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
use crate::bucket::*; use crate::utils::{cpu_relax, make_elem}; use std::ops::Add; use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering}; use std::thread; use std::time::{Duration, Instant}; const POOL_SIZE: usize = 8; const EXPANSION_CAP: usize = 512; const SPIN_PERIOD: usize = 4; /// Configuration flag (@ bit positions): /// 1 -> If the pool is allowed to expand when under pressure const CONFIG_ALLOW_EXPANSION: usize = 1; pub(crate) enum ElemBuilder<T> { Default(fn() -> Box<T>), Builder(fn() -> T), Packer(fn(Box<T>) -> Box<T>), } struct VisitorGuard<'a>(&'a AtomicUsize); impl<'a> VisitorGuard<'a> { fn register(base: &'a (AtomicUsize, AtomicBool), get: bool) -> Option<Self> { let mut count = 8; // wait if the underlying storage is in protection mode while base.1.load(Ordering::Relaxed) { if get { return None; } cpu_relax(count); if count > 4 { count -= 1; } } base.0.fetch_add(1, Ordering::AcqRel); Some(VisitorGuard(&base.0)) } } impl<'a> Drop for VisitorGuard<'a> { fn drop(&mut self) { self.0.fetch_sub(1, Ordering::AcqRel); } } pub struct SyncPool<T> { /// The slots storage slots: Vec<Bucket2<T>>, /// the next bucket to try curr: (AtomicUsize, AtomicUsize), /// First node: how many threads are concurrently accessing the struct: /// 0 -> updating the `slots` field; /// 1 -> no one is using the pool; /// num -> number of visitors /// Second node: write barrier: /// true -> write barrier raised /// false -> no write barrier visitor_counter: (AtomicUsize, AtomicBool), /// the number of times we failed to find an in-store struct to offer miss_count: AtomicUsize, /// if we allow expansion of the pool configure: AtomicUsize, /// the handle to be invoked before putting the struct back reset_handle: Option<fn(&mut T)>, /// The builder that will be tasked to create a new instance of the data when the pool is unable /// to render one. builder: ElemBuilder<T>, } impl<T: Default> SyncPool<T> { /// Create a pool with default size of 64 pre-allocated elements in it. pub fn new() -> Self { Self::make_pool(POOL_SIZE, ElemBuilder::Default(Default::default)) } /// Create a `SyncPool` with pre-defined number of elements. Note that we will round-up /// the size such that the total number of elements in the pool will mod to 8. pub fn with_size(size: usize) -> Self { let mut pool_size = size / SLOT_CAP; if pool_size < 1 { pool_size = 1 } Self::make_pool(pool_size, ElemBuilder::Default(Default::default)) } } impl<T> SyncPool<T> { /// Create a pool with default size of 64 pre-allocated elements in it, which will use the `builder` /// handler to obtain the initialized instance of the struct, and then place the object into the /// `syncpool` for later use. /// /// Note that the handler shall be responsible for creating and initializing the struct object /// with all fields being valid. After all, they will be the same objects provided to the caller /// when invoking the `get` call. /// /// # Examples /// /// ```rust /// use syncpool::*; /// use std::vec; /// /// struct BigStruct { /// a: u32, /// b: u32, /// c: Vec<u8>, /// } /// /// let mut pool = SyncPool::with_builder(|| { /// BigStruct { /// a: 1, /// b: 42, /// c: vec::from_elem(0u8, 0x1_000_000), /// } /// }); /// /// let big_box: Box<BigStruct> = pool.get(); /// /// assert_eq!(big_box.a, 1); /// assert_eq!(big_box.b, 42); /// assert_eq!(big_box.c.len(), 0x1_000_000); /// /// pool.put(big_box); /// ``` pub fn with_builder(builder: fn() -> T) -> Self { Self::make_pool(POOL_SIZE, ElemBuilder::Builder(builder)) } /// Create a `SyncPool` with pre-defined number of elements and a packer handler. The `builder` /// handler shall essentially function the same way as in the `with_builder`, that it shall take /// the responsibility to create and initialize the element, and return the instance at the end /// of the `builder` closure. Note that we will round-up the size such that the total number of /// elements in the pool will mod to 8. pub fn with_builder_and_size(size: usize, builder: fn() -> T) -> Self { let mut pool_size = size / SLOT_CAP; if pool_size < 1 { pool_size = 1 } Self::make_pool(pool_size, ElemBuilder::Builder(builder)) } /// Create a pool with default size of 64 pre-allocated elements in it, which will use the `packer` /// handler to initialize the element that's being provided by the pool. /// /// Note that the handler shall take a boxed instance of the element that only contains /// placeholder fields, and it is the caller/handler's job to initialize the fields and pack it /// with valid and meaningful values. If the struct is valid with all-zero values, the handler /// can just return the input element. /// /// # Examples /// /// ```rust /// use syncpool::*; /// use std::vec; /// /// struct BigStruct { /// a: u32, /// b: u32, /// c: Vec<u8>, /// } /// /// let mut pool = SyncPool::with_packer(|mut src: Box<BigStruct>| { /// src.a = 1; /// src.b = 42; /// src.c = vec::from_elem(0u8, 0x1_000_000); /// src /// }); /// /// let big_box: Box<BigStruct> = pool.get(); /// /// assert_eq!(big_box.a, 1); /// assert_eq!(big_box.b, 42); /// assert_eq!(big_box.c.len(), 0x1_000_000); /// /// pool.put(big_box); /// ``` pub fn with_packer(packer: fn(Box<T>) -> Box<T>) -> Self { Self::make_pool(POOL_SIZE, ElemBuilder::Packer(packer)) } /// Create a `SyncPool` with pre-defined number of elements and a packer handler. The `packer` /// handler shall essentially function the same way as in `with_packer`, that it shall take the /// responsibility to initialize all the fields of a placeholder struct on the heap, otherwise /// the element returned by the pool will be essentially undefined, unless all the struct's /// fields can be represented by a 0 value. In addition, we will round-up the size such that /// the total number of elements in the pool will mod to 8. pub fn with_packer_and_size(size: usize, packer: fn(Box<T>) -> Box<T>) -> Self { let mut pool_size = size / SLOT_CAP; if pool_size < 1 { pool_size = 1 } Self::make_pool(pool_size, ElemBuilder::Packer(packer)) } /// Try to obtain a pre-allocated element from the pool. This method will always succeed even if /// the pool is empty or not available for anyone to access, and in this case, a new boxed-element /// will be created. pub fn get(&mut self) -> Box<T> { // update user count let guard = VisitorGuard::register(&self.visitor_counter, true); if guard.is_none() { return make_elem(&self.builder); } // start from where we're left let cap = self.slots.len(); let mut trials = cap; let mut pos: usize = self.curr.0.load(Ordering::Acquire) % cap; loop { // check this slot let slot = &mut self.slots[pos]; // try the access or move on if let Ok(i) = slot.access(true) { // try to checkout one slot let checkout = slot.checkout(i); slot.leave(i as u16); /* if slot.access(true) { // try to checkout one slot let checkout = slot.checkout(); slot.leave();*/ if let Ok(val) = checkout { // now we're locked, get the val and update internal states self.curr.0.store(pos, Ordering::Release); // done return val; } // failed to checkout, break and let the remainder logic to handle the rest break; } // hold off a bit to reduce contentions cpu_relax(SPIN_PERIOD); // update to the next position now. pos = self.curr.0.fetch_add(1, Ordering::AcqRel) % cap; trials -= 1; // we've finished 1 loop but not finding a value to extract, quit if trials == 0 { break; } } // make sure our guard has been returned if we want the correct visitor count drop(guard); self.miss_count.fetch_add(1, Ordering::Relaxed); // create a new object make_elem(&self.builder) } /// Try to return an element to the `SyncPool`. If succeed, we will return `None` to indicate that /// the value has been placed in an empty slot; otherwise, we will return `Option<Box<T>>` such /// that the caller can decide if the element shall be just discarded, or try put it back again. pub fn put(&mut self, val: Box<T>) -> Option<Box<T>> { // update user count let _guard = VisitorGuard::register(&self.visitor_counter, false); // start from where we're left let cap = self.slots.len(); let mut trials = 2 * cap; let mut pos: usize = self.curr.1.load(Ordering::Acquire) % cap; loop { // check this slot let slot = &mut self.slots[pos]; // try the access or move on if let Ok(i) = slot.access(false) { // now we're locked, get the val and update internal states self.curr.1.store(pos, Ordering::Release); // put the value back and reset slot.release(i, val, self.reset_handle); slot.leave(i as u16); return None; } /* if slot.access(false) { // now we're locked, get the val and update internal states self.curr.1.store(pos, Ordering::Release); // put the value back into the slot slot.release(val, self.reset_handle.load(Ordering::Acquire)); slot.leave(); return true; }*/ // hold off a bit to reduce contentions if trials < cap { cpu_relax(SPIN_PERIOD); } else { thread::yield_now(); } // update states pos = self.curr.1.fetch_add(1, Ordering::AcqRel) % cap; trials -= 1; // we've finished 1 loop but not finding a value to extract, quit if trials == 0 { return Some(val); } } } fn make_pool(size: usize, builder: ElemBuilder<T>) -> Self { let mut pool = SyncPool { slots: Vec::with_capacity(size), curr: (AtomicUsize::new(0), AtomicUsize::new(0)), visitor_counter: (AtomicUsize::new(1), AtomicBool::new(false)), miss_count: AtomicUsize::new(0), configure: AtomicUsize::new(0), reset_handle: None, builder, }; pool.add_slots(size, true); pool } #[inline] fn add_slots(&mut self, count: usize, fill: bool) { let filler = if fill { Some(&self.builder) } else { None }; for _ in 0..count { // self.slots.push(Bucket::new(fill)); self.slots.push(Bucket2::new(filler)); } } fn update_config(&mut self, mask: usize, target: bool) { let mut config = self.configure.load(Ordering::SeqCst); while let Err(old) = self.configure.compare_exchange( config, config ^ mask, Ordering::SeqCst, Ordering::Relaxed, ) { if !((old & mask > 0) ^ target) { // the configure already matches, we're done return; } config = old; } } } impl<T> Default for SyncPool<T> where T: Default, { fn default() -> Self { SyncPool::new() } } impl<T> Drop for SyncPool<T> { fn drop(&mut self) { self.slots.clear(); // now drop the reset handle if it's not null self.reset_handle.take(); } } pub trait PoolState { fn expansion_enabled(&self) -> bool; fn miss_count(&self) -> usize; fn capacity(&self) -> usize; fn len(&self) -> usize; fn is_empty(&self) -> bool { self.len() == 0 } } impl<T> PoolState for SyncPool<T> { fn expansion_enabled(&self) -> bool { let configure = self.configure.load(Ordering::SeqCst); configure & CONFIG_ALLOW_EXPANSION > 0 } fn miss_count(&self) -> usize { self.miss_count.load(Ordering::Acquire) } fn capacity(&self) -> usize { self.slots.len() * SLOT_CAP } fn len(&self) -> usize { self.slots .iter() .fold(0, |sum, item| sum + item.size_hint()) } } pub trait PoolManager<T> { fn reset_handle(&mut self, handle: fn(&mut T)) -> &mut Self; fn allow_expansion(&mut self, allow: bool) -> &mut Self; fn expand(&mut self, additional: usize, block: bool) -> bool; fn refill(&mut self, count: usize) -> usize; } /// The pool manager that provide many useful utilities to keep the SyncPool close to the needs of /// the caller program. impl<T> PoolManager<T> for SyncPool<T> { /// Set or update the reset handle. If set, the reset handle will be invoked every time an element /// has been returned back to the pool (i.e. calling the `put` method), regardless of if the element /// is created by the pool or not. fn reset_handle(&mut self, handle: fn(&mut T)) -> &mut Self { // busy waiting ... for the first chance a barrier owned by someone else is lowered let mut count: usize = 8; let timeout = Instant::now().add(Duration::from_millis(16)); loop { match self.visitor_counter.1.compare_exchange( false, true, Ordering::SeqCst, Ordering::Relaxed, ) { Ok(_) => break, Err(_) => { cpu_relax(count); // update the counter (and the busy wait period) count -= 1; if count < 4 { // yield the thread for later try thread::yield_now(); } else if Instant::now() > timeout { // don't block for more than 16ms return self; } } } } self.reset_handle.replace(handle); self.visitor_counter.1.store(false, Ordering::SeqCst); self } /// Set or update the settings that if we will allow the `SyncPool` to be expanded. fn allow_expansion(&mut self, allow: bool) -> &mut Self { if !(self.expansion_enabled() ^ allow) { // not flipping the configuration, return return self; } self.update_config(CONFIG_ALLOW_EXPANSION, allow); self } /// Try to expand the `SyncPool` and add more elements to it. Usually invoke this API only when /// the caller is certain that the pool is under pressure, and that a short block to the access /// of the pool won't cause serious issues, since the function will block the current caller's /// thread until it's finished (i.e. get the opportunity to raise the writer's barrier and wait /// everyone to leave). /// /// If we're unable to expand the pool, it's due to one of the following reasons: 1) someone has /// already raised the writer's barrier and is likely modifying the pool, we will leave immediately, /// and it's up to the caller if they want to try again; 2) we've waited too long but still couldn't /// obtain an exclusive access to the pool, and similar to reason 1), we will quit now. fn expand(&mut self, additional: usize, block: bool) -> bool { // if the pool isn't allowed to expand, just return if !self.expansion_enabled() { return false; } // if exceeding the upper limit, quit if self.slots.len() > EXPANSION_CAP { return false; } // raise the write barrier now, if someone has already raised the flag to indicate the // intention to write, let me go away. if self .visitor_counter .1 .compare_exchange_weak(false, true, Ordering::Acquire, Ordering::Acquire) .is_err() { return false; } // busy waiting ... for all visitors to leave let mut count: usize = 8; let safe = loop { match self .visitor_counter .0 .compare_exchange(1, 0, Ordering::SeqCst, Ordering::Relaxed) { Ok(_) => break true, Err(_) => { cpu_relax(2); count -= 1; if count < 4 { thread::yield_now(); } else if !block { break false; } } } }; if safe { // update the slots by pushing `additional` slots self.add_slots(additional, true); self.miss_count.store(0, Ordering::Release); } // update the internal states self.visitor_counter.0.store(1, Ordering::SeqCst); self.visitor_counter.1.store(false, Ordering::Release); safe } /// Due to contentious access to the pool, sometimes the `put` action could not finish and return /// the element to the pool successfully. Overtime, this could cause the number of elements in the /// pool to dwell. This would only happen slowly if we're running a very contentious multithreading /// program, but it surely could happen. If the caller detects such situation, they can invoke the /// `refill` API and try to refill the pool with elements. /// /// We will try to refill as many elements as requested fn refill(&mut self, additional: usize) -> usize { let cap = self.capacity(); let empty_slots = cap - self.len(); if empty_slots == 0 { return 0; } let quota = if additional > empty_slots { empty_slots } else { additional }; let mut count = 0; let timeout = Instant::now().add(Duration::from_millis(16)); // try to put `quota` number of elements into the pool while count < quota { let mut val = make_elem(&self.builder); let mut runs = 0; // retry to put the allocated element into the pool. while let Some(ret) = self.put(val) { val = ret; runs += 1; // timeout if Instant::now() > timeout { return count; } // check the pool length for every 4 failed attempts to put the element into the pool. if runs % 4 == 0 && self.len() == cap { return count; } // relax a bit if runs > 8 { thread::yield_now(); } else { cpu_relax(runs / 2); } } count += 1; } count } } #[cfg(test)] mod pool_tests { use super::*; use std::vec; struct BigStruct { a: u32, b: u32, c: Vec<u8>, } impl BigStruct { fn new() -> Self { BigStruct { a: 1, b: 42, c: vec::from_elem(0u8, 0x1_000_000), } } fn initializer(mut self: Box<Self>) -> Box<Self> { self.a = 1; self.b = 42; self.c = vec::from_elem(0u8, 0x1_000_000); self } } #[test] fn use_packer() { let mut pool = SyncPool::with_packer(BigStruct::initializer); let big_box = pool.get(); assert_eq!(big_box.a, 1); assert_eq!(big_box.b, 42); assert_eq!(big_box.c.len(), 0x1_000_000); } #[test] fn use_builder() { let mut pool = SyncPool::with_builder(BigStruct::new); let big_box = pool.get(); assert_eq!(big_box.a, 1); assert_eq!(big_box.b, 42); assert_eq!(big_box.c.len(), 0x1_000_000); } }