1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215
//! A thread pool for isolating blocking I/O in async programs. //! //! Sometimes there's no way to avoid blocking I/O. Consider files or stdin, which have weak async //! support on modern operating systems. While [IOCP], [AIO], and [io_uring] are possible //! solutions, they're not always available or ideal. //! //! Since blocking is not allowed inside futures, we must move blocking I/O onto a special thread //! pool provided by this crate. The pool dynamically spawns and stops threads depending on the //! current number of running I/O jobs. //! //! Note that there is a limit on the number of active threads. Once that limit is hit, a running //! job has to finish before others get a chance to run. When a thread is idle, it waits for the //! next job or shuts down after a certain timeout. //! //! [IOCP]: https://en.wikipedia.org/wiki/Input/output_completion_port //! [AIO]: http://man7.org/linux/man-pages/man2/io_submit.2.html //! [io_uring]: https://lwn.net/Articles/776703 //! //! # Examples //! //! Read the contents of a file: //! //! ```no_run //! use blocking::unblock; //! use std::fs; //! //! # futures_lite::future::block_on(async { //! let contents = unblock(|| fs::read_to_string("file.txt")).await?; //! println!("{}", contents); //! # std::io::Result::Ok(()) }); //! ``` //! //! Read a file and pipe its contents to stdout: //! //! ```no_run //! use blocking::{unblock, Unblock}; //! use futures_lite::io; //! use std::fs::File; //! //! # futures_lite::future::block_on(async { //! let input = unblock(|| File::open("file.txt")).await?; //! let input = Unblock::new(input); //! let mut output = Unblock::new(std::io::stdout()); //! //! io::copy(input, &mut output).await?; //! # std::io::Result::Ok(()) }); //! ``` //! //! Iterate over the contents of a directory: //! //! ```no_run //! use blocking::Unblock; //! use futures_lite::prelude::*; //! use std::fs; //! //! # futures_lite::future::block_on(async { //! let mut dir = Unblock::new(fs::read_dir(".")?); //! while let Some(item) = dir.next().await { //! println!("{}", item?.file_name().to_string_lossy()); //! } //! # std::io::Result::Ok(()) }); //! ``` //! //! Spawn a process: //! //! ```no_run //! use blocking::unblock; //! use std::process::Command; //! //! # futures_lite::future::block_on(async { //! let out = unblock(|| Command::new("dir").output()).await?; //! # std::io::Result::Ok(()) }); //! ``` #![warn(missing_docs, missing_debug_implementations, rust_2018_idioms)] use std::any::Any; use std::collections::VecDeque; use std::fmt; use std::io::{self, Read, Seek, SeekFrom, Write}; use std::mem; use std::panic; use std::pin::Pin; use std::slice; use std::sync::atomic::{self, AtomicBool, AtomicUsize, Ordering}; use std::sync::{Arc, Condvar, Mutex, MutexGuard}; use std::task::{Context, Poll}; use std::thread; use std::time::Duration; use async_channel::{bounded, Receiver}; use async_task::{Runnable, Task}; use atomic_waker::AtomicWaker; use futures_lite::{future, prelude::*, ready}; use once_cell::sync::Lazy; /// Lazily initialized global executor. static EXECUTOR: Lazy<Executor> = Lazy::new(|| Executor { inner: Mutex::new(Inner { idle_count: 0, thread_count: 0, queue: VecDeque::new(), }), cvar: Condvar::new(), }); /// The blocking executor. struct Executor { /// Inner state of the executor. inner: Mutex<Inner>, /// Used to put idle threads to sleep and wake them up when new work comes in. cvar: Condvar, } /// Inner state of the blocking executor. struct Inner { /// Number of idle threads in the pool. /// /// Idle threads are sleeping, waiting to get a task to run. idle_count: usize, /// Total number of threads in the pool. /// /// This is the number of idle threads + the number of active threads. thread_count: usize, /// The queue of blocking tasks. queue: VecDeque<Runnable>, } impl Executor { /// Spawns a future onto this executor. /// /// Returns a [`Task`] handle for the spawned task. fn spawn<T: Send + 'static>(future: impl Future<Output = T> + Send + 'static) -> Task<T> { let (runnable, task) = async_task::spawn(future, |r| EXECUTOR.schedule(r)); runnable.schedule(); task } /// Runs the main loop on the current thread. /// /// This function runs blocking tasks until it becomes idle and times out. fn main_loop(&'static self) { let mut inner = self.inner.lock().unwrap(); loop { // This thread is not idle anymore because it's going to run tasks. inner.idle_count -= 1; // Run tasks in the queue. while let Some(runnable) = inner.queue.pop_front() { // We have found a task - grow the pool if needed. self.grow_pool(inner); // Run the task. panic::catch_unwind(|| runnable.run()).ok(); // Re-lock the inner state and continue. inner = self.inner.lock().unwrap(); } // This thread is now becoming idle. inner.idle_count += 1; // Put the thread to sleep until another task is scheduled. let timeout = Duration::from_millis(500); let (lock, res) = self.cvar.wait_timeout(inner, timeout).unwrap(); inner = lock; // If there are no tasks after a while, stop this thread. if res.timed_out() && inner.queue.is_empty() { inner.idle_count -= 1; inner.thread_count -= 1; break; } } } /// Schedules a runnable task for execution. fn schedule(&'static self, runnable: Runnable) { let mut inner = self.inner.lock().unwrap(); inner.queue.push_back(runnable); // Notify a sleeping thread and spawn more threads if needed. self.cvar.notify_one(); self.grow_pool(inner); } /// Spawns more blocking threads if the pool is overloaded with work. fn grow_pool(&'static self, mut inner: MutexGuard<'static, Inner>) { // If runnable tasks greatly outnumber idle threads and there aren't too many threads // already, then be aggressive: wake all idle threads and spawn one more thread. while inner.queue.len() > inner.idle_count * 5 && inner.thread_count < 500 { // The new thread starts in idle state. inner.idle_count += 1; inner.thread_count += 1; // Notify all existing idle threads because we need to hurry up. self.cvar.notify_all(); // Generate a new thread ID. static ID: AtomicUsize = AtomicUsize::new(1); let id = ID.fetch_add(1, Ordering::Relaxed); // Spawn the new thread. thread::Builder::new() .name(format!("blocking-{}", id)) .spawn(move || self.main_loop()) .unwrap(); } } } /// Runs blocking code on a thread pool. /// /// # Examples /// /// Read the contents of a file: /// /// ```no_run /// use blocking::unblock; /// use std::fs; /// /// # futures_lite::future::block_on(async { /// let contents = unblock(|| fs::read_to_string("file.txt")).await?; /// # std::io::Result::Ok(()) }); /// ``` /// /// Spawn a process: /// /// ```no_run /// use blocking::unblock; /// use std::process::Command; /// /// # futures_lite::future::block_on(async { /// let out = unblock(|| Command::new("dir").output()).await?; /// # std::io::Result::Ok(()) }); /// ``` pub async fn unblock<T, F>(f: F) -> T where F: FnOnce() -> T + Send + 'static, T: Send + 'static, { Executor::spawn(async move { f() }).await } /// Runs blocking I/O on a thread pool. /// /// Blocking I/O must be isolated from async code. This type moves blocking I/O operations onto a /// special thread pool while exposing a familiar async interface. /// /// This type implements traits [`Stream`], [`AsyncRead`], [`AsyncWrite`], or [`AsyncSeek`] if the /// inner type implements [`Iterator`], [`Read`], [`Write`], or [`Seek`], respectively. /// /// # Caveats /// /// [`Unblock`] is a low-level primitive, and as such it comes with some caveats. /// /// For higher-level primitives built on top of [`Unblock`], look into [`async-fs`] or /// [`async-process`] (on Windows). /// /// [`async-fs`]: https://github.com/stjepang/async-fs /// [`async-process`]: https://github.com/stjepang/async-process /// /// [`Unblock`] communicates with I/O operations on the thread pool through a pipe. That means an /// async read/write operation simply receives/sends some bytes from/into the pipe. When in reading /// mode, the thread pool reads bytes from the I/O handle and forwards them into the pipe until it /// becomes full. When in writing mode, the thread pool reads bytes from the pipe and forwards them /// into the I/O handle. /// /// Use [`Unblock::with_capacity()`] to configure the capacity of the pipe. /// /// ### Reading /// /// If you create an [`Unblock`]`<`[`Stdin`][`std::io::Stdin`]`>`, read some bytes from it, /// and then drop it, a blocked read operation may keep hanging on the thread pool. The next /// attempt to read from stdin will lose bytes read by the hanging operation. This is a difficult /// problem to solve, so make sure you only use a single stdin handle for the duration of the /// entire program. /// /// ### Writing /// /// If writing data through the [`AsyncWrite`] trait, make sure to flush before dropping the /// [`Unblock`] handle or some buffered data might get lost. /// /// ### Seeking /// /// Because of buffering in the pipe, if [`Unblock`] wraps a [`File`][`std::fs::File`], a single /// read operation may move the file cursor farther than is the span of the operation. In fact, /// reading just keeps going in the background until the pipe gets full. Keep this mind when /// using [`AsyncSeek`] with [relative][`SeekFrom::Current`] offsets. /// /// # Examples /// /// ``` /// use blocking::Unblock; /// use futures_lite::prelude::*; /// /// # futures_lite::future::block_on(async { /// let mut stdout = Unblock::new(std::io::stdout()); /// stdout.write_all(b"Hello world!").await?; /// stdout.flush().await?; /// # std::io::Result::Ok(()) }); /// ``` pub struct Unblock<T> { state: State<T>, cap: Option<usize>, } impl<T> Unblock<T> { /// Wraps a blocking I/O handle into the async [`Unblock`] interface. /// /// # Examples /// /// ```no_run /// use blocking::Unblock; /// /// let stdin = Unblock::new(std::io::stdin()); /// ``` pub fn new(io: T) -> Unblock<T> { Unblock { state: State::Idle(Some(Box::new(io))), cap: None, } } /// Wraps a blocking I/O handle into the async [`Unblock`] interface with a custom buffer /// capacity. /// /// When communicating with the inner [`Stream`]/[`Read`]/[`Write`] type from async code, data /// transferred between blocking and async code goes through a buffer of limited capacity. This /// constructor configures that capacity. /// /// The default capacity is: /// /// * For [`Iterator`] types: 8192 items. /// * For [`Read`]/[`Write`] types: 8 MB. /// /// # Examples /// /// ```no_run /// use blocking::Unblock; /// /// let stdout = Unblock::with_capacity(64 * 1024, std::io::stdout()); /// ``` pub fn with_capacity(cap: usize, io: T) -> Unblock<T> { Unblock { state: State::Idle(Some(Box::new(io))), cap: Some(cap), } } /// Gets a mutable reference to the blocking I/O handle. /// /// This is an async method because the I/O handle might be on the thread pool and needs to /// be moved onto the current thread before we can get a reference to it. /// /// # Examples /// /// ```no_run /// use blocking::{unblock, Unblock}; /// use std::fs::File; /// /// # futures_lite::future::block_on(async { /// let file = unblock(|| File::create("file.txt")).await?; /// let mut file = Unblock::new(file); /// /// let metadata = file.get_mut().await.metadata()?; /// # std::io::Result::Ok(()) }); /// ``` pub async fn get_mut(&mut self) -> &mut T { // Wait for the running task to stop and ignore I/O errors if there are any. future::poll_fn(|cx| self.poll_stop(cx)).await.ok(); // Assume idle state and get a reference to the inner value. match &mut self.state { State::Idle(t) => t.as_mut().expect("inner value was taken out"), State::WithMut(..) | State::Streaming(..) | State::Reading(..) | State::Writing(..) | State::Seeking(..) => { unreachable!("when stopped, the state machine must be in idle state"); } } } /// Performs a blocking operation on the I/O handle. /// /// # Examples /// /// ```no_run /// use blocking::{unblock, Unblock}; /// use std::fs::File; /// /// # futures_lite::future::block_on(async { /// let file = unblock(|| File::create("file.txt")).await?; /// let mut file = Unblock::new(file); /// /// let metadata = file.with_mut(|f| f.metadata()).await?; /// # std::io::Result::Ok(()) }); /// ``` pub async fn with_mut<R, F>(&mut self, op: F) -> R where F: FnOnce(&mut T) -> R + Send + 'static, R: Send + 'static, T: Send + 'static, { // Wait for the running task to stop and ignore I/O errors if there are any. future::poll_fn(|cx| self.poll_stop(cx)).await.ok(); // Assume idle state and take out the inner value. let mut t = match &mut self.state { State::Idle(t) => t.take().expect("inner value was taken out"), State::WithMut(..) | State::Streaming(..) | State::Reading(..) | State::Writing(..) | State::Seeking(..) => { unreachable!("when stopped, the state machine must be in idle state"); } }; let (sender, receiver) = bounded(1); let task = Executor::spawn(async move { sender.try_send(op(&mut t)).ok(); t }); self.state = State::WithMut(task); receiver .recv() .await .expect("`Unblock::with_mut()` operation has panicked") } /// Extracts the inner blocking I/O handle. /// /// This is an async method because the I/O handle might be on the thread pool and needs to /// be moved onto the current thread before we can extract it. /// /// # Examples /// /// ```no_run /// use blocking::{unblock, Unblock}; /// use futures_lite::prelude::*; /// use std::fs::File; /// /// # futures_lite::future::block_on(async { /// let file = unblock(|| File::create("file.txt")).await?; /// let file = Unblock::new(file); /// /// let file = file.into_inner().await; /// # std::io::Result::Ok(()) }); /// ``` pub async fn into_inner(self) -> T { // There's a bug in rustdoc causing it to render `mut self` as `__arg0: Self`, so we just // bind `self` to a local mutable variable. let mut this = self; // Wait for the running task to stop and ignore I/O errors if there are any. future::poll_fn(|cx| this.poll_stop(cx)).await.ok(); // Assume idle state and extract the inner value. match &mut this.state { State::Idle(t) => *t.take().expect("inner value was taken out"), State::WithMut(..) | State::Streaming(..) | State::Reading(..) | State::Writing(..) | State::Seeking(..) => { unreachable!("when stopped, the state machine must be in idle state"); } } } /// Waits for the running task to stop. /// /// On success, the state machine is moved into the idle state. fn poll_stop(&mut self, cx: &mut Context<'_>) -> Poll<io::Result<()>> { loop { match &mut self.state { State::Idle(_) => return Poll::Ready(Ok(())), State::WithMut(task) => { // Poll the task to wait for it to finish. let io = ready!(Pin::new(task).poll(cx)); self.state = State::Idle(Some(io)); } State::Streaming(any, task) => { // Drop the receiver to close the channel. This stops the `send()` operation in // the task, after which the task returns the iterator back. any.take(); // Poll the task to retrieve the iterator. let iter = ready!(Pin::new(task).poll(cx)); self.state = State::Idle(Some(iter)); } State::Reading(reader, task) => { // Drop the reader to close the pipe. This stops copying inside the task, after // which the task returns the I/O handle back. reader.take(); // Poll the task to retrieve the I/O handle. let (res, io) = ready!(Pin::new(task).poll(cx)); // Make sure to move into the idle state before reporting errors. self.state = State::Idle(Some(io)); res?; } State::Writing(writer, task) => { // Drop the writer to close the pipe. This stops copying inside the task, after // which the task flushes the I/O handle and writer.take(); // Poll the task to retrieve the I/O handle. let (res, io) = ready!(Pin::new(task).poll(cx)); // Make sure to move into the idle state before reporting errors. self.state = State::Idle(Some(io)); res?; } State::Seeking(task) => { // Poll the task to wait for it to finish. let (_, res, io) = ready!(Pin::new(task).poll(cx)); // Make sure to move into the idle state before reporting errors. self.state = State::Idle(Some(io)); res?; } } } } } impl<T: fmt::Debug> fmt::Debug for Unblock<T> { fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { struct Closed; impl fmt::Debug for Closed { fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { f.write_str("<closed>") } } struct Blocked; impl fmt::Debug for Blocked { fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { f.write_str("<blocked>") } } match &self.state { State::Idle(None) => f.debug_struct("Unblock").field("io", &Closed).finish(), State::Idle(Some(io)) => { let io: &T = &*io; f.debug_struct("Unblock").field("io", io).finish() } State::WithMut(..) | State::Streaming(..) | State::Reading(..) | State::Writing(..) | State::Seeking(..) => f.debug_struct("Unblock").field("io", &Blocked).finish(), } } } /// Current state of a blocking task. enum State<T> { /// There is no blocking task. /// /// The inner value is readily available, unless it has already been extracted. The value is /// extracted out by [`Unblock::into_inner()`], [`AsyncWrite::poll_close()`], or by awaiting /// [`Unblock`]. Idle(Option<Box<T>>), /// A [`Unblock::with_mut()`] closure was spawned and is still running. WithMut(Task<Box<T>>), /// The inner value is an [`Iterator`] currently iterating in a task. /// /// The `dyn Any` value here is a `mpsc::Receiver<<T as Iterator>::Item>`. Streaming(Option<Box<dyn Any + Send + Sync>>, Task<Box<T>>), /// The inner value is a [`Read`] currently reading in a task. Reading(Option<Reader>, Task<(io::Result<()>, Box<T>)>), /// The inner value is a [`Write`] currently writing in a task. Writing(Option<Writer>, Task<(io::Result<()>, Box<T>)>), /// The inner value is a [`Seek`] currently seeking in a task. Seeking(Task<(SeekFrom, io::Result<u64>, Box<T>)>), } impl<T: Iterator + Send + 'static> Stream for Unblock<T> where T::Item: Send + 'static, { type Item = T::Item; fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<T::Item>> { loop { match &mut self.state { // If not in idle or active streaming state, stop the running task. State::WithMut(..) | State::Streaming(None, _) | State::Reading(..) | State::Writing(..) | State::Seeking(..) => { // Wait for the running task to stop. ready!(self.poll_stop(cx)).ok(); } // If idle, start a streaming task. State::Idle(iter) => { // Take the iterator out to run it on a blocking task. let mut iter = iter.take().expect("inner iterator was taken out"); // This channel capacity seems to work well in practice. If it's too low, there // will be too much synchronization between tasks. If too high, memory // consumption increases. let (sender, receiver) = bounded(self.cap.unwrap_or(8 * 1024)); // 8192 items // Spawn a blocking task that runs the iterator and returns it when done. let task = Executor::spawn(async move { for item in &mut iter { if sender.send(item).await.is_err() { break; } } iter }); // Move into the busy state and poll again. self.state = State::Streaming(Some(Box::new(receiver)), task); } // If streaming, receive an item. State::Streaming(Some(any), task) => { let receiver = any.downcast_mut::<Receiver<T::Item>>().unwrap(); // Poll the channel. let opt = ready!(Pin::new(receiver).poll_next(cx)); // If the channel is closed, retrieve the iterator back from the blocking task. // This is not really a required step, but it's cleaner to drop the iterator on // the same thread that created it. if opt.is_none() { // Poll the task to retrieve the iterator. let iter = ready!(Pin::new(task).poll(cx)); self.state = State::Idle(Some(iter)); } return Poll::Ready(opt); } } } } } impl<T: Read + Send + 'static> AsyncRead for Unblock<T> { fn poll_read( mut self: Pin<&mut Self>, cx: &mut Context<'_>, buf: &mut [u8], ) -> Poll<io::Result<usize>> { loop { match &mut self.state { // If not in idle or active reading state, stop the running task. State::WithMut(..) | State::Reading(None, _) | State::Streaming(..) | State::Writing(..) | State::Seeking(..) => { // Wait for the running task to stop. ready!(self.poll_stop(cx))?; } // If idle, start a reading task. State::Idle(io) => { // Take the I/O handle out to read it on a blocking task. let mut io = io.take().expect("inner value was taken out"); // This pipe capacity seems to work well in practice. If it's too low, there // will be too much synchronization between tasks. If too high, memory // consumption increases. let (reader, mut writer) = pipe(self.cap.unwrap_or(8 * 1024 * 1024)); // 8 MB // Spawn a blocking task that reads and returns the I/O handle when done. let task = Executor::spawn(async move { // Copy bytes from the I/O handle into the pipe until the pipe is closed or // an error occurs. loop { match future::poll_fn(|cx| writer.fill(cx, &mut io)).await { Ok(0) => return (Ok(()), io), Ok(_) => {} Err(err) => return (Err(err), io), } } }); // Move into the busy state and poll again. self.state = State::Reading(Some(reader), task); } // If reading, read bytes from the pipe. State::Reading(Some(reader), task) => { // Poll the pipe. let n = ready!(reader.drain(cx, buf))?; // If the pipe is closed, retrieve the I/O handle back from the blocking task. // This is not really a required step, but it's cleaner to drop the handle on // the same thread that created it. if n == 0 { // Poll the task to retrieve the I/O handle. let (res, io) = ready!(Pin::new(task).poll(cx)); // Make sure to move into the idle state before reporting errors. self.state = State::Idle(Some(io)); res?; } return Poll::Ready(Ok(n)); } } } } } impl<T: Write + Send + 'static> AsyncWrite for Unblock<T> { fn poll_write( mut self: Pin<&mut Self>, cx: &mut Context<'_>, buf: &[u8], ) -> Poll<io::Result<usize>> { loop { match &mut self.state { // If not in idle or active writing state, stop the running task. State::WithMut(..) | State::Writing(None, _) | State::Streaming(..) | State::Reading(..) | State::Seeking(..) => { // Wait for the running task to stop. ready!(self.poll_stop(cx))?; } // If idle, start the writing task. State::Idle(io) => { // Take the I/O handle out to write on a blocking task. let mut io = io.take().expect("inner value was taken out"); // This pipe capacity seems to work well in practice. If it's too low, there will // be too much synchronization between tasks. If too high, memory consumption // increases. let (mut reader, writer) = pipe(self.cap.unwrap_or(8 * 1024 * 1024)); // 8 MB // Spawn a blocking task that writes and returns the I/O handle when done. let task = Executor::spawn(async move { // Copy bytes from the pipe into the I/O handle until the pipe is closed or an // error occurs. Flush the I/O handle at the end. loop { match future::poll_fn(|cx| reader.drain(cx, &mut io)).await { Ok(0) => return (io.flush(), io), Ok(_) => {} Err(err) => { io.flush().ok(); return (Err(err), io); } } } }); // Move into the busy state and poll again. self.state = State::Writing(Some(writer), task); } // If writing, write more bytes into the pipe. State::Writing(Some(writer), _) => return writer.fill(cx, buf), } } } fn poll_flush(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> { loop { match &mut self.state { // If not in idle state, stop the running task. State::WithMut(..) | State::Streaming(..) | State::Writing(..) | State::Reading(..) | State::Seeking(..) => { // Wait for the running task to stop. ready!(self.poll_stop(cx))?; } // Idle implies flushed. State::Idle(_) => return Poll::Ready(Ok(())), } } } fn poll_close(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> { // First, make sure the I/O handle is flushed. ready!(Pin::new(&mut self).poll_flush(cx))?; // Then move into the idle state with no I/O handle, thus dropping it. self.state = State::Idle(None); Poll::Ready(Ok(())) } } impl<T: Seek + Send + 'static> AsyncSeek for Unblock<T> { fn poll_seek( mut self: Pin<&mut Self>, cx: &mut Context<'_>, pos: SeekFrom, ) -> Poll<io::Result<u64>> { loop { match &mut self.state { // If not in idle state, stop the running task. State::WithMut(..) | State::Streaming(..) | State::Reading(..) | State::Writing(..) => { // Wait for the running task to stop. ready!(self.poll_stop(cx))?; } State::Idle(io) => { // Take the I/O handle out to seek on a blocking task. let mut io = io.take().expect("inner value was taken out"); let task = Executor::spawn(async move { let res = io.seek(pos); (pos, res, io) }); self.state = State::Seeking(task); } State::Seeking(task) => { // Poll the task to wait for it to finish. let (original_pos, res, io) = ready!(Pin::new(task).poll(cx)); // Make sure to move into the idle state before reporting errors. self.state = State::Idle(Some(io)); let current = res?; // If the `pos` argument matches the original one, return the result. if original_pos == pos { return Poll::Ready(Ok(current)); } } } } } } /// Creates a bounded single-producer single-consumer pipe. /// /// A pipe is a ring buffer of `cap` bytes that can be asynchronously read from and written to. /// /// When the sender is dropped, remaining bytes in the pipe can still be read. After that, attempts /// to read will result in `Ok(0)`, i.e. they will always 'successfully' read 0 bytes. /// /// When the receiver is dropped, the pipe is closed and no more bytes and be written into it. /// Further writes will result in `Ok(0)`, i.e. they will always 'successfully' write 0 bytes. fn pipe(cap: usize) -> (Reader, Writer) { assert!(cap > 0, "capacity must be positive"); assert!(cap.checked_mul(2).is_some(), "capacity is too large"); // Allocate the ring buffer. let mut v = Vec::with_capacity(cap); let buffer = v.as_mut_ptr(); mem::forget(v); let inner = Arc::new(Pipe { head: AtomicUsize::new(0), tail: AtomicUsize::new(0), reader: AtomicWaker::new(), writer: AtomicWaker::new(), closed: AtomicBool::new(false), buffer, cap, }); let r = Reader { inner: inner.clone(), head: 0, tail: 0, }; let w = Writer { inner, head: 0, tail: 0, zeroed_until: 0, }; (r, w) } /// The reading side of a pipe. struct Reader { /// The inner ring buffer. inner: Arc<Pipe>, /// The head index, moved by the reader, in the range `0..2*cap`. /// /// This index always matches `inner.head`. head: usize, /// The tail index, moved by the writer, in the range `0..2*cap`. /// /// This index is a snapshot of `index.tail` that might become stale at any point. tail: usize, } /// The writing side of a pipe. struct Writer { /// The inner ring buffer. inner: Arc<Pipe>, /// The head index, moved by the reader, in the range `0..2*cap`. /// /// This index is a snapshot of `index.head` that might become stale at any point. head: usize, /// The tail index, moved by the writer, in the range `0..2*cap`. /// /// This index always matches `inner.tail`. tail: usize, /// How many bytes at the beginning of the buffer have been zeroed. /// /// The pipe allocates an uninitialized buffer, and we must be careful about passing /// uninitialized data to user code. Zeroing the buffer right after allocation would be too /// expensive, so we zero it in smaller chunks as the writer makes progress. zeroed_until: usize, } unsafe impl Send for Reader {} unsafe impl Send for Writer {} /// The inner ring buffer. /// /// Head and tail indices are in the range `0..2*cap`, even though they really map onto the /// `0..cap` range. The distance between head and tail indices is never more than `cap`. /// /// The reason why indices are not in the range `0..cap` is because we need to distinguish between /// the pipe being empty and being full. If head and tail were in `0..cap`, then `head == tail` /// could mean the pipe is either empty or full, but we don't know which! struct Pipe { /// The head index, moved by the reader, in the range `0..2*cap`. head: AtomicUsize, /// The tail index, moved by the writer, in the range `0..2*cap`. tail: AtomicUsize, /// A waker representing the blocked reader. reader: AtomicWaker, /// A waker representing the blocked writer. writer: AtomicWaker, /// Set to `true` if the reader or writer was dropped. closed: AtomicBool, /// The byte buffer. buffer: *mut u8, /// The buffer capacity. cap: usize, } unsafe impl Sync for Pipe {} unsafe impl Send for Pipe {} impl Drop for Pipe { fn drop(&mut self) { // Deallocate the byte buffer. unsafe { Vec::from_raw_parts(self.buffer, 0, self.cap); } } } impl Drop for Reader { fn drop(&mut self) { // Dropping closes the pipe and then wakes the writer. self.inner.closed.store(true, Ordering::SeqCst); self.inner.writer.wake(); } } impl Drop for Writer { fn drop(&mut self) { // Dropping closes the pipe and then wakes the reader. self.inner.closed.store(true, Ordering::SeqCst); self.inner.reader.wake(); } } impl Reader { /// Reads bytes from this reader and writes into blocking `dest`. fn drain(&mut self, cx: &mut Context<'_>, mut dest: impl Write) -> Poll<io::Result<usize>> { let cap = self.inner.cap; // Calculates the distance between two indices. let distance = |a: usize, b: usize| { if a <= b { b - a } else { 2 * cap - (a - b) } }; // If the pipe appears to be empty... if distance(self.head, self.tail) == 0 { // Reload the tail in case it's become stale. self.tail = self.inner.tail.load(Ordering::Acquire); // If the pipe is now really empty... if distance(self.head, self.tail) == 0 { // Register the waker. self.inner.reader.register(cx.waker()); atomic::fence(Ordering::SeqCst); // Reload the tail after registering the waker. self.tail = self.inner.tail.load(Ordering::Acquire); // If the pipe is still empty... if distance(self.head, self.tail) == 0 { // Check whether the pipe is closed or just empty. if self.inner.closed.load(Ordering::Relaxed) { return Poll::Ready(Ok(0)); } else { return Poll::Pending; } } } } // The pipe is not empty so remove the waker. self.inner.reader.take(); // Yield with some small probability - this improves fairness. ready!(maybe_yield(cx)); // Given an index in `0..2*cap`, returns the real index in `0..cap`. let real_index = |i: usize| { if i < cap { i } else { i - cap } }; // Number of bytes read so far. let mut count = 0; loop { // Calculate how many bytes to read in this iteration. let n = (128 * 1024) // Not too many bytes in one go - better to wake the writer soon! .min(distance(self.head, self.tail)) // No more than bytes in the pipe. .min(cap - real_index(self.head)); // Don't go past the buffer boundary. // Create a slice of data in the pipe buffer. let pipe_slice = unsafe { slice::from_raw_parts(self.inner.buffer.add(real_index(self.head)), n) }; // Copy bytes from the pipe buffer into `dest`. let n = dest.write(pipe_slice)?; count += n; // If pipe is empty or `dest` is full, return. if n == 0 { return Poll::Ready(Ok(count)); } // Move the head forward. if self.head + n < 2 * cap { self.head += n; } else { self.head = 0; } // Store the current head index. self.inner.head.store(self.head, Ordering::Release); // Wake the writer because the pipe is not full. self.inner.writer.wake(); } } } impl Writer { /// Reads bytes from blocking `src` and writes into this writer. fn fill(&mut self, cx: &mut Context<'_>, mut src: impl Read) -> Poll<io::Result<usize>> { // Just a quick check if the pipe is closed, which is why a relaxed load is okay. if self.inner.closed.load(Ordering::Relaxed) { return Poll::Ready(Ok(0)); } // Calculates the distance between two indices. let cap = self.inner.cap; let distance = |a: usize, b: usize| { if a <= b { b - a } else { 2 * cap - (a - b) } }; // If the pipe appears to be full... if distance(self.head, self.tail) == cap { // Reload the head in case it's become stale. self.head = self.inner.head.load(Ordering::Acquire); // If the pipe is now really empty... if distance(self.head, self.tail) == cap { // Register the waker. self.inner.writer.register(cx.waker()); atomic::fence(Ordering::SeqCst); // Reload the head after registering the waker. self.head = self.inner.head.load(Ordering::Acquire); // If the pipe is still full... if distance(self.head, self.tail) == cap { // Check whether the pipe is closed or just full. if self.inner.closed.load(Ordering::Relaxed) { return Poll::Ready(Ok(0)); } else { return Poll::Pending; } } } } // The pipe is not full so remove the waker. self.inner.writer.take(); // Yield with some small probability - this improves fairness. ready!(maybe_yield(cx)); // Given an index in `0..2*cap`, returns the real index in `0..cap`. let real_index = |i: usize| { if i < cap { i } else { i - cap } }; // Number of bytes written so far. let mut count = 0; loop { // Calculate how many bytes to write in this iteration. let n = (128 * 1024) // Not too many bytes in one go - better to wake the reader soon! .min(self.zeroed_until * 2 + 4096) // Don't zero too many bytes when starting. .min(cap - distance(self.head, self.tail)) // No more than space in the pipe. .min(cap - real_index(self.tail)); // Don't go past the buffer boundary. // Create a slice of available space in the pipe buffer. let pipe_slice_mut = unsafe { let from = real_index(self.tail); let to = from + n; // Make sure all bytes in the slice are initialized. if self.zeroed_until < to { self.inner .buffer .add(self.zeroed_until) .write_bytes(0u8, to - self.zeroed_until); self.zeroed_until = to; } slice::from_raw_parts_mut(self.inner.buffer.add(from), n) }; // Copy bytes from `src` into the piper buffer. let n = src.read(pipe_slice_mut)?; count += n; // If the pipe is full or closed, or `src` is empty, return. if n == 0 || self.inner.closed.load(Ordering::Relaxed) { return Poll::Ready(Ok(count)); } // Move the tail forward. if self.tail + n < 2 * cap { self.tail += n; } else { self.tail = 0; } // Store the current tail index. self.inner.tail.store(self.tail, Ordering::Release); // Wake the reader because the pipe is not empty. self.inner.reader.wake(); } } } /// Yield with some small probability. fn maybe_yield(cx: &mut Context<'_>) -> Poll<()> { if fastrand::usize(..100) == 0 { cx.waker().wake_by_ref(); Poll::Pending } else { Poll::Ready(()) } }