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
use std::time::Duration; #[macro_use(defer)] extern crate scopeguard; use crossbeam::channel::Receiver; use crossbeam::crossbeam_channel::{select, tick, unbounded}; use crossbeam::sync::WaitGroup; use std::fmt; use std::error::Error; use std::result; use std::thread; use rand::Rng; use std::time::Instant; type Result = result::Result<(), Box<dyn Error>>; /// Group allows to start a group of threads and wait for their completion. pub struct Group { wg: WaitGroup, } impl Group { pub fn new() -> Group { Group { wg: WaitGroup::new(), } } pub fn wait(self) { self.wg.wait(); } /// start_with_channel starts f in a new thread in the group. /// stop_ch is passed to f as an argument. /// f should stop when stop_ch is available. pub fn start_with_channel<F>(&self, stop_ch: Receiver<bool>, f: F) where F: Fn(Receiver<bool>) -> () + 'static + std::marker::Send + std::marker::Sync, { self.start(move || f(stop_ch.clone())); } /// start starts f in a new thread in the group. pub fn start<F>(&self, f: F) where F: Fn() -> () + std::marker::Send + 'static, { let wg = self.wg.clone(); thread::spawn(move || { f(); drop(wg); }); } } /// forever calls f every period for ever. /// /// forever is syntactic sugar on top of until. pub fn forever<F>(f: F, period: Duration) where F: Fn() -> (), { let (_s, r) = unbounded(); until(f, period, r) } /// until loops until stop channel is closed, running f every period. /// /// until is syntactic sugar on top of jitter_until with zero jitter factor and /// with sliding = true (which means the timer for period starts after the f /// completes). pub fn until<F>(f: F, period: Duration, stop_ch: Receiver<bool>) where F: Fn() -> (), { jitter_until(f, period, 0.0, true, stop_ch) } /// non_sliding_until loops until stop channel is closed, running f every /// period. /// /// non_sliding_until is syntactic sugar on top of jitter_until with zero jitter /// factor, with sliding = false (meaning the timer for period starts at the same /// time as the function starts). pub fn non_sliding_until<F>(f: F, period: Duration, stop_ch: Receiver<bool>) where F: Fn() -> (), { jitter_until(f, period, 0.0, false, stop_ch) } /// jitter_until loops until stop channel is closed, running f every period. /// /// If jitter_factor is positive, the period is jittered before every run of f. /// If jitter_factor is not positive, the period is unchanged and not jittered. /// /// If sliding is true, the period is computed after f runs. If it is false then /// period includes the runtime for f. /// /// Close stop_ch to stop. f may not be invoked if stop channel is already /// closed. Pass NeverStop to if you don't want it stop. pub fn jitter_until<F>( f: F, period: Duration, jitter_factor: f64, sliding: bool, stop_ch: Receiver<bool>, ) where F: Fn() -> (), { backoff_until( f, JitteredBackoffManager::new_jittered_backoff_manager(period, jitter_factor), sliding, stop_ch, ) } /// backoff_until loops until stop channel is closed, run f every duration given by BackoffManager. /// /// If sliding is true, the period is computed after f runs. If it is false then /// period includes the runtime for f. pub fn backoff_until<F>( f: F, mut backoff: Box<dyn BackoffManager>, sliding: bool, stop_ch: Receiver<bool>, ) where F: Fn() -> (), { loop { select! { recv(stop_ch) -> _ => return , default => {} } let mut t = backoff.backoff(); // FIXME handle crach // func() { // defer runtime.HandleCrash() // f() // }() f(); if sliding { t = backoff.backoff(); } // NOTE: b/c there is no priority selection in golang // it is possible for this to race, meaning we could // trigger t.C and stop_ch, and t.C select falls through. // In order to mitigate we re-check stop_ch at the beginning // of every loop to prevent extra executions of f(). select! { recv(stop_ch) -> _ => return, recv(t) -> _msg => { } } } } /// jitter returns a Duration between duration and duration + max_factor * /// duration. /// /// This allows clients to avoid converging on periodic behavior. If max_factor /// is 0.0, a suggested default value will be chosen. pub fn jitter(duration: Duration, max_factor: f64) -> Duration { let mut mf = max_factor; if mf <= 0.0 { mf = 1.0; } let mut rng = rand::thread_rng(); Duration::from_nanos((duration.as_nanos() as f64 * (1.0 + rng.gen::<u64>() as f64 * mf)) as u64) } /// WaitTimeoutError is returned when the condition exited without success. #[derive(Debug, Clone)] struct WaitTimeoutError; impl fmt::Display for WaitTimeoutError { fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { write!(f, "timed out waiting for the condition") } } impl Error for WaitTimeoutError {} /// run_condition_with_crash_protection runs a ConditionFunc with crash protection fn run_condition_with_crash_protection<F>(condition: F) -> std::result::Result<bool, Box<dyn Error>> where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { // defer runtime.HandleCrash() condition() } /// Backoff holds parameters applied to a Backoff function. pub struct Backoff { /// The initial duration. duration: Duration, /// Duration is multiplied by factor each iteration, if factor is not zero /// and the limits imposed by steps and cap have not been reached. /// Should not be negative. /// The jitter does not contribute to the updates to the duration parameter. factor: f64, /// The sleep at each iteration is the duration plus an additional /// amount chosen uniformly at random from the interval between /// zero and `jitter*duration`. jitter: f64, /// The remaining number of iterations in which the duration /// parameter may change (but progress can be stopped earlier by /// hitting the cap). If not positive, the duration is not /// changed. Used for exponential backoff in combination with /// factor and cap. steps: i32, /// A limit on revised values of the duration parameter. If a /// multiplication by the factor parameter would make the duration /// exceed the cap then the duration is set to the cap and the /// steps parameter is set to zero. cap: Duration, } impl Backoff { /// step (1) returns an amount of time to sleep determined by the /// original duration and jitter and (2) mutates the provided Backoff /// to update its steps and duration. pub fn step(&mut self) -> Duration { if self.steps < 1 { if self.jitter > 0.0 { return jitter(self.duration, self.jitter); } return self.duration; } self.steps = self.steps - 1; let mut duration = self.duration; // calculate the next step if self.factor != 0.0 { self.duration = Duration::from_nanos((self.duration.as_nanos() as f64 * self.factor) as u64); if !(self.cap.as_nanos() == 0) && self.duration > self.cap { self.duration = self.cap; self.steps = 0; } } if self.jitter > 0.0 { duration = jitter(duration, self.jitter); } duration } } /// BackoffManager manages backoff with a particular scheme based on its underlying implementation. It provides /// an interface to return a timer for backoff, and caller shall backoff until Timer.C() drains. If the second backoff() /// is called before the timer from the first backoff() call finishes, the first timer will NOT be drained and result in /// undetermined behavior. /// The BackoffManager is supposed to be called in a single-threaded environment. pub trait BackoffManager { fn backoff(&mut self) -> Receiver<Instant>; } pub struct ExponentialBackoffManager { backoff: Backoff, last_backoff_start: Instant, initial_backoff: Duration, backoff_reset_duration: Duration, } impl BackoffManager for ExponentialBackoffManager { /// Backoff implements BackoffManager.backoff, /// it returns a timer so caller can block on the timer for exponential backoff. /// The returned timer must be drained before calling backoff() the second time fn backoff(&mut self) -> Receiver<Instant> { tick(self.get_next_backoff()) } } impl ExponentialBackoffManager { /// new_exponential_backoff_manager returns a manager for managing exponential backoff. /// Each backoff is jittered and /// backoff will not exceed the given max. /// If the backoff is not called within reset_duration, the backoff is reset. /// This backoff manager is used to reduce load during upstream unhealthiness. pub fn new_exponential_backoff_manager( init_backoff: Duration, max_backoff: Duration, reset_duration: Duration, backoff_factor: f64, jitter: f64, ) -> Box<dyn BackoffManager> { Box::new(ExponentialBackoffManager { backoff: Backoff { duration: init_backoff, factor: backoff_factor, jitter: jitter, // the current impl of wait.backoff returns Backoff.duration once steps are used up, which is not // what we ideally need here, we set it to max int and assume we will never use up the steps steps: std::i32::MAX, cap: max_backoff, }, initial_backoff: init_backoff, last_backoff_start: Instant::now(), backoff_reset_duration: reset_duration, }) } fn get_next_backoff(&mut self) -> Duration { if Instant::now().duration_since(self.last_backoff_start) > self.backoff_reset_duration { self.backoff.steps = std::i32::MAX; self.backoff.duration = self.initial_backoff; } self.last_backoff_start = Instant::now(); return self.backoff.step(); } } pub struct JitteredBackoffManager { duration: Duration, jitter: f64, } impl BackoffManager for JitteredBackoffManager { /// backoff implements BackoffManager.backoff, /// it returns a timer so caller can block on the timer for jittered backoff. /// The returned timer must be drained before calling backoff() the second time fn backoff(&mut self) -> Receiver<Instant> { tick(self.get_next_backoff()) } } impl JitteredBackoffManager { /// new_jittered_backoff_manager returns a BackoffManager that backoffs with given duration plus given jitter. /// If the jitter /// is negative, backoff will not be jittered. pub fn new_jittered_backoff_manager( duration: Duration, jitter: f64, ) -> Box<dyn BackoffManager> { Box::new(JitteredBackoffManager { duration: duration, jitter: jitter, }) as Box<dyn BackoffManager> } fn get_next_backoff(&self) -> Duration { if self.jitter > 0.0 { jitter(self.duration, self.jitter) } else { self.duration } } } /// exponential_backoff repeats a condition check with exponential backoff. /// /// It repeatedly checks the condition and then sleeps, using `backoff.step()` /// to determine the length of the sleep and adjust Duration and steps. /// Stops and returns as soon as: /// 1. the condition check returns true or an error, /// 2. `backoff.steps` checks of the condition have been done, or /// 3. a sleep truncated by the cap on duration has been completed. /// In case (1) the returned error is what the condition function returned. /// In all other cases, WaitTimeoutError is returned. pub fn exponential_backoff<F>(backoff: &mut Backoff, condition: F) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { while backoff.steps > 0 { let ok = run_condition_with_crash_protection(condition)?; if ok { return Ok(()); } if backoff.steps == 1 { break; } thread::sleep(backoff.step()); } Err(Box::new(WaitTimeoutError)) } /// poll tries a condition func until it returns true, an error, or the timeout /// is reached. /// /// poll always waits the interval before the run of 'condition'. /// 'condition' will always be invoked at least once. /// /// Some intervals may be missed if the condition takes too long or the time /// window is too short. /// /// If you want to poll something forever, see poll_infinite. pub fn poll<F>(interval: Duration, timeout: Duration, condition: F) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { poll_internal(poller(interval, timeout), condition) } fn poll_internal<F>(wait: Box<dyn Fn(Receiver<bool>) -> Receiver<bool>>, condition: F) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { let (_s, r) = unbounded(); wait_for(wait, condition, r) } /// poll_immediate tries a condition func until it returns true, an error, or the timeout /// is reached. /// /// poll_immediate always checks 'condition' before waiting for the interval. 'condition' /// will always be invoked at least once. /// /// Some intervals may be missed if the condition takes too long or the time /// window is too short. /// /// If you want to immediately poll something forever, see poll_immediate_infinite. pub fn poll_immediate<F>(interval: Duration, timeout: Duration, condition: F) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { poll_immediate_internal(poller(interval, timeout), condition) } fn poll_immediate_internal<F>( wait: Box<dyn Fn(Receiver<bool>) -> Receiver<bool>>, condition: F, ) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { let done = run_condition_with_crash_protection(condition)?; if done { return Ok(()); } poll_internal(wait, condition) } /// poll_infinite tries a condition func until it returns true or an error /// /// poll_infinite always waits the interval before the run of 'condition'. /// /// Some intervals may be missed if the condition takes too long or the time /// window is too short. pub fn poll_infinite<F>(interval: Duration, condition: F) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { let (_s, r) = unbounded(); return poll_until(interval, condition, r); } /// poll_immediate_infinite tries a condition func until it returns true or an error /// /// poll_immediate_infinite runs the 'condition' before waiting for the interval. /// /// Some intervals may be missed if the condition takes too long or the time /// window is too short. pub fn poll_immediate_infinite<F>(interval: Duration, condition: F) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { let done = run_condition_with_crash_protection(condition)?; if done { return Ok(()); } poll_infinite(interval, condition) } /// poll_until tries a condition func until it returns true, an error or stop_ch is /// closed. /// /// poll_until always waits interval before the first run of 'condition'. /// 'condition' will always be invoked at least once. pub fn poll_until<F>(interval: Duration, condition: F, stop_ch: Receiver<bool>) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { return wait_for(poller(interval, Duration::new(0, 0)), condition, stop_ch); } /// poll_immediate_until tries a condition func until it returns true, an error or stop_ch is closed. /// /// poll_immediate_until runs the 'condition' before waiting for the interval. /// 'condition' will always be invoked at least once. pub fn poll_immediate_until<F>(interval: Duration, condition: F, stop_ch: Receiver<bool>) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { let done = condition()?; if done { return Ok(()); } select! { recv(stop_ch) -> _ => return Err(Box::new(WaitTimeoutError)) , default => return poll_until(interval, condition, stop_ch) } } /// wait_for continually checks 'fn' as driven by 'wait'. /// /// wait_for gets a channel from 'wait()', and then invokes 'fn' once for every value /// placed on the channel and once more when the channel is closed. If the channel is closed /// and 'fn' returns false without error, wait_for returns WaitTimeoutError. /// /// If 'fn' returns an error the loop ends and that error is returned. If /// 'fn' returns true the loop ends and nil is returned. /// /// WaitTimeoutError will be returned if the 'done' channel is closed without fn ever /// returning true. /// /// When the done channel is closed, because the golang `select` statement is /// "uniform pseudo-random", the `fn` might still run one or multiple time, /// though eventually `wait_for` will return. pub fn wait_for<F>( wait: Box<dyn Fn(Receiver<bool>) -> Receiver<bool>>, func: F, done: Receiver<bool>, ) -> Result where F: Fn() -> std::result::Result<bool, Box<dyn Error>> + Copy, { let (s, r) = unbounded(); let c = wait(r); // notify wait thread to quit. defer! { drop(s);}; loop { select! { recv(c) -> msg => { let ok = run_condition_with_crash_protection(func)?; if ok { return Ok(()); } if msg.is_err() { return Err(Box::new(WaitTimeoutError)); } }, recv(done) -> _ => return Err(Box::new(WaitTimeoutError)), } } } /// poller returns a Box<dyn Fn(Receiver<bool>) -> Receiver<bool>> /// that will send to the channel every interval until /// timeout has elapsed and then closes the channel. /// /// Over very short intervals you may receive no ticks before the channel is /// closed. A timeout of 0.0 is interpreted as an infinity, and in such a case /// it would be the caller's responsibility to close the done channel. /// Failure to do so would result in a leaked goroutine. /// /// Output ticks are not buffered. If the channel is not ready to receive an /// item, the tick is skipped. fn poller(interval: Duration, timeout: Duration) -> Box<dyn Fn(Receiver<bool>) -> Receiver<bool>> { let func = move |done: Receiver<bool>| -> Receiver<bool> { let (s, r) = unbounded(); let rr = r.clone(); thread::spawn(move || { let ticker = tick(interval); // FIXME: workaround for compile error use of possibly-uninitialized `after` let mut after = tick(Duration::from_secs(1000000000)); if !(timeout.as_nanos() == 0) { after = tick(timeout); } loop { select! { recv(ticker) -> _ => { // If the consumer isn't ready for this signal drop it and // check the other channels. s.send(true).unwrap(); }, recv(after) -> _ => { return }, recv(done) -> _ => { return }, } } }); rr }; Box::new(func) } #[cfg(test)] mod tests { use super::*; use std::sync::{Arc, Mutex}; #[test] fn test_poll() { let (s, r) = unbounded(); let cond_fn = || { select! { recv(r) -> msg => { if msg.unwrap() == 3 { return Ok(true); }else { return Ok(false); } }, default => { return Ok(false) } , } }; thread::spawn(move || { for i in 1..4 { thread::sleep(Duration::from_millis(20)); s.send(i).unwrap(); } drop(s); println!("sender dropped"); }); let ret = poll( Duration::from_millis(100), Duration::from_millis(300), cond_fn, ); assert_eq!(true, ret.is_ok()); let cond_fn = || Ok(false); let ret = poll( Duration::from_millis(100), Duration::from_millis(300), cond_fn, ); assert_eq!(true, ret.is_err()); } #[test] fn test_until() { let (s, r) = unbounded(); let counter = Arc::new(Mutex::new(0)); let counter1 = counter.clone(); let worker_fn = move || { let mut counter = counter1.lock().unwrap(); *counter += 1; println!("do work {}", *counter); }; let counter2 = counter.clone(); std::thread::spawn(move || loop { { let counter = counter2.lock().unwrap(); if *counter > 4 { drop(s); println!("sender dropped"); return; } } thread::sleep(Duration::from_millis(10)); }); until(worker_fn, Duration::from_millis(10), r); let counter = counter.lock().unwrap(); println!("final counter {}", *counter); assert_eq!(true, *counter > 4); } #[test] fn test_xxx() { let zero_seconds = Duration::new(0, 0); assert_eq!(0, zero_seconds.as_nanos()); let (s, r) = unbounded(); thread::spawn(move || { s.send(1).unwrap(); s.send(2).unwrap(); }); let msg1 = r.recv().unwrap(); let msg2 = r.recv().unwrap(); assert_eq!(msg1 + msg2, 3); } #[test] fn test_groups() { let counter = Arc::new(Mutex::new(0)); let counter1 = counter.clone(); let worker_fn1 = move || { let mut counter = counter1.lock().unwrap(); *counter += 1; thread::sleep(Duration::from_millis(50)); println!("worker 1 finished"); }; let (s, r) = unbounded(); let counter2 = counter.clone(); let worker_fn2 = move |x| { let mut counter = counter2.lock().unwrap(); *counter += 1; select! { recv(x) -> _ => { println!("worker 2 finished"); } } }; thread::spawn(move || { thread::sleep(Duration::from_millis(300)); drop(s); println!("notify worker 2 to finish"); }); let group = Group::new(); group.start(worker_fn1); println!("worker 1 started"); group.start_with_channel(r, worker_fn2); println!("worker 2 started"); println!("wait two workeres to finish"); group.wait(); println!("two workeres are finished"); } #[test] fn test_exponential_backoff_manager() { // backoff at least 1ms, 2ms, 4ms, 8ms, 10ms, 10ms, 10ms let duration_factors = vec![1, 2, 4, 8, 10, 10, 10]; let mut backoff_mgr = ExponentialBackoffManager::new_exponential_backoff_manager( Duration::from_millis(1), Duration::from_millis(10), Duration::from_secs(3600), 2.0, 0.0, ); for i in duration_factors { let start = Instant::now(); let r = backoff_mgr.backoff(); select! { recv(r) -> _ => {}, } let passed = Instant::now().duration_since(start).as_millis(); assert_eq!( true, passed >= i, "backoff should be at least {} ms, but got {}", i, passed ); } } #[test] fn test_jitter_backoff_manager_with_real_clock() { let mut backoff_mgr = JitteredBackoffManager::new_jittered_backoff_manager(Duration::from_millis(1), 0.0); for _ in 0..5 { let start = Instant::now(); let r = backoff_mgr.backoff(); select! { recv(r) -> _ => {}, } let passed = Instant::now().duration_since(start).as_millis(); assert_eq!( true, passed >= 1, "backoff should be at least 1ms, but got {}", passed ); } } }