001/* 002 * Copyright (C) 2018 The Guava Authors 003 * 004 * Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except 005 * in compliance with the License. You may obtain a copy of the License at 006 * 007 * http://www.apache.org/licenses/LICENSE-2.0 008 * 009 * Unless required by applicable law or agreed to in writing, software distributed under the License 010 * is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express 011 * or implied. See the License for the specific language governing permissions and limitations under 012 * the License. 013 */ 014 015package com.google.common.util.concurrent; 016 017import static com.google.common.base.Preconditions.checkNotNull; 018import static com.google.common.base.Preconditions.checkState; 019import static com.google.common.util.concurrent.ExecutionSequencer.RunningState.CANCELLED; 020import static com.google.common.util.concurrent.ExecutionSequencer.RunningState.NOT_RUN; 021import static com.google.common.util.concurrent.ExecutionSequencer.RunningState.STARTED; 022import static com.google.common.util.concurrent.Futures.immediateCancelledFuture; 023import static com.google.common.util.concurrent.Futures.immediateFuture; 024import static com.google.common.util.concurrent.Futures.immediateVoidFuture; 025import static com.google.common.util.concurrent.MoreExecutors.directExecutor; 026import static java.util.Objects.requireNonNull; 027 028import com.google.common.annotations.Beta; 029import java.util.concurrent.Callable; 030import java.util.concurrent.Executor; 031import java.util.concurrent.atomic.AtomicReference; 032import javax.annotation.CheckForNull; 033import org.checkerframework.checker.nullness.qual.Nullable; 034 035/** 036 * Serializes execution of tasks, somewhat like an "asynchronous {@code synchronized} block." Each 037 * {@linkplain #submit enqueued} callable will not be submitted to its associated executor until the 038 * previous callable has returned -- and, if the previous callable was an {@link AsyncCallable}, not 039 * until the {@code Future} it returned is {@linkplain Future#isDone done} (successful, failed, or 040 * cancelled). 041 * 042 * <p>This class has limited support for cancellation and other "early completion": 043 * 044 * <ul> 045 * <li>While calls to {@code submit} and {@code submitAsync} return a {@code Future} that can be 046 * cancelled, cancellation never propagates to a task that has started to run -- neither to 047 * the callable itself nor to any {@code Future} returned by an {@code AsyncCallable}. 048 * (However, cancellation can prevent an <i>unstarted</i> task from running.) Therefore, the 049 * next task will wait for any running callable (or pending {@code Future} returned by an 050 * {@code AsyncCallable}) to complete, without interrupting it (and without calling {@code 051 * cancel} on the {@code Future}). So beware: <i>Even if you cancel every precededing {@code 052 * Future} returned by this class, the next task may still have to wait.</i>. 053 * <li>Once an {@code AsyncCallable} returns a {@code Future}, this class considers that task to 054 * be "done" as soon as <i>that</i> {@code Future} completes in any way. Notably, a {@code 055 * Future} is "completed" even if it is cancelled while its underlying work continues on a 056 * thread, an RPC, etc. The {@code Future} is also "completed" if it fails "early" -- for 057 * example, if the deadline expires on a {@code Future} returned from {@link 058 * Futures#withTimeout} while the {@code Future} it wraps continues its underlying work. So 059 * beware: <i>Your {@code AsyncCallable} should not complete its {@code Future} until it is 060 * safe for the next task to start.</i> 061 * </ul> 062 * 063 * <p>An additional limitation: this class serializes execution of <i>tasks</i> but not any 064 * <i>listeners</i> of those tasks. 065 * 066 * <p>This class is similar to {@link MoreExecutors#newSequentialExecutor}. This class is different 067 * in a few ways: 068 * 069 * <ul> 070 * <li>Each task may be associated with a different executor. 071 * <li>Tasks may be of type {@code AsyncCallable}. 072 * <li>Running tasks <i>cannot</i> be interrupted. (Note that {@code newSequentialExecutor} does 073 * not return {@code Future} objects, so it doesn't support interruption directly, either. 074 * However, utilities that <i>use</i> that executor have the ability to interrupt tasks 075 * running on it. This class, by contrast, does not expose an {@code Executor} API.) 076 * </ul> 077 * 078 * <p>If you don't need the features of this class, you may prefer {@code newSequentialExecutor} for 079 * its simplicity and ability to accommodate interruption. 080 * 081 * @since 26.0 082 */ 083@Beta 084@ElementTypesAreNonnullByDefault 085public final class ExecutionSequencer { 086 087 private ExecutionSequencer() {} 088 089 /** Creates a new instance. */ 090 public static ExecutionSequencer create() { 091 return new ExecutionSequencer(); 092 } 093 094 /** This reference acts as a pointer tracking the head of a linked list of ListenableFutures. */ 095 private final AtomicReference<ListenableFuture<@Nullable Void>> ref = 096 new AtomicReference<>(immediateVoidFuture()); 097 098 private ThreadConfinedTaskQueue latestTaskQueue = new ThreadConfinedTaskQueue(); 099 100 /** 101 * This object is unsafely published, but avoids problematic races by relying exclusively on the 102 * identity equality of its Thread field so that the task field is only accessed by a single 103 * thread. 104 */ 105 private static final class ThreadConfinedTaskQueue { 106 /** 107 * This field is only used for identity comparisons with the current thread. Field assignments 108 * are atomic, but do not provide happens-before ordering; however: 109 * 110 * <ul> 111 * <li>If this field's value == currentThread, we know that it's up to date, because write 112 * operations in a thread always happen-before subsequent read operations in the same 113 * thread 114 * <li>If this field's value == null because of unsafe publication, we know that it isn't the 115 * object associated with our thread, because if it was the publication wouldn't have been 116 * unsafe and we'd have seen our thread as the value. This state is also why a new 117 * ThreadConfinedTaskQueue object must be created for each inline execution, because 118 * observing a null thread does not mean the object is safe to reuse. 119 * <li>If this field's value is some other thread object, we know that it's not our thread. 120 * <li>If this field's value == null because it originally belonged to another thread and that 121 * thread cleared it, we still know that it's not associated with our thread 122 * <li>If this field's value == null because it was associated with our thread and was 123 * cleared, we know that we're not executing inline any more 124 * </ul> 125 * 126 * All the states where thread != currentThread are identical for our purposes, and so even 127 * though it's racy, we don't care which of those values we get, so no need to synchronize. 128 */ 129 @CheckForNull Thread thread; 130 /** Only used by the thread associated with this object */ 131 @CheckForNull Runnable nextTask; 132 /** Only used by the thread associated with this object */ 133 @CheckForNull Executor nextExecutor; 134 } 135 136 /** 137 * Enqueues a task to run when the previous task (if any) completes. 138 * 139 * <p>Cancellation does not propagate from the output future to a callable that has begun to 140 * execute, but if the output future is cancelled before {@link Callable#call()} is invoked, 141 * {@link Callable#call()} will not be invoked. 142 */ 143 public <T extends @Nullable Object> ListenableFuture<T> submit( 144 final Callable<T> callable, Executor executor) { 145 checkNotNull(callable); 146 checkNotNull(executor); 147 return submitAsync( 148 new AsyncCallable<T>() { 149 @Override 150 public ListenableFuture<T> call() throws Exception { 151 return immediateFuture(callable.call()); 152 } 153 154 @Override 155 public String toString() { 156 return callable.toString(); 157 } 158 }, 159 executor); 160 } 161 162 /** 163 * Enqueues a task to run when the previous task (if any) completes. 164 * 165 * <p>Cancellation does not propagate from the output future to the future returned from {@code 166 * callable} or a callable that has begun to execute, but if the output future is cancelled before 167 * {@link AsyncCallable#call()} is invoked, {@link AsyncCallable#call()} will not be invoked. 168 */ 169 public <T extends @Nullable Object> ListenableFuture<T> submitAsync( 170 final AsyncCallable<T> callable, final Executor executor) { 171 checkNotNull(callable); 172 checkNotNull(executor); 173 final TaskNonReentrantExecutor taskExecutor = new TaskNonReentrantExecutor(executor, this); 174 final AsyncCallable<T> task = 175 new AsyncCallable<T>() { 176 @Override 177 public ListenableFuture<T> call() throws Exception { 178 if (!taskExecutor.trySetStarted()) { 179 return immediateCancelledFuture(); 180 } 181 return callable.call(); 182 } 183 184 @Override 185 public String toString() { 186 return callable.toString(); 187 } 188 }; 189 /* 190 * Four futures are at play here: 191 * taskFuture is the future tracking the result of the callable. 192 * newFuture is a future that completes after this and all prior tasks are done. 193 * oldFuture is the previous task's newFuture. 194 * outputFuture is the future we return to the caller, a nonCancellationPropagating taskFuture. 195 * 196 * newFuture is guaranteed to only complete once all tasks previously submitted to this instance 197 * have completed - namely after oldFuture is done, and taskFuture has either completed or been 198 * cancelled before the callable started execution. 199 */ 200 final SettableFuture<@Nullable Void> newFuture = SettableFuture.create(); 201 202 final ListenableFuture<@Nullable Void> oldFuture = ref.getAndSet(newFuture); 203 204 // Invoke our task once the previous future completes. 205 final TrustedListenableFutureTask<T> taskFuture = TrustedListenableFutureTask.create(task); 206 oldFuture.addListener(taskFuture, taskExecutor); 207 208 final ListenableFuture<T> outputFuture = Futures.nonCancellationPropagating(taskFuture); 209 210 // newFuture's lifetime is determined by taskFuture, which can't complete before oldFuture 211 // unless taskFuture is cancelled, in which case it falls back to oldFuture. This ensures that 212 // if the future we return is cancelled, we don't begin execution of the next task until after 213 // oldFuture completes. 214 Runnable listener = 215 new Runnable() { 216 @Override 217 public void run() { 218 if (taskFuture.isDone()) { 219 // Since the value of oldFuture can only ever be immediateFuture(null) or setFuture of 220 // a future that eventually came from immediateFuture(null), this doesn't leak 221 // throwables or completion values. 222 newFuture.setFuture(oldFuture); 223 } else if (outputFuture.isCancelled() && taskExecutor.trySetCancelled()) { 224 // If this CAS succeeds, we know that the provided callable will never be invoked, 225 // so when oldFuture completes it is safe to allow the next submitted task to 226 // proceed. Doing this immediately here lets the next task run without waiting for 227 // the cancelled task's executor to run the noop AsyncCallable. 228 // 229 // --- 230 // 231 // If the CAS fails, the provided callable already started running (or it is about 232 // to). Our contract promises: 233 // 234 // 1. not to execute a new callable until the old one has returned 235 // 236 // If we were to cancel taskFuture, that would let the next task start while the old 237 // one is still running. 238 // 239 // Now, maybe we could tweak our implementation to not start the next task until the 240 // callable actually completes. (We could detect completion in our wrapper 241 // `AsyncCallable task`.) However, our contract also promises: 242 // 243 // 2. not to cancel any Future the user returned from an AsyncCallable 244 // 245 // We promise this because, once we cancel that Future, we would no longer be able to 246 // tell when any underlying work it is doing is done. Thus, we might start a new task 247 // while that underlying work is still running. 248 // 249 // So that is why we cancel only in the case of CAS success. 250 taskFuture.cancel(false); 251 } 252 } 253 }; 254 // Adding the listener to both futures guarantees that newFuture will aways be set. Adding to 255 // taskFuture guarantees completion if the callable is invoked, and adding to outputFuture 256 // propagates cancellation if the callable has not yet been invoked. 257 outputFuture.addListener(listener, directExecutor()); 258 taskFuture.addListener(listener, directExecutor()); 259 260 return outputFuture; 261 } 262 263 enum RunningState { 264 NOT_RUN, 265 CANCELLED, 266 STARTED, 267 } 268 269 /** 270 * This class helps avoid a StackOverflowError when large numbers of tasks are submitted with 271 * {@link MoreExecutors#directExecutor}. Normally, when the first future completes, all the other 272 * tasks would be called recursively. Here, we detect that the delegate executor is executing 273 * inline, and maintain a queue to dispatch tasks iteratively. There is one instance of this class 274 * per call to submit() or submitAsync(), and each instance supports only one call to execute(). 275 * 276 * <p>This class would certainly be simpler and easier to reason about if it were built with 277 * ThreadLocal; however, ThreadLocal is not well optimized for the case where the ThreadLocal is 278 * non-static, and is initialized/removed frequently - this causes churn in the Thread specific 279 * hashmaps. Using a static ThreadLocal to avoid that overhead would mean that different 280 * ExecutionSequencer objects interfere with each other, which would be undesirable, in addition 281 * to increasing the memory footprint of every thread that interacted with it. In order to release 282 * entries in thread-specific maps when the ThreadLocal object itself is no longer referenced, 283 * ThreadLocal is usually implemented with a WeakReference, which can have negative performance 284 * properties; for example, calling WeakReference.get() on Android will block during an 285 * otherwise-concurrent GC cycle. 286 */ 287 private static final class TaskNonReentrantExecutor extends AtomicReference<RunningState> 288 implements Executor, Runnable { 289 290 /** 291 * Used to update and read the latestTaskQueue field. Set to null once the runnable has been run 292 * or queued. 293 */ 294 @CheckForNull ExecutionSequencer sequencer; 295 296 /** 297 * Executor the task was set to run on. Set to null when the task has been queued, run, or 298 * cancelled. 299 */ 300 @CheckForNull Executor delegate; 301 302 /** 303 * Set before calling delegate.execute(); set to null once run, so that it can be GCed; this 304 * object may live on after, if submitAsync returns an incomplete future. 305 */ 306 @CheckForNull Runnable task; 307 308 /** Thread that called execute(). Set in execute, cleared when delegate.execute() returns. */ 309 @CheckForNull Thread submitting; 310 311 private TaskNonReentrantExecutor(Executor delegate, ExecutionSequencer sequencer) { 312 super(NOT_RUN); 313 this.delegate = delegate; 314 this.sequencer = sequencer; 315 } 316 317 @Override 318 public void execute(Runnable task) { 319 // If this operation was successfully cancelled already, calling the runnable will be a noop. 320 // This also avoids a race where if outputFuture is cancelled, it will call taskFuture.cancel, 321 // which will call newFuture.setFuture(oldFuture), to allow the next task in the queue to run 322 // without waiting for the user's executor to run our submitted Runnable. However, this can 323 // interact poorly with the reentrancy-avoiding behavior of this executor - when the operation 324 // before the cancelled future completes, it will synchronously complete both the newFuture 325 // from the cancelled operation and its own. This can cause one runnable to queue two tasks, 326 // breaking the invariant this method relies on to iteratively run the next task after the 327 // previous one completes. 328 if (get() == RunningState.CANCELLED) { 329 delegate = null; 330 sequencer = null; 331 return; 332 } 333 submitting = Thread.currentThread(); 334 335 try { 336 /* 337 * requireNonNull is safe because we don't null out `sequencer` except: 338 * 339 * - above, where we return (in which case we never get here) 340 * 341 * - in `run`, which can't run until this Runnable is submitted to an executor, which 342 * doesn't happen until below. (And this Executor -- yes, the object is both a Runnable 343 * and an Executor -- is used for only a single `execute` call.) 344 */ 345 ThreadConfinedTaskQueue submittingTaskQueue = requireNonNull(sequencer).latestTaskQueue; 346 if (submittingTaskQueue.thread == submitting) { 347 sequencer = null; 348 // Submit from inside a reentrant submit. We don't know if this one will be reentrant (and 349 // can't know without submitting something to the executor) so queue to run iteratively. 350 // Task must be null, since each execution on this executor can only produce one more 351 // execution. 352 checkState(submittingTaskQueue.nextTask == null); 353 submittingTaskQueue.nextTask = task; 354 // requireNonNull(delegate) is safe for reasons similar to requireNonNull(sequencer). 355 submittingTaskQueue.nextExecutor = requireNonNull(delegate); 356 delegate = null; 357 } else { 358 // requireNonNull(delegate) is safe for reasons similar to requireNonNull(sequencer). 359 Executor localDelegate = requireNonNull(delegate); 360 delegate = null; 361 this.task = task; 362 localDelegate.execute(this); 363 } 364 } finally { 365 // Important to null this out here - if we did *not* execute inline, we might still 366 // run() on the same thread that called execute() - such as in a thread pool, and think 367 // that it was happening inline. As a side benefit, avoids holding on to the Thread object 368 // longer than necessary. 369 submitting = null; 370 } 371 } 372 373 @SuppressWarnings("ShortCircuitBoolean") 374 @Override 375 public void run() { 376 Thread currentThread = Thread.currentThread(); 377 if (currentThread != submitting) { 378 /* 379 * requireNonNull is safe because we set `task` before submitting this Runnable to an 380 * Executor, and we don't null it out until here. 381 */ 382 Runnable localTask = requireNonNull(task); 383 task = null; 384 localTask.run(); 385 return; 386 } 387 // Executor called reentrantly! Make sure that further calls don't overflow stack. Further 388 // reentrant calls will see that their current thread is the same as the one set in 389 // latestTaskQueue, and queue rather than calling execute() directly. 390 ThreadConfinedTaskQueue executingTaskQueue = new ThreadConfinedTaskQueue(); 391 executingTaskQueue.thread = currentThread; 392 /* 393 * requireNonNull is safe because we don't null out `sequencer` except: 394 * 395 * - after the requireNonNull call below. (And this object has its Runnable.run override 396 * called only once, just as it has its Executor.execute override called only once.) 397 * 398 * - if we return immediately from `execute` (in which case we never get here) 399 * 400 * - in the "reentrant submit" case of `execute` (in which case we must have started running a 401 * user task -- which means that we already got past this code (or else we exited early 402 * above)) 403 */ 404 // Unconditionally set; there is no risk of throwing away a queued task from another thread, 405 // because in order for the current task to run on this executor the previous task must have 406 // already started execution. Because each task on a TaskNonReentrantExecutor can only produce 407 // one execute() call to another instance from the same ExecutionSequencer, we know by 408 // induction that the task that launched this one must not have added any other runnables to 409 // that thread's queue, and thus we cannot be replacing a TaskAndThread object that would 410 // otherwise have another task queued on to it. Note the exception to this, cancellation, is 411 // specially handled in execute() - execute() calls triggered by cancellation are no-ops, and 412 // thus don't count. 413 requireNonNull(sequencer).latestTaskQueue = executingTaskQueue; 414 sequencer = null; 415 try { 416 // requireNonNull is safe, as discussed above. 417 Runnable localTask = requireNonNull(task); 418 task = null; 419 localTask.run(); 420 // Now check if our task attempted to reentrantly execute the next task. 421 Runnable queuedTask; 422 Executor queuedExecutor; 423 // Intentionally using non-short-circuit operator 424 while ((queuedTask = executingTaskQueue.nextTask) != null 425 & (queuedExecutor = executingTaskQueue.nextExecutor) != null) { 426 executingTaskQueue.nextTask = null; 427 executingTaskQueue.nextExecutor = null; 428 queuedExecutor.execute(queuedTask); 429 } 430 } finally { 431 // Null out the thread field, so that we don't leak a reference to Thread, and so that 432 // future `thread == currentThread()` calls from this thread don't incorrectly queue instead 433 // of executing. Don't null out the latestTaskQueue field, because the work done here 434 // may have scheduled more operations on another thread, and if those operations then 435 // trigger reentrant calls that thread will have updated the latestTaskQueue field, and 436 // we'd be interfering with their operation. 437 executingTaskQueue.thread = null; 438 } 439 } 440 441 private boolean trySetStarted() { 442 return compareAndSet(NOT_RUN, STARTED); 443 } 444 445 private boolean trySetCancelled() { 446 return compareAndSet(NOT_RUN, CANCELLED); 447 } 448 } 449}