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
//! # Yaque: Yet Another QUEue //! //! Yaque is yet another disk-backed persistent queue for Rust. It implements //! an SPSC channel using your OS' filesystem. Its main advantages over a simple //! `VecDeque<T>` are that //! * You are not constrained by your RAM size, just by your disk size. This //! means you can store gigabytes of data without getting OOM killed. //! * Your data is safe even if you program panics. All the queue state is //! written to the disk when the queue is dropped. //! * Your data can *persist*, that is, can exist through multiple executions //! of your program. Think of it as a very rudimentary kind of database. //! * You can pass data between two processes. //! //! Yaque is _asynchronous_ and built directly on top of `mio` and `notify`. //! It is therefore completely agnostic to the runtime you are using for you //! application. It will work smoothly with `tokio`, with `async-std` or any //! other executor of your choice. //! //! ## Sample usage //! //! To create a new queue, just use the [`channel`] function, passing a //! directory path on which to mount the queue. If the directory does not exist //! on creation, it (and possibly all its parent directories) will be created. //! ``` //! use yaque::channel; //! //! futures::executor::block_on(async { //! let (mut sender, mut receiver) = channel("data/my-queue").unwrap(); //! }) //! ``` //! You can also use [`Sender::open`] and [`Receiver::open`] to open only one //! half of the channel, if you need to. //! //! The usage is similar to the MPSC channel in the standard library, except //! that the receiving method, [`Receiver::recv`] is asynchronous. Writing to //! the queue with the sender is basically lock-free and atomic. //! ``` //! use yaque::{channel, try_clear}; //! //! futures::executor::block_on(async { //! // Open using the `channel` function or directly with the constructors. //! let (mut sender, mut receiver) = channel("data/my-queue").unwrap(); //! //! // Send stuff with the sender... //! sender.send(b"some data").unwrap(); //! //! // ... and receive it in the other side. //! let data = receiver.recv().await.unwrap(); //! //! assert_eq!(&*data, b"some data"); //! //! // Call this to make the changes to the queue permanent. //! // Not calling it will revert the state of the queue. //! data.commit(); //! }); //! //! // After everything is said and done, you may delete the queue. //! // Use `clear` for awaiting for the queue to be released. //! try_clear("data/my-queue").unwrap(); //! ``` //! The returned value `data` is a kind of guard that implements `Deref` and //! `DerefMut` on the underlying type. //! //! ## [`RecvGuard`] and transactional behavior //! //! One important thing to notice is that reads from the queue are //! _transactional_. The `Receiver::recv` returns a [`RecvGuard`] that acts as //! a _dead man switch_. If dropped, it will revert the dequeue operation, //! unless [`RecvGuard::commit`] is explicitly called. This ensures that //! the operation reverts on panics and early returns from errors (such as when //! using the `?` notation). However, it is necessary to perform one more //! filesystem operation while rolling back. During drop, this is done on a //! "best effort" basis: if an error occurs, it is logged and ignored. This is done //! because errors cannot propagate outside a drop and panics in drops risk the //! program being aborted. If you _have_ any cleanup behavior for an error from //! rolling back, you may call [`RecvGuard::rollback`] which _will_ return the //! underlying error. //! //! ## Batches //! //! You can use the `yaque` queue to send and receive batches of data , //! too. The guarantees are the same as with single reads and writes, except //! that you may save on OS overhead when you send items, since only one disk //! operation is made. See [`Sender::send_batch`], [`Receiver::recv_batch`] and //! [`Receiver::recv_while`] for more information on receiver batches. //! //! ## `Ctrl+C` and other unexpected events //! //! During some anomalous behavior, the queue might enter an inconsistent state. //! This inconsistency is mainly related to the position of the sender and of //! the receiver in the queue. Writing to the queue is an atomic operation. //! Therefore, unless there is something really wrong with your OS, you should be //! fine. //! //! The queue is (almost) guaranteed to save all the most up-to-date metadata //! for both receiving and sending parts during a panic. The only exception is //! if the saving operation fails. However, this is not the case if the process //! receives a signal from the OS. Signals from the OS are not handled //! automatically by this library. It is understood that the application //! programmer knows best how to handle them. If you chose to close queue on //! `Ctrl+C` or other signals, you are in luck! Saving both sides of the queue //! is [async-signal-safe](https://man7.org/linux/man-pages/man7/signal-safety.7.html) //! so you may set up a bare signal hook directly using, for example, //! [`signal_hook`](https://docs.rs/signal-hook/), if you are the sort of person //! that enjoys `unsafe` code. If not, there are a ton of completely safe //! alternatives out there. Choose the one that suits you the best. //! //! Unfortunately, there are times when you get `Aborted` or `Killed`. When this //! happens, maybe not everything is lost yet. First of all, you will end up //! with a queue that is locked by no process. If you know that the process //! owning the locks has indeed past away, you may safely delete the lock files //! identified by the `.lock` extension. You will also end up with queue //! metadata pointing to an earlier state in time. Is is easy to guess what the //! sending metadata should be. Is is the top of the last segment file. However, //! things get trickier in the receiver side. You know that it is the greatest //! of two positions: //! //! 1. the bottom of the smallest segment still present in the directory. //! //! 2. the position indicated in the metadata file. //! //! Depending on your use case, this might be information enough so that not all //! hope is lost. However, this is all you will get. //! //! If you really want to err on the safer side, you may use [`Sender::save`] //! and [`Receiver::save`] to periodically back the queue state up. Just choose //! you favorite timer implementation and set a simple periodical task up every //! hundreds of milliseconds. However, be warned that this is only a _mitigation_ //! of consistency problems, not a solution. //! //! ## Known issues and next steps //! //! * This is a brand new project. Although I have tested it and it will //! certainly not implode your computer, don't trust your life on it yet. //! * Wastes too much kernel time when the queue is small enough and the sender //! sends many frequent small messages non-atomically. You can mitigate that by //! writing in batches to the queue. //! * I intend to make this an MPSC queue in the future. //! * There are probably unknown bugs hidden in some corner case. If you find //! one, please fill an issue in GitHub. Pull requests and contributions are //! also greatly appreciated. mod state; mod sync; mod watcher; #[cfg(feature = "recovery")] pub mod recovery; pub use sync::FileGuard; use std::fs::*; use std::io::{self, Write}; use std::ops::{Deref, DerefMut}; use std::path::{Path, PathBuf}; use state::FilePersistence; use state::QueueState; use sync::TailFollower; /// The name of segment file in the queue folder. fn segment_filename<P: AsRef<Path>>(base: P, segment: u64) -> PathBuf { base.as_ref().join(format!("{}.q", segment)) } /// The name of the receiver lock in the queue folder. fn recv_lock_filename<P: AsRef<Path>>(base: P) -> PathBuf { base.as_ref().join("recv.lock") } /// Tries to acquire the receiver lock for a queue. fn try_acquire_recv_lock<P: AsRef<Path>>(base: P) -> io::Result<FileGuard> { FileGuard::try_lock(recv_lock_filename(base.as_ref()))?.ok_or_else(|| { io::Error::new( io::ErrorKind::Other, format!( "queue `{}` receiver side already in use", base.as_ref().to_string_lossy() ), ) }) } /// Acquire the receiver lock for a queue, awaiting if locked. async fn acquire_recv_lock<P: AsRef<Path>>(base: P) -> io::Result<FileGuard> { FileGuard::lock(recv_lock_filename(base.as_ref())).await } /// The name of the sender lock in the queue folder. fn send_lock_filename<P: AsRef<Path>>(base: P) -> PathBuf { base.as_ref().join("send.lock") } /// Tries to acquire the sender lock for a queue. fn try_acquire_send_lock<P: AsRef<Path>>(base: P) -> io::Result<FileGuard> { FileGuard::try_lock(send_lock_filename(base.as_ref()))?.ok_or_else(|| { io::Error::new( io::ErrorKind::Other, format!( "queue `{}` sender side already in use", base.as_ref().to_string_lossy() ), ) }) } /// Acquire the sender lock for a queue, awaiting if locked. async fn acquire_send_lock<P: AsRef<Path>>(base: P) -> io::Result<FileGuard> { FileGuard::lock(send_lock_filename(base.as_ref())).await } /// The value of a header EOF. const HEADER_EOF: u32 = std::u32::MAX; /// The sender part of the queue. This part is lock-free and therefore can be /// used outside an asynchronous context. pub struct Sender { _file_guard: FileGuard, file: io::BufWriter<File>, state: QueueState, base: PathBuf, persistence: FilePersistence, } impl Sender { /// Opens a queue on a folder indicated by the `base` path for sending. The /// folder will be created if it does not already exist. /// /// # Errors /// /// This function will return an IO error if the queue is already in use for /// sending, which is indicated by a lock file. Also, any other IO error /// encountered while opening will be sent. pub fn open<P: AsRef<Path>>(base: P) -> io::Result<Sender> { // Guarantee that the queue exists: create_dir_all(base.as_ref())?; log::trace!("created queue directory"); // Acquire lock and state: let file_guard = try_acquire_send_lock(base.as_ref())?; let mut persistence = FilePersistence::new(); let state = persistence.open_send(base.as_ref())?; log::trace!("sender lock acquired. Sender state now is {:?}", state); // See the docs on OpenOptions::append for why the BufWriter here. let file = io::BufWriter::new( OpenOptions::new() .create(true) .append(true) .open(segment_filename(base.as_ref(), state.segment))?, ); log::trace!("last segment opened for appending"); Ok(Sender { _file_guard: file_guard, file, state, base: PathBuf::from(base.as_ref()), persistence, }) } /// Saves the sender queue state. You do not need to use method in most /// circumstances, since it is automatically done on drop (yes, it will be /// called eve if your thread panics). However, you can use this function to /// /// 1. Make periodical backups. Use an external timer implementation for this. /// /// 2. Handle possible IO errors in sending. The `drop` implementation will /// ignore (but log) any io errors, which may lead to data loss in an /// unreliable filesystem. This happens because no errors are allowed to /// propagate on drop and panicking will abort the program if drop is called /// during a panic. pub fn save(&mut self) -> io::Result<()> { self.persistence.save(&self.state) } /// Just writes to the internal buffer, but doesn't flush it. fn write(&mut self, data: &[u8]) -> io::Result<u64> { // Get length of the data and write the header: let len = data.as_ref().len(); assert!(len < std::u32::MAX as usize); let header = (len as u32).to_be_bytes(); // Write stuff to the file: self.file.write_all(&header)?; self.file.write_all(data.as_ref())?; Ok(4 + len as u64) } /// Caps off a segment by writing an EOF header and then moves segment. fn cap_off_and_move(&mut self) -> io::Result<()> { // Write EOF header: self.file.write(&HEADER_EOF.to_be_bytes())?; self.file.flush()?; // Preserves the already allocated buffer: *self.file.get_mut() = OpenOptions::new() .create(true) .append(true) .open(segment_filename(&self.base, self.state.advance_segment()))?; Ok(()) } /// Sends some data into the queue. One send is always atomic. /// /// # Errors /// /// This function returns any underlying errors encountered while writing or /// flushing the queue. pub fn send<D: AsRef<[u8]>>(&mut self, data: D) -> io::Result<()> { // Write to the queue and flush: let written = self.write(data.as_ref())?; self.file.flush()?; // guarantees atomic operation. See `new`. self.state.advance_position(written); // See if you are past the end of the file if self.state.is_past_end() { // If so, create a new file: self.cap_off_and_move()?; } Ok(()) } /// Sends all the contents of an iterable into the queue. All is buffered /// to be sent atomically, in one flush operation. pub fn send_batch<I>(&mut self, it: I) -> io::Result<()> where I: IntoIterator, I::Item: AsRef<[u8]>, { let mut written = 0; // Drain iterator into the buffer. for item in it { written += self.write(item.as_ref())?; } self.file.flush()?; // guarantees atomic operation. See `new`. self.state.advance_position(written); // See if you are past the end of the file if self.state.is_past_end() { // If so, create a new file: self.cap_off_and_move()?; } Ok(()) } } impl Drop for Sender { fn drop(&mut self) { if let Err(err) = self.persistence.save(&self.state) { log::error!("could not release sender lock: {}", err); } } } /// The receiver part of the queue. This part is asynchronous and therefore /// needs an executor that will the poll the futures to completion. pub struct Receiver { _file_guard: FileGuard, tail_follower: TailFollower, state: QueueState, base: PathBuf, persistence: FilePersistence, } impl Receiver { /// Opens a queue for reading. The access will be exclusive, based on the /// existence of the temporary file `recv.lock` inside the queue folder. /// /// # Errors /// /// This function will return an IO error if the queue is already in use for /// receiving, which is indicated by a lock file. Also, any other IO error /// encountered while opening will be sent. /// /// # Panics /// /// This function will panic if it is not able to set up the notification /// handler to watch for file changes. pub fn open<P: AsRef<Path>>(base: P) -> io::Result<Receiver> { // Guarantee that the queue exists: create_dir_all(base.as_ref())?; log::trace!("created queue directory"); // Acquire guard and state: let file_guard = try_acquire_recv_lock(base.as_ref())?; let mut persistence = FilePersistence::new(); let state = persistence.open_recv(base.as_ref())?; log::trace!("receiver lock acquired. Receiver state now is {:?}", state); // Put the needle on the groove (oh! the 70's): let mut tail_follower = TailFollower::open(segment_filename(base.as_ref(), state.segment))?; tail_follower.seek(io::SeekFrom::Start(state.position))?; log::trace!("last segment opened fo reading"); Ok(Receiver { _file_guard: file_guard, tail_follower, state, base: PathBuf::from(base.as_ref()), persistence, }) } /// Maybe advance the segment of this receiver. async fn advance(&mut self) -> io::Result<()> { log::trace!( "advancing segment from {} to {}", self.state.segment, self.state.segment + 1 ); // Advance segment and checkpoint! self.state.advance_segment(); if let Err(err) = self.persistence.save(&self.state) { log::error!("failed to save receiver: {}", err); self.state.retreat_segment(); return Err(err); } // Start listening to new segments: self.tail_follower = TailFollower::open(segment_filename(&self.base, self.state.segment))?; log::trace!("acquired new tail follower"); // Remove old file: remove_file(segment_filename(&self.base, self.state.segment - 1))?; log::trace!("removed old segment file"); Ok(()) } /// Reads one element from the queue, inevitably advancing the file reader. async fn read_one(&mut self) -> io::Result<Vec<u8>> { // Read the header to get the length: let mut header = [0; 4]; self.tail_follower.read_exact(&mut header).await?; let mut len = u32::from_be_bytes(header); // If the header is EOF, advance segment: if len == HEADER_EOF { log::trace!("got EOF header. Advancing..."); self.advance().await?; // Re-read the header: log::trace!("re-reading new header from new file"); self.tail_follower.read_exact(&mut header).await?; len = u32::from_be_bytes(header); } // With the length, read the data: let mut data = (0..len).map(|_| 0).collect::<Vec<_>>(); self.tail_follower .read_exact(&mut data) .await .expect("poisoned queue"); Ok(data) } /// Saves the receiver queue state. You do not need to use method in most /// circumstances, since it is automatically done on drop (yes, it will be /// called eve if your thread panics). However, you can use this function to /// /// 1. Make periodical backups. Use an external timer implementation for this. /// /// 2. Handle possible IO errors in sending. The `drop` implementation will /// ignore (but log) any io errors, which may lead to data loss in an /// unreliable filesystem. This happens because no errors are allowed to /// propagate on drop and panicking will abort the program if drop is called /// during a panic. pub fn save(&mut self) -> io::Result<()> { self.persistence.save(&self.state) } /// Tries to retrieve an element from the queue. The returned value is a /// guard that will only commit state changes to the queue when dropped. /// /// # Panics /// /// This function will panic if it has to start reading a new segment and /// it is not able to set up the notification handler to watch for file /// changes. pub async fn recv(&mut self) -> io::Result<RecvGuard<'_, Vec<u8>>> { let data = self.read_one().await?; Ok(RecvGuard { receiver: self, len: 4 + data.len(), item: Some(data), override_drop: false, }) } /// Tries to remove a number of elements from the queue. The returned value /// is a guard that will only commit state changes to the queue when dropped. /// /// # Panics /// /// This function will panic if it has to start reading a new segment and /// it is not able to set up the notification handler to watch for file /// changes. pub async fn recv_batch(&mut self, n: usize) -> io::Result<RecvGuard<'_, Vec<Vec<u8>>>> { let mut data = vec![]; for _ in 0..n { data.push(self.read_one().await?); } Ok(RecvGuard { receiver: self, len: data.iter().map(|item| 4 + item.len()).sum(), item: Some(data), override_drop: false, }) } /// Takes a number of elements from the queue until a certain asynchronous /// condition is met. You will receive a `None` as the first element so that /// you may return early and leave the queue intact (this would not be /// possible otherwise). The returned value is a guard that will only commit /// state changes to the queue when dropped. /// /// # Panics /// /// This function will panic if it has to start reading a new segment and /// it is not able to set up the notification handler to watch for file /// changes. pub async fn recv_until<P, Fut>( &mut self, mut predicate: P, ) -> io::Result<RecvGuard<'_, Vec<Vec<u8>>>> where P: FnMut(Option<&[u8]>) -> Fut, Fut: std::future::Future<Output = bool>, { let mut data = vec![]; // Prepare: predicate(None).await; // Poor man's do-while (aka. until) loop { let item = self.read_one().await?; if !predicate(Some(&item)).await { data.push(item); } else { break; } } Ok(RecvGuard { receiver: self, len: data.iter().map(|item| 4 + item.len()).sum(), item: Some(data), override_drop: false, }) } } impl Drop for Receiver { fn drop(&mut self) { if let Err(err) = self.persistence.save(&self.state) { log::error!("could not release receiver lock: {}", err); } } } /// A guard that will only log changes on the queue state when dropped. /// /// If it is dropped without a call to `RecvGuard::commit`, changes will be /// rolled back in a "best effort" policy: if any IO error is encountered /// during rollback, the state will be committed. If you *can* do something /// with the IO error, you may use `RecvGuard::rollback` explicitly to catch /// the error. /// /// This struct implements `Deref` and `DerefMut`. If you really, really want /// ownership, there is `RecvGuard::into_inner`, but be careful, because you /// lose your chance to rollback if anything unexpected occurs. pub struct RecvGuard<'a, T> { receiver: &'a mut Receiver, len: usize, item: Option<T>, override_drop: bool, } impl<'a, T> Drop for RecvGuard<'a, T> { fn drop(&mut self) { if self.override_drop { } else { if let Err(err) = self .receiver .tail_follower .seek(io::SeekFrom::Current(-(self.len as i64))) { log::error!("unable to rollback on drop: {}", err); } } } } impl<'a, T> Deref for RecvGuard<'a, T> { type Target = T; fn deref(&self) -> &T { self.item.as_ref().expect("unreachable") } } impl<'a, T> DerefMut for RecvGuard<'a, T> { fn deref_mut(&mut self) -> &mut T { self.item.as_mut().expect("unreachable") } } impl<'a, T> RecvGuard<'a, T> { /// Commits the transaction and returns the underlying value. If you /// accidentally lose this value from now on, it's your own fault! pub fn into_inner(mut self) -> T { let item = self.item.take().expect("unreachable"); self.commit(); item } /// Commits the changes to the queue, consuming this `RecvGuard`. pub fn commit(mut self) { self.override_drop = true; self.receiver.state.position += self.len as u64; drop(self); } /// Rolls the reader back to the previous point, negating the changes made /// on the queue. This is also done on drop. However, on drop, the possible /// IO error is ignored (but logged as an error) because we cannot have /// errors inside drops. Use this if you want to control errors at rollback. /// /// # Errors /// /// If there is some error while moving the reader back, this error will be /// return. pub fn rollback(mut self) -> io::Result<()> { self.override_drop = true; // Do it manually. self.receiver .tail_follower .seek(io::SeekFrom::Current(-(self.len as i64))) } } /// Convenience function for opening the queue for both sending and receiving. pub fn channel<P: AsRef<Path>>(base: P) -> io::Result<(Sender, Receiver)> { Ok((Sender::open(base.as_ref())?, Receiver::open(base.as_ref())?)) } /// Tries to deletes a queue at the given path. This function will fail if the /// queue is in use either for sending or receiving. pub fn try_clear<P: AsRef<Path>>(base: P) -> io::Result<()> { let mut send_lock = try_acquire_send_lock(base.as_ref())?; let mut recv_lock = try_acquire_recv_lock(base.as_ref())?; // Sets the the locks to ignore when their files magically disappear. send_lock.ignore(); recv_lock.ignore(); remove_dir_all(base.as_ref())?; Ok(()) } /// Deletes a queue at the given path. This function will await the queue to /// become available for both sending and receiving. pub async fn clear<P: AsRef<Path>>(base: P) -> io::Result<()> { let mut send_lock = acquire_send_lock(base.as_ref()).await?; let mut recv_lock = acquire_recv_lock(base.as_ref()).await?; // Sets the the locks to ignore when their files magically disappear. send_lock.ignore(); recv_lock.ignore(); remove_dir_all(base.as_ref())?; Ok(()) } #[cfg(test)] mod tests { use super::*; use rand::{Rng, SeedableRng}; use rand_xorshift::XorShiftRng; use std::sync::Arc; fn data_lots_of_data() -> impl Iterator<Item = Vec<u8>> { let mut rng = XorShiftRng::from_rng(rand::thread_rng()).expect("can init"); (0..).map(move |_| { (0..rng.gen::<usize>() % 128 + 1) .map(|_| rng.gen()) .collect::<Vec<_>>() }) } #[test] fn create_and_clear() { let _ = simple_logger::init_with_level(log::Level::Trace); let _ = Sender::open("data/create-and-clear").unwrap(); try_clear("data/create-and-clear").unwrap(); } #[test] #[should_panic] fn create_and_clear_fails() { let _ = simple_logger::init_with_level(log::Level::Trace); let sender = Sender::open("data/create-and-clear-fails").unwrap(); try_clear("data/create-and-clear-fails").unwrap(); drop(sender); } #[test] fn create_and_clear_async() { let _ = simple_logger::init_with_level(log::Level::Trace); let _ = Sender::open("data/create-and-clear-async").unwrap(); futures::executor::block_on(async { clear("data/create-and-clear-async").await.unwrap() }); } #[test] fn test_enqueue() { let _ = simple_logger::init_with_level(log::Level::Trace); let mut sender = Sender::open("data/enqueue").unwrap(); for data in data_lots_of_data().take(100_000) { sender.send(&data).unwrap(); } } /// Test enqueuing everything and then dequeueing everything, with no persistence. #[test] fn test_enqueue_then_dequeue() { let _ = simple_logger::init_with_level(log::Level::Trace); // Enqueue: let dataset = data_lots_of_data().take(100_000).collect::<Vec<_>>(); let mut sender = Sender::open("data/enqueue-then-dequeue").unwrap(); for data in &dataset { sender.send(data).unwrap(); } log::trace!("enqueued"); // Dequeue: futures::executor::block_on(async { let mut receiver = Receiver::open("data/enqueue-then-dequeue").unwrap(); let dataset_iter = dataset.iter(); let mut i = 0u64; for should_be in dataset_iter { let data = receiver.recv().await.unwrap(); assert_eq!(&*data, should_be, "at sample {}", i); i += 1; data.commit(); } }); } /// Test enqueuing and dequeueing, round robin, with no persistence. #[test] fn test_enqueue_and_dequeue() { let _ = simple_logger::init_with_level(log::Level::Trace); // Enqueue: let dataset = data_lots_of_data().take(100_000).collect::<Vec<_>>(); let mut sender = Sender::open("data/enqueue-and-dequeue").unwrap(); futures::executor::block_on(async { let mut receiver = Receiver::open("data/enqueue-and-dequeue").unwrap(); let mut i = 0; for data in &dataset { sender.send(data).unwrap(); let received = receiver.recv().await.unwrap(); assert_eq!(&*received, data, "at sample {}", i); i += 1; received.commit(); } }); } /// Test enqueuing and dequeueing in parallel. #[test] fn test_enqueue_dequeue_parallel() { let _ = simple_logger::init_with_level(log::Level::Trace); // Generate data: let dataset = data_lots_of_data().take(1_000_000).collect::<Vec<_>>(); let arc_sender = Arc::new(dataset); let arc_receiver = arc_sender.clone(); // Enqueue: let enqueue = std::thread::spawn(move || { let mut sender = Sender::open("data/enqueue-dequeue-parallel").unwrap(); for data in &*arc_sender { sender.send(data).unwrap(); } }); // Dequeue: let dequeue = std::thread::spawn(move || { futures::executor::block_on(async { let mut receiver = Receiver::open("data/enqueue-dequeue-parallel").unwrap(); let dataset_iter = arc_receiver.iter(); let mut i = 0u64; for should_be in dataset_iter { let data = receiver.recv().await.unwrap(); assert_eq!(&*data, should_be, "at sample {}", i); i += 1; data.commit(); } }); }); enqueue.join().expect("enqueue thread panicked"); dequeue.join().expect("dequeue thread panicked"); } /// Test enqueuing and dequeueing in parallel, using batches. #[test] fn test_enqueue_dequeue_parallel_with_batches() { let _ = simple_logger::init_with_level(log::Level::Trace); // Generate data: let mut dataset = vec![]; let mut batch = vec![]; for data in data_lots_of_data().take(10_000_000) { batch.push(data); if batch.len() >= 256 { dataset.push(batch); batch = vec![]; } } let arc_sender = Arc::new(dataset); let arc_receiver = arc_sender.clone(); // Enqueue: let enqueue = std::thread::spawn(move || { let mut sender = Sender::open("data/enqueue-dequeue-parallel-with-batches").unwrap(); for batch in &*arc_sender { sender.send_batch(batch).unwrap(); } }); // Dequeue: let dequeue = std::thread::spawn(move || { futures::executor::block_on(async { let mut receiver = Receiver::open("data/enqueue-dequeue-parallel-with-batches").unwrap(); let dataset_iter = arc_receiver.iter(); let mut i = 0u64; for should_be in dataset_iter { let batch = receiver.recv_batch(256).await.unwrap(); assert_eq!(&*batch, should_be, "at sample {}", i); i += 1; batch.commit(); } }); }); enqueue.join().expect("enqueue thread panicked"); dequeue.join().expect("dequeue thread panicked"); } }