前置知识
Kotlin协程不是什么空中阁楼,Kotlin源代码会被编译成class字节码文件,最终会运行到虚拟机中。所以从本质上讲,Kotlin和Java是类似的,都是可以编译产生class的语言,但最终还是会受到虚拟机的限制,它们的代码最终会在虚拟机上的某个线程上被执行。
之前我们分析了launch的原理,但当时我们没有去分析协程创建出来后是如何与线程产生关联的,怎么被分发到具体的线程上执行的,本篇文章就带大家分析一下。
要想搞懂Dispatchers,我们先来看一下Dispatchers、CoroutineDispatcher、ContinuationInterceptor、CoroutineContext之间的关系
public actual object Dispatchers { @JvmStatic public actual val Default: CoroutineDispatcher = DefaultScheduler @JvmStatic public actual val Main: MainCoroutineDispatcher get() = MainDispatcherLoader.dispatcher @JvmStatic public actual val Unconfined: CoroutineDispatcher = kotlinx.coroutines.Unconfined @JvmStatic public val IO: CoroutineDispatcher = DefaultIoScheduler } public abstract class CoroutineDispatcher : AbstractCoroutineContextElement(ContinuationInterceptor), ContinuationInterceptor { } public interface ContinuationInterceptor : CoroutineContext.Element {} public interface Element : CoroutineContext {}
Dispatchers中存放的是协程调度器(它本身是一个单例),有我们平时常用的IO、Default、Main等。这些协程调度器都是CoroutineDispatcher的子类,这些协程调度器其实都是CoroutineContext。
demo
我们先来看一个关于launch的demo:
fun main() { val coroutineScope = CoroutineScope(Job()) coroutineScope.launch { println("Thread : ${Thread.currentThread().name}") } Thread.sleep(5000L) }
在生成CoroutineScope时,demo中没有传入相关的协程调度器,也就是Dispatchers。那这个launch会运行到哪个线程之上?
运行试一下:
Thread : DefaultDispatcher-worker-1
居然运行到了DefaultDispatcher-worker-1
线程上,这看起来明显是Dispatchers.Default
协程调度器里面的线程。我明明没传Dispatchers相关的context,居然会运行到子线程上。说明运行到default线程是launch默认的。
它是怎么与default线程产生关联的?打开源码一探究竟:
public fun CoroutineScope.launch( context: CoroutineContext = EmptyCoroutineContext, start: CoroutineStart = CoroutineStart.DEFAULT, block: suspend CoroutineScope.() -> Unit ): Job { //代码1 val newContext = newCoroutineContext(context) //代码2 val coroutine = if (start.isLazy) LazyStandaloneCoroutine(newContext, block) else StandaloneCoroutine(newContext, active = true) //代码3 coroutine.start(start, coroutine, block) return coroutine }
- 将传入的CoroutineContext构造出新的context
- 启动模式,判断是否为懒加载,如果是懒加载则构建懒加载协程对象,否则就是标准的
- 启动协程
我们重点关注代码1,这是与CoroutineContext相关的。
public actual fun CoroutineScope.newCoroutineContext(context: CoroutineContext): CoroutineContext { //从父协程那里继承过来的context+这次的context val combined = coroutineContext.foldCopiesForChildCoroutine() + context val debug = if (DEBUG) combined + CoroutineId(COROUTINE_ID.incrementAndGet()) else combined //combined可以简单的把它看成是一个map,它是CoroutineContext类型的 //如果当前context不等于Dispatchers.Default,而且从map里面取ContinuationInterceptor(用于拦截之后分发线程的)值为空,说明没有传入协程应该在哪个线程上运行的相关参数 return if (combined !== Dispatchers.Default && combined[ContinuationInterceptor] == null) debug + Dispatchers.Default else debug }
调用launch的时候,我们没有传入context,默认参数是EmptyCoroutineContext。这里的combined,它其实是CoroutineContext类型的,可以简单的看成是map(其实不是,只是类似)。
通过combined[ContinuationInterceptor]可以将传入的线程调度相关的参数给取出来,这里如果取出来为空,是给该context添加了一个Dispatchers.Default,然后把新的context返回出去了。所以launch默认情况下,会走到default线程去执行。
补充一点:CoroutineContext能够通过+
连接是因为它内部有个public operator fun plus
函数。能够通过combined[ContinuationInterceptor]这种方式访问元素是因为有个public operator fun get
函数。
public interface CoroutineContext { /** * Returns the element with the given [key] from this context or `null`. */ public operator funget(key: Key ): E? /** * Returns a context containing elements from this context and elements from other [context]. * The elements from this context with the same key as in the other one are dropped. */ public operator fun plus(context: CoroutineContext): CoroutineContext { ...... } }
startCoroutineCancellable
上面我们分析了launch默认情况下,context中会增加Dispatchers.Default的这个协程调度器,到时launch的Lambda会在default线程上执行,其中具体流程是怎么样的,我们分析一下。
在之前的文章 Kotlin协程之launch原理 中我们分析过,launch默认情况下会最终执行到startCoroutineCancellable
函数。
public fun(suspend () -> T).startCoroutineCancellable(completion: Continuation ): Unit = runSafely(completion) { //构建ContinuationImpl createCoroutineUnintercepted(completion).intercepted().resumeCancellableWith(Result.success(Unit)) } public actual fun (suspend () -> T).createCoroutineUnintercepted( completion: Continuation ): Continuation { val probeCompletion = probeCoroutineCreated(completion) return if (this is BaseContinuationImpl) //走这里 create(probeCompletion) else createCoroutineFromSuspendFunction(probeCompletion) { (this as Function1 , Any?>).invoke(it) } }
在Kotlin协程之launch原理 文章中,咱们分析过create(probeCompletion)这里创建出来的是launch的那个Lambda,编译器会产生一个匿名内部类,它继承自SuspendLambda,而SuspendLambda是继承自ContinuationImpl。
所以 createCoroutineUnintercepted(completion)一开始构建出来的是一个ContinuationImpl,接下来需要去看它的intercepted()函数。
intercepted()函数
internal abstract class ContinuationImpl( completion: Continuation?, private val _context: CoroutineContext? ) : BaseContinuationImpl(completion) { constructor(completion: Continuation ?) : this(completion, completion?.context) public override val context: CoroutineContext get() = _context!! @Transient private var intercepted: Continuation ? = null public fun intercepted(): Continuation = intercepted ?: (context[ContinuationInterceptor]?.interceptContinuation(this) ?: this) .also { intercepted = it } }
第一次走到intercepted()函数时,intercepted肯定是为null的,还没初始化。此时会通过context[ContinuationInterceptor]取出Dispatcher对象,然后调用该Dispatcher对象的interceptContinuation()函数。这个Dispatcher对象在demo这里其实就是Dispatchers.Default。
public actual object Dispatchers { @JvmStatic public actual val Default: CoroutineDispatcher = DefaultScheduler }
可以看到,Dispatchers.Default是一个CoroutineDispatcher对象,interceptContinuation()函数就在CoroutineDispatcher中。
public abstract class CoroutineDispatcher : AbstractCoroutineContextElement(ContinuationInterceptor), ContinuationInterceptor { public final override fun <T> interceptContinuation(continuation: Continuation<T>): Continuation<T> = DispatchedContinuation(this, continuation) } public fun <T> (suspend () -> T).startCoroutineCancellable(completion: Continuation<T>): Unit = runSafely(completion) { createCoroutineUnintercepted(completion).intercepted().resumeCancellableWith(Result.success(Unit)) }
这个方法非常简单,就是新建并且返回了一个DispatchedContinuation对象,将this和continuation给传入进去。这里的this是Dispatchers.Default。
所以,最终我们发现走完startCoroutineCancellable的前2步之后,也就是走完intercepted()之后,创建的是DispatchedContinuation对象,最后是调用的DispatchedContinuation的resumeCancellableWith函数。最后这步比较关键,这是真正将协程的具体执行逻辑放到线程上执行的部分。
internal class DispatchedContinuation( //这里传入的dispatcher在demo中是Dispatchers.Default @JvmField val dispatcher: CoroutineDispatcher, @JvmField val continuation: Continuation ) : DispatchedTask (MODE_UNINITIALIZED), CoroutineStackFrame, Continuation by continuation { inline fun resumeCancellableWith( result: Result , noinline onCancellation: ((cause: Throwable) -> Unit)? ) { val state = result.toState(onCancellation) //代码1 if (dispatcher.isDispatchNeeded(context)) { _state = state resumeMode = MODE_CANCELLABLE //代码2 dispatcher.dispatch(context, this) } else { //代码3 executeUnconfined(state, MODE_CANCELLABLE) { if (!resumeCancelled(state)) { resumeUndispatchedWith(result) } } } } } internal abstract class DispatchedTask ( @JvmField public var resumeMode: Int ) : SchedulerTask() { ...... } internal actual typealias SchedulerTask = Task internal abstract class Task( @JvmField var submissionTime: Long, @JvmField var taskContext: TaskContext ) : Runnable { ...... } public abstract class CoroutineDispatcher : AbstractCoroutineContextElement(ContinuationInterceptor), ContinuationInterceptor { public abstract fun dispatch(context: CoroutineContext, block: Runnable) public open fun isDispatchNeeded(context: CoroutineContext): Boolean = true }
从DispatchedContinuation的继承结构来看,它既是一个Continuation(通过委托给传入的continuation参数),也是一个Runnable。
- 首先看代码1:这个dispatcher在demo中其实是Dispatchers.Default ,然后调用它的isDispatchNeeded(),这个函数定义在CoroutineDispatcher中,默认就是返回true,只有Dispatchers.Unconfined返回false
- 代码2:调用Dispatchers.Default的dispatch函数,将context和自己(DispatchedContinuation,也就是Runnable)传过去了
- 代码3:对应Dispatchers.Unconfined的情况,它的isDispatchNeeded()返回false
现在我们要分析代码2之后的执行逻辑,也就是将context和Runnable传入到dispatch函数之后是怎么执行的。按道理,看到Runnable,那可能这个与线程执行相关,应该离我们想要的答案不远了。回到Dispatchers,我们发现Dispatchers.Default是DefaultScheduler类型的,那我们就去DefaultScheduler中或者其父类中去找dispatch函数。
DefaultScheduler中找dispatch函数
public actual object Dispatchers { @JvmStatic public actual val Default: CoroutineDispatcher = DefaultScheduler } internal object DefaultScheduler : SchedulerCoroutineDispatcher( CORE_POOL_SIZE, MAX_POOL_SIZE, IDLE_WORKER_KEEP_ALIVE_NS, DEFAULT_SCHEDULER_NAME ) { ...... } internal open class SchedulerCoroutineDispatcher( private val corePoolSize: Int = CORE_POOL_SIZE, private val maxPoolSize: Int = MAX_POOL_SIZE, private val idleWorkerKeepAliveNs: Long = IDLE_WORKER_KEEP_ALIVE_NS, private val schedulerName: String = "CoroutineScheduler", ) : ExecutorCoroutineDispatcher() { private var coroutineScheduler = createScheduler() private fun createScheduler() = CoroutineScheduler(corePoolSize, maxPoolSize, idleWorkerKeepAliveNs, schedulerName) override fun dispatch(context: CoroutineContext, block: Runnable): Unit = coroutineScheduler.dispatch(block) }
最后发现dispatch函数在其父类SchedulerCoroutineDispatcher中,在这里构建了一个CoroutineScheduler,直接调用了CoroutineScheduler对象的dispatch,然后将Runnable(也就是上面的DispatchedContinuation对象)传入。
Runnable传入
internal class CoroutineScheduler( @JvmField val corePoolSize: Int, @JvmField val maxPoolSize: Int, @JvmField val idleWorkerKeepAliveNs: Long = IDLE_WORKER_KEEP_ALIVE_NS, @JvmField val schedulerName: String = DEFAULT_SCHEDULER_NAME ) : Executor, Closeable { override fun execute(command: Runnable) = dispatch(command) fun dispatch(block: Runnable, taskContext: TaskContext = NonBlockingContext, tailDispatch: Boolean = false) { trackTask() // this is needed for virtual time support //代码1:构建Task,Task实现了Runnable接口 val task = createTask(block, taskContext) //代码2:取当前线程转为Worker对象,Worker是一个继承自Thread的类 val currentWorker = currentWorker() //代码3:尝试将Task提交到本地队列并根据结果执行相应的操作 val notAdded = currentWorker.submitToLocalQueue(task, tailDispatch) if (notAdded != null) { //代码4:notAdded不为null,则再将notAdded(Task)添加到全局队列中 if (!addToGlobalQueue(notAdded)) { throw RejectedExecutionException("$schedulerName was terminated") } } val skipUnpark = tailDispatch && currentWorker != null // Checking 'task' instead of 'notAdded' is completely okay if (task.mode == TASK_NON_BLOCKING) { if (skipUnpark) return //代码5: 创建Worker并开始执行该线程 signalCpuWork() } else { // Increment blocking tasks anyway signalBlockingWork(skipUnpark = skipUnpark) } } private fun currentWorker(): Worker? = (Thread.currentThread() as? Worker)?.takeIf { it.scheduler == this } internal inner class Worker private constructor() : Thread() { ..... } }
观察发现,原来CoroutineScheduler类实现了java.util.concurrent.Executor接口,同时实现了它的execute方法,这个方法也会调用dispatch()。
- 代码1:首先是通过Runnable构建了一个Task,这个Task其实也是实现了Runnable接口,只是把传入的Runnable包装了一下
- 代码2:将当前线程取出来转换成Worker,当然第一次时,这个转换不会成功,这个Worker是继承自Thread的一个类
- 代码3:将task提交到本地队列中,这个本地队列待会儿会在Worker这个线程执行时取出Task,并执行Task
- 代码4:如果task提交到本地队列的过程中没有成功,那么会添加到全局队列中,待会儿也会被Worker取出来Task并执行
- 代码5:创建Worker线程,并开始执行
开始执行Worker线程之后,我们需要看一下这个线程的run方法执行的是啥,也就是它的具体执行逻辑。
Worker线程执行逻辑
internal inner class Worker private constructor() : Thread() { override fun run() = runWorker() private fun runWorker() { var rescanned = false while (!isTerminated && state != WorkerState.TERMINATED) { //代码1 val task = findTask(mayHaveLocalTasks) if (task != null) { rescanned = false minDelayUntilStealableTaskNs = 0L //代码2 executeTask(task) continue } else { mayHaveLocalTasks = false } if (minDelayUntilStealableTaskNs != 0L) { if (!rescanned) { rescanned = true } else { rescanned = false tryReleaseCpu(WorkerState.PARKING) interrupted() LockSupport.parkNanos(minDelayUntilStealableTaskNs) minDelayUntilStealableTaskNs = 0L } continue } tryPark() } tryReleaseCpu(WorkerState.TERMINATED) } fun findTask(scanLocalQueue: Boolean): Task? { if (tryAcquireCpuPermit()) return findAnyTask(scanLocalQueue) // If we can't acquire a CPU permit -- attempt to find blocking task val task = if (scanLocalQueue) { localQueue.poll() ?: globalBlockingQueue.removeFirstOrNull() } else { globalBlockingQueue.removeFirstOrNull() } return task ?: trySteal(blockingOnly = true) } private fun executeTask(task: Task) { val taskMode = task.mode idleReset(taskMode) beforeTask(taskMode) runSafely(task) afterTask(taskMode) } fun runSafely(task: Task) { try { task.run() } catch (e: Throwable) { val thread = Thread.currentThread() thread.uncaughtExceptionHandler.uncaughtException(thread, e) } finally { unTrackTask() } } }
run方法直接调用的runWorker(),在里面是一个while循环,不断从队列中取Task来执行。
- 代码1:从本地队列或者全局队列中取出Task
- 代码2:执行这个task,最终其实就是调用这个Runnable的run方法。
也就是说,在Worker这个线程中,执行了这个Runnable的run方法。还记得这个Runnable是谁么?它就是上面我们看过的DispatchedContinuation,这里的run方法执行的就是协程任务,那这块具体的run方法的实现逻辑,我们应该到DispatchedContinuation中去找。
internal class DispatchedContinuation( @JvmField val dispatcher: CoroutineDispatcher, @JvmField val continuation: Continuation ) : DispatchedTask (MODE_UNINITIALIZED), CoroutineStackFrame, Continuation by continuation { ...... } internal abstract class DispatchedTask ( @JvmField public var resumeMode: Int ) : SchedulerTask() { public final override fun run() { assert { resumeMode != MODE_UNINITIALIZED } // should have been set before dispatching val taskContext = this.taskContext var fatalException: Throwable? = null try { val delegate = delegate as DispatchedContinuation val continuation = delegate.continuation withContinuationContext(continuation, delegate.countOrElement) { val context = continuation.context val state = takeState() // NOTE: Must take state in any case, even if cancelled val exception = getExceptionalResult(state) /* * Check whether continuation was originally resumed with an exception. * If so, it dominates cancellation, otherwise the original exception * will be silently lost. */ val job = if (exception == null && resumeMode.isCancellableMode) context[Job] else null //非空,且未处于active状态 if (job != null && !job.isActive) { //开始之前,协程已经被取消,将具体的Exception传出去 val cause = job.getCancellationException() cancelCompletedResult(state, cause) continuation.resumeWithStackTrace(cause) } else { //有异常,传递异常 if (exception != null) { continuation.resumeWithException(exception) } else { //代码1 continuation.resume(getSuccessfulResult(state)) } } } } catch (e: Throwable) { // This instead of runCatching to have nicer stacktrace and debug experience fatalException = e } finally { val result = runCatching { taskContext.afterTask() } handleFatalException(fatalException, result.exceptionOrNull()) } } }
我们主要看一下代码1处,调用了resume开启协程。前面没有异常,才开始启动协程,这里才是真正的开始启动协程,开始执行launch传入的Lambda表达式。这个时候,协程的逻辑是在Worker这个线程上执行的了,切到某个线程上执行的逻辑已经完成了。
ps: rusume会走到BaseContinuationImpl的rusumeWith,然后走到launch传入的Lambda匿名内部类的invokeSuspend方法,开始执行状态机逻辑。前面的文章 Kotlin协程createCoroutine和startCoroutine原理 我们分析过这里,这里就只是简单提一下。
到这里,Dispatchers的执行流程就算完了,前后都串起来了。
小结
Dispatchers是协程框架中与线程交互的关键。底层会有不同的线程池,Dispatchers.Default、IO,协程任务来了的时候会封装成一个个的Runnable,丢到线程中执行,这些Runnable的run方法中执行的其实就是continuation.resume,也就是launch的Lambda生成的SuspendLambda匿名内部类,也就是开启协程状态机,开始协程的真正执行。
以上就是Kotlin协程Dispatchers原理示例详解的详细内容,更多关于Kotlin协程Dispatchers的资料请关注脚本之家其它相关文章!