adaptive_barrier/lib.rs
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
//! A synchronization barrier that adapts to the number of subscribing threads.
//!
//! This has the same goal as the [`std::sync::Barrier`], but it handles runtime additions
//! or removals of thread subscriptions ‒ the number of threads waiting for the barrier can change
//! (even while some threads are already waiting).
//!
//! It can be convenient if your algorithm changes the number of working threads during lifetime.
//! You don't need a different barrier for different phases of the algorithm.
//!
//! But most importantly, the [`Barrier`] is robust in face of panics.
//!
//! # Problems with panics and the [`std::sync::Barrier`]
//!
//! If we have a barrier that was set up for `n` threads, some of the threads park on it and wait
//! for the rest to finish, but one of the other threads has a bug and panics, the already parked
//! threads will never get a chance to continue and the whole algorithm deadlocks. This is usually
//! worse than propagating the panic and cleaning up the whole algorithm or even shutting down the
//! whole application, because then something can recover by restarting it. If the application
//! deadlocks in the computation phase, but otherwise looks healthy, it will never recover.
//!
//! This makes applications less robust and makes tests which use barriers very annoying and
//! fragile to write.
//!
//! Our [`Barrier`] watches the number of subscribed threads (by counting the number of its own
//! clones, unlike the standard barrier, this one can and need to be cloned for each thread). If a
//! thread disappears (or is added), the expectations are adjusted.
//!
//! It also has a mode in which it'll get poisoned and propagate the panic to the rest of the
//! group.
#![doc(test(attr(deny(warnings))))]
#![forbid(unsafe_code)]
#![warn(missing_docs)]
use std::fmt::{Debug, Formatter, Result as FmtResult};
use std::mem;
use std::panic::UnwindSafe;
use std::sync::{Arc, Condvar, Mutex, MutexGuard};
use std::thread;
/// What to do if a [`Barrier`] is destroyed during a panic.
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
pub enum PanicMode {
/// Nothing special.
///
/// Just decrement the number of expected threads, just like during any normal destruction.
Decrement,
/// Poison the barrier.
///
/// All calls to [`wait`][Barrier::wait], including the ones that are already in progress, will
/// panic too. Once poisoned, there's no way to "unpoison" the barrier.
///
/// This is useful in case a failure in one thread makes the whole group unusable (very often
/// in tests).
Poison,
}
/// A result after waiting.
///
/// This can be used to designate a single thread as the leader after calling
/// [`wait`][Barrier::wait].
#[derive(Debug)]
pub struct WaitResult {
is_leader: bool,
}
impl WaitResult {
/// Returns true for exactly one thread from a waiting group.
///
/// An algorithm can use that to pick a thread between equals that'll do some singleton thing
/// (consolidate the results, for example).
pub fn is_leader(&self) -> bool {
self.is_leader
}
}
struct Inner {
active: usize,
waiting: usize,
gen: usize,
leader: bool,
poisoned: bool,
}
impl Inner {
fn check_release(&mut self) -> bool {
if self.waiting >= self.active || self.poisoned {
self.leader = true;
self.gen = self.gen.wrapping_add(1);
self.waiting = 0;
true
} else {
false
}
}
}
struct Shared {
inner: Mutex<Inner>,
condvar: Condvar,
panic_mode: PanicMode,
}
/// A Barrier to synchronize multiple threads.
///
/// Multiple threads can meet on a single barrier to synchronize a "meeting point" in a computation
/// (eg. when they need to pass results to others), much like the [`Barrier`][std::sync::Barrier]
/// from the standard library.
///
/// Unlike that, the expected number of threads waiting for the barrier is not preset in the `new`
/// call, but autodetected and adapted to at runtime.
///
/// The way this is done is by cloning the original [`Barrier`] ‒ for a group to continue after
/// wait, a [`wait`][Barrier::wait] needs to be called on each clone. This allows to add or remove
/// (even implicitly by panicking) the clones as needed.
///
/// # Examples
///
/// ```rust
/// # use std::thread;
/// # use adaptive_barrier::{Barrier, PanicMode};
///
/// let barrier = Barrier::new(PanicMode::Poison);
/// let mut threads = Vec::new();
/// for _ in 0..4 {
/// // Each thread gets its own clone of the barrier. They are tied together, not independent.
/// let mut barrier = barrier.clone();
/// let thread = thread::spawn(move || {
/// // Wait to start everything at the same time
/// barrier.wait();
///
/// // ... Do some work that needs to start synchronously ...
/// // Now, if this part panics, it will *not* deadlock, it'll unlock the others just fine
/// // and propagate the panic (see the parameter to new(..)
///
/// // Wait for all threads to finish
/// if barrier.wait().is_leader() {
/// // Pick one thread to consolidate the results here
///
/// // Note that as we don't call wait any more, if we panic here, it'll not get
/// // propagated through the barrier any more.
/// }
/// });
/// threads.push(thread);
/// }
///
/// // Watch out for the last instance here in the main/controlling thread. You can either call
/// // wait on it too, or make sure it is dropped. If you don't, others will keep waiting for it.
/// drop(barrier);
///
/// for thread in threads {
/// thread.join().expect("Propagating thread panic");
/// }
/// ```
pub struct Barrier(Arc<Shared>);
impl Barrier {
/// Creates a new (independent) barrier.
///
/// To create more handles to the same barrier, clone it.
///
/// The panic mode specifies what to do if a barrier observes a panic (is dropped while
/// panicking).
pub fn new(panic_mode: PanicMode) -> Self {
Barrier(Arc::new(Shared {
inner: Mutex::new(Inner {
active: 1, // this thread
waiting: 0,
gen: 0,
leader: false,
poisoned: false,
}),
condvar: Condvar::new(),
panic_mode,
}))
}
fn check_release(&self, lock: &mut MutexGuard<'_, Inner>) {
if lock.check_release() {
self.0.condvar.notify_all();
}
}
/// Wait for all the other threads to wait too.
///
/// This'll block until all threads holding clones of the same barrier call `wait`.
///
/// # Panics
///
/// If the barrier was created with [`PanicMode::Poison`] and some other clone of the barrier
/// observed a panic, this'll also panic (even if it was already parked inside).
pub fn wait(&mut self) -> WaitResult {
let mut lock = self.0.inner.lock().unwrap();
lock.waiting += 1;
let gen = lock.gen;
self.check_release(&mut lock);
while gen == lock.gen {
lock = self.0.condvar.wait(lock).unwrap();
}
if lock.poisoned {
drop(lock); // Make sure we don't poison the mutex too
panic!("Barrier is poisoned");
}
WaitResult {
is_leader: mem::replace(&mut lock.leader, false),
}
}
}
impl Clone for Barrier {
fn clone(&self) -> Self {
let new = Arc::clone(&self.0);
new.inner.lock().unwrap().active += 1;
Barrier(new)
}
}
impl Drop for Barrier {
fn drop(&mut self) {
let mut lock = self.0.inner.lock().unwrap();
lock.active -= 1;
if self.0.panic_mode == PanicMode::Poison && thread::panicking() {
lock.poisoned = true;
}
self.check_release(&mut lock);
}
}
impl Debug for Barrier {
fn fmt(&self, fmt: &mut Formatter) -> FmtResult {
fmt.pad("Barrier { .. }")
}
}
impl Default for Barrier {
fn default() -> Self {
Self::new(PanicMode::Decrement)
}
}
// We deal with panics explicitly.
impl UnwindSafe for Barrier {}
#[cfg(test)]
mod tests {
use std::panic;
use std::sync::atomic::Ordering::*;
use std::sync::atomic::{AtomicBool, AtomicUsize};
use std::thread::{self, sleep};
use std::time::Duration;
use super::*;
/// When we have just one instance, it doesn't wait.
#[test]
fn single() {
let mut bar = Barrier::new(PanicMode::Decrement);
assert!(bar.wait().is_leader());
}
/// Check the barriers wait for each other.
#[test]
fn dispatch() {
let mut bar = Barrier::new(PanicMode::Decrement);
let waited = Arc::new(AtomicBool::new(false));
let t = thread::spawn({
let mut bar = bar.clone();
let waited = Arc::clone(&waited);
move || {
bar.wait();
waited.store(true, SeqCst);
bar.wait();
}
});
sleep(Duration::from_millis(50));
assert!(!waited.load(SeqCst));
bar.wait();
bar.wait();
assert!(waited.load(SeqCst));
t.join().unwrap();
}
#[test]
fn adjust_up() {
let mut bar = Barrier::new(PanicMode::Decrement);
let woken = Arc::new(AtomicUsize::new(0));
let t1 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
let t2 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
bar.wait();
bar.wait();
assert_eq!(woken.load(SeqCst), 2);
t1.join().unwrap();
t2.join().unwrap();
}
#[test]
fn adjust_down() {
let mut bar = Barrier::new(PanicMode::Decrement);
let woken = Arc::new(AtomicUsize::new(0));
let t1 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
}
});
let t2 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
// Only one wait, the second one will be done on only 2 copies
bar.wait();
woken.fetch_add(1, SeqCst);
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
bar.wait();
t2.join().unwrap();
bar.wait();
assert_eq!(woken.load(SeqCst), 2);
t1.join().unwrap();
}
#[test]
fn adjust_panic() {
let mut bar = Barrier::new(PanicMode::Decrement);
let woken = Arc::new(AtomicUsize::new(0));
let t1 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
woken.fetch_add(1, SeqCst);
}
});
let t2 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
// Only one wait, the second one will be done on only 2 copies
bar.wait();
woken.fetch_add(1, SeqCst);
panic!("We are going to panic, woohooo, the thing still adjusts");
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
bar.wait();
t2.join().unwrap_err();
bar.wait();
t1.join().unwrap();
assert_eq!(woken.load(SeqCst), 3);
}
#[test]
fn adjust_drop() {
let bar = Barrier::new(PanicMode::Decrement);
let woken = Arc::new(AtomicUsize::new(0));
let t1 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
let t2 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::clone(&woken);
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
drop(bar);
t1.join().unwrap();
t2.join().unwrap();
assert_eq!(woken.load(SeqCst), 2);
}
/// Poisoning of the barrier.
#[test]
#[cfg_attr(clippy, allow(clippy::redundant_clone))]
fn poisoning() {
let mut bar = Barrier::new(PanicMode::Poison);
let woken = Arc::new(AtomicUsize::new(0));
let t1 = thread::spawn({
let mut bar = bar.clone();
let woken = Arc::new(AtomicUsize::new(0));
move || {
bar.wait();
woken.fetch_add(1, SeqCst);
bar.wait();
}
});
sleep(Duration::from_millis(50));
assert_eq!(woken.load(SeqCst), 0);
let t2 = thread::spawn({
let bar = bar.clone();
move || {
// Make sure this one gets into the closure so we destroy it on the panic.
// Not issue in practice where it would get pulled in by .wait(), but we don't have
// one here in test.
let _bar = bar;
panic!("Testing a panic");
}
});
// The thread 2 panics
t2.join().unwrap_err();
// And the panic propagates to t1, even though we still hold our copy of barrier.
t1.join().unwrap_err();
// Our last instance would panic too.
panic::catch_unwind(move || bar.wait()).unwrap_err();
}
}