkedQueue<Runnable>()
    //Whoever CASes this false->true schedules execution of runproc
    private val pending = AtomicBoolean(false)
    //Only one of these runs at a time
    private val runproc = object: Runnable {
        override fun run() {
            while(true) {
                val proc = q.poll();
                if (proc != null) {
                    try {
                        proc.run()
                    }
                    catch (e: Throwable) {
                        target.dispatch(EmptyCoroutineContext, this)
                        throw e
                    }
                } else {
                    pending.set(false);
                    if (q.isEmpty() || !pending.compareAndSet(false, true)) {
                        return
                    }
                }
            }
        }
    }

    override fun dispatch(context: CoroutineContext, block: Runnable) {
        q.add(block)
        if (pending.compareAndSet(false, true)) {
            target.dispatch(EmptyCoroutineContext, runproc)
        }
    }
}

Answer 2

I found, that there are not simple solution to create such context.

There is open issue on githuib - https://github.com/Kotlin/kotlinx.coroutines/issues/261

I think I will update this question then I will found right solution.

Answer 3

Starting from the 1.6.0 version of the kotlinx.coroutines library we can use limitedParallelism function on the CoroutineDispatcher object, which lets you limit parallelism without creating additional thread pools and provides a unified way to create dispatchers for unbound parallelism.

Example of usage:

class UserRepository {
    private val dbDispatcher = Dispatchers.IO.limitedParallelism(1)

    suspend fun getUserById(userId: Int): User? = withContext(dbDispatcher) {
        executeQuery("SELECT * FROM users WHERE id = $1", userId).singleOrNull()
   }
}

It should solve the problems:

Maximum one task can be executed at the same time and context should hold no additional threads.