How to implement parallel mapping for sequences in kotlin

天大地大妈咪最大 提交于 2021-01-28 04:40:54

问题


I'm trying to implement a parallel implementation for both Iterable and Sequence in Kotlin. I got a little file, it consists of 4 extension functions, but the third one gives me an compiler error:

suspend fun <T, R> Iterable<T>.parallelMap(block: suspend(T) -> R) =
    coroutineScope { map { async { block(it) } }.map { it.await() } }

suspend fun <T> Iterable<T>.parallelForEach(block: suspend (T) -> Unit) =
    coroutineScope { map { async { block(it) } }.forEach { it.await() } }

suspend fun <T, R> Sequence<T>.parallelMap(block: suspend(T) -> R) =
    coroutineScope { map { async { block(it) } }.map { it.await() } }

suspend fun <T> Sequence<T>.parallelForEach(block: suspend (T) -> Unit) =
    coroutineScope { map { async { block(it) } }.forEach { it.await() } }

The compiler comes back and says that suspension functions can only be called inside suspension functions. Is there a way to implement this?

Edit: fixed bad copy/paste

Edit2: I thought of an implementation:

suspend fun <T, R> Sequence<T>.parrallelMap(block: suspend (T) -> R) =
        asIterable().map { coroutineScope { async { block(it) } } }
              .asSequence().map { runBlocking { it.await() } }

I was hoping that this would fire all the suspending functions and await them lazily. I'm just not sure if this is safe, or this saves time or not.


回答1:


There is a problem with the core semantics of parallel execution for lazy sequences. Your current implementation does not start block(it) until the resulting sequence is iterated:

suspend fun <T, R> Sequence<T>.parallelMap(block: suspend(T) -> R) =
    coroutineScope { map { async { block(it) } }.map { it.await() } }

Consider the following example:

sequenceOf(1, 2, 3).parallelMap { it * it }.forEach { println(it) }

For this sample the order of execution will be

val p1 = async { 1 * 1 } 
val r1 = p1.await()
println(r1)
val p2 = async { 2 * 2 } 
val r2 = p2.await()
println(r2)
val p3 = async { 3 * 3 } 
val r3 = p3.await()
println(r3)

Note that the execution of the mapping operations is sequientional, not parallel.

What the compiler tells you is that the lambda of Sequence<T>.map {} is performed lazily on-demand outside of the context of the call (read: outside of your coroutine), so you can't use the coroutine you are currently in.

Frankly, I am not sure how one can both perform lazy computation and do it in parallel.



来源:https://stackoverflow.com/questions/55830626/how-to-implement-parallel-mapping-for-sequences-in-kotlin

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!