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
//! EbrCell - A concurrently readable cell with Ebr
//!
//! An `EbrCell` can be used in place of a `RwLock`. Readers are guaranteed that
//! the data will not change during the lifetime of the read. Readers do
//! not block writers, and writers do not block readers. Writers are serialised
//! same as the write in a `RwLock`.
//!
//! This is the Ebr collected implementation.
//! Ebr is the crossbeam-epoch based reclaim system for async memory
//! garbage collection. Ebr is faster than `Arc`,
//! but long transactions can cause the memory usage to grow very quickly
//! before a garbage reclaim. This is a space time trade, where you gain
//! performance at the expense of delaying garbage collection. Holding Ebr
//! reads for too long may impact garbage collection of other epoch structures
//! or crossbeam library components.
//! If you need accurate memory reclaim, use the Arc (`CowCell`) implementation.
use crossbeam_epoch as epoch;
use crossbeam_epoch::{Atomic, Guard, Owned};
use std::sync::atomic::Ordering::{Acquire, Release};
use parking_lot::{Mutex, MutexGuard};
use std::mem;
use std::ops::{Deref, DerefMut};
/// An `EbrCell` Write Transaction handle.
///
/// This allows mutation of the content of the `EbrCell` without blocking or
/// affecting current readers.
///
/// Changes are only stored in the structure until you call commit: to
/// abort a change, don't call commit and allow the write transaction to
/// go out of scope. This causes the `EbrCell` to unlock allowing other
/// writes to proceed.
pub struct EbrCellWriteTxn<'a, T: 'a> {
data: Option<T>,
// This way we know who to contact for updating our data ....
caller: &'a EbrCell<T>,
_guard: MutexGuard<'a, ()>,
}
impl<'a, T> EbrCellWriteTxn<'a, T>
where
T: Clone + Send + 'static,
{
/// Access a mutable pointer of the data in the `EbrCell`. This data is only
/// visible to this write transaction object in this thread until you call
/// 'commit'.
pub fn get_mut(&mut self) -> &mut T {
self.data.as_mut().unwrap()
}
/// Commit the changes in this write transaction to the `EbrCell`. This will
/// consume the transaction so that further changes can not be made to it
/// after this function is called.
pub fn commit(mut self) {
/* Write our data back to the EbrCell */
// Now make a new dummy element, and swap it into the mutex
// This fixes up ownership of some values for lifetimes.
let mut element: Option<T> = None;
mem::swap(&mut element, &mut self.data);
self.caller.commit(element);
}
}
impl<'a, T> Deref for EbrCellWriteTxn<'a, T> {
type Target = T;
#[inline]
fn deref(&self) -> &T {
self.data.as_ref().unwrap()
}
}
impl<'a, T> DerefMut for EbrCellWriteTxn<'a, T> {
fn deref_mut(&mut self) -> &mut T {
self.data.as_mut().unwrap()
}
}
/// A concurrently readable cell.
///
/// This structure behaves in a similar manner to a `RwLock<Box<T>>`. However
/// unlike a read-write lock, writes and parallel reads can be performed
/// simultaneously. This means writes do not block reads or reads do not
/// block writes.
///
/// To achieve this a form of "copy-on-write" (or for Rust, clone on write) is
/// used. As a write transaction begins, we clone the existing data to a new
/// location that is capable of being mutated.
///
/// Readers are guaranteed that the content of the `EbrCell` will live as long
/// as the read transaction is open, and will be consistent for the duration
/// of the transaction. There can be an "unlimited" number of readers in parallel
/// accessing different generations of data of the `EbrCell`.
///
/// Data that is copied is garbage collected using the crossbeam-epoch library.
///
/// Writers are serialised and are guaranteed they have exclusive write access
/// to the structure.
///
/// # Examples
/// ```
/// use concread::ebrcell::EbrCell;
///
/// let data: i64 = 0;
/// let ebrcell = EbrCell::new(data);
///
/// // Begin a read transaction
/// let read_txn = ebrcell.read();
/// assert_eq!(*read_txn, 0);
/// {
/// // Now create a write, and commit it.
/// let mut write_txn = ebrcell.write();
/// *write_txn = 1;
/// // Commit the change
/// write_txn.commit();
/// }
/// // Show the previous generation still reads '0'
/// assert_eq!(*read_txn, 0);
/// let new_read_txn = ebrcell.read();
/// // And a new read transaction has '1'
/// assert_eq!(*new_read_txn, 1);
/// ```
#[derive(Debug)]
pub struct EbrCell<T> {
write: Mutex<()>,
active: Atomic<T>,
}
impl<T> EbrCell<T>
where
T: Clone + Send + 'static,
{
/// Create a new `EbrCell` storing type `T`. `T` must implement `Clone`.
pub fn new(data: T) -> Self {
EbrCell {
write: Mutex::new(()),
active: Atomic::new(data),
}
}
/// Begin a write transaction, returning a write guard.
pub fn write(&self) -> EbrCellWriteTxn<T> {
/* Take the exclusive write lock first */
let mguard = self.write.lock();
/* Do an atomic load of the current value */
let guard = epoch::pin();
let cur_shared = self.active.load(Acquire, &guard);
/* Now build the write struct, we'll discard the pin shortly! */
EbrCellWriteTxn {
/* This is the 'copy' of the copy on write! */
data: Some(unsafe { cur_shared.deref().clone() }),
caller: self,
_guard: mguard,
}
}
/// Attempt to begin a write transaction. If it's already held,
/// `None` is returned.
pub fn try_write(&self) -> Option<EbrCellWriteTxn<T>> {
self.write.try_lock().map(|mguard| {
let guard = epoch::pin();
let cur_shared = self.active.load(Acquire, &guard);
/* Now build the write struct, we'll discard the pin shortly! */
EbrCellWriteTxn {
/* This is the 'copy' of the copy on write! */
data: Some(unsafe { cur_shared.deref().clone() }),
caller: self,
_guard: mguard,
}
})
}
/// This is an internal compontent of the commit cycle. It takes ownership
/// of the value stored in the writetxn, and commits it to the main EbrCell
/// safely.
///
/// In theory you could use this as a "lock free" version, but you don't
/// know if you are trampling a previous change, so it's private and we
/// let the writetxn struct serialise and protect this interface.
fn commit(&self, element: Option<T>) {
// Yield a read txn?
let guard = epoch::pin();
// Load the previous data ready for unlinking
let prev_data = self.active.load(Acquire, &guard);
// Make the data Owned, and set it in the active.
let owned_data: Owned<T> = Owned::new(element.unwrap());
let _shared_data = self
.active
.compare_and_set(prev_data, owned_data, Release, &guard);
// Finally, set our previous data for cleanup.
unsafe {
guard.defer(move || {
drop(prev_data.into_owned());
});
}
// Then return the current data with a readtxn. Do we need a new guard scope?
}
/// Begin a read transaction. The returned [`EbrCellReadTxn'] guarantees
/// the data lives long enough via crossbeam's Epoch type. When this is
/// dropped the data *may* be freed at some point in the future.
pub fn read(&self) -> EbrCellReadTxn<T> {
let guard = epoch::pin();
// This option returns None on null pointer, but we can never be null
// as we have to init with data, and all replacement ALWAYS gives us
// a ptr, so unwrap?
let cur = {
let c = self.active.load(Acquire, &guard);
c.as_raw()
};
EbrCellReadTxn {
_guard: guard,
data: cur,
}
}
}
impl<T> Drop for EbrCell<T> {
fn drop(&mut self) {
// Right, we are dropping! Everything is okay here *except*
// that we need to tell our active data to be unlinked, else it may
// be dropped "unsafely".
let guard = epoch::pin();
let prev_data = self.active.load(Acquire, &guard);
unsafe {
guard.defer(move || {
drop(prev_data.into_owned());
});
}
}
}
/// A read transaction. This stores a reference to the data from the main
/// `EbrCell`, and guarantees it is alive for the duration of the read.
// #[derive(Debug)]
pub struct EbrCellReadTxn<T> {
_guard: Guard,
data: *const T,
}
impl<T> Deref for EbrCellReadTxn<T> {
type Target = T;
/// Derference and access the value within the read transaction.
fn deref(&self) -> &T {
unsafe { &(*self.data) }
}
}
#[cfg(test)]
mod tests {
extern crate crossbeam_utils;
extern crate time;
use std::sync::atomic::{AtomicUsize, Ordering};
use super::EbrCell;
use crossbeam_utils::thread::scope;
#[test]
fn test_deref_mut() {
let data: i64 = 0;
let cc = EbrCell::new(data);
{
/* Take a write txn */
let mut cc_wrtxn = cc.write();
*cc_wrtxn = 1;
cc_wrtxn.commit();
}
let cc_rotxn = cc.read();
assert_eq!(*cc_rotxn, 1);
}
#[test]
fn test_try_write() {
let data: i64 = 0;
let cc = EbrCell::new(data);
/* Take a write txn */
let cc_wrtxn_a = cc.try_write();
assert!(cc_wrtxn_a.is_some());
/* Because we already hold the writ, the second is guaranteed to fail */
let cc_wrtxn_a = cc.try_write();
assert!(cc_wrtxn_a.is_none());
}
#[test]
fn test_simple_create() {
let data: i64 = 0;
let cc = EbrCell::new(data);
let cc_rotxn_a = cc.read();
assert_eq!(*cc_rotxn_a, 0);
{
/* Take a write txn */
let mut cc_wrtxn = cc.write();
/* Get the data ... */
{
let mut_ptr = cc_wrtxn.get_mut();
/* Assert it's 0 */
assert_eq!(*mut_ptr, 0);
*mut_ptr = 1;
assert_eq!(*mut_ptr, 1);
}
assert_eq!(*cc_rotxn_a, 0);
let cc_rotxn_b = cc.read();
assert_eq!(*cc_rotxn_b, 0);
/* The write txn and it's lock is dropped here */
cc_wrtxn.commit();
}
/* Start a new txn and see it's still good */
let cc_rotxn_c = cc.read();
assert_eq!(*cc_rotxn_c, 1);
assert_eq!(*cc_rotxn_a, 0);
}
const MAX_TARGET: i64 = 2000;
#[test]
fn test_multithread_create() {
let start = time::now();
// Create the new ebrcell.
let data: i64 = 0;
let cc = EbrCell::new(data);
scope(|scope| {
let cc_ref = &cc;
let _readers: Vec<_> = (0..7)
.map(|_| {
scope.spawn(move || {
let mut last_value: i64 = 0;
while last_value < MAX_TARGET {
let cc_rotxn = cc_ref.read();
{
assert!(*cc_rotxn >= last_value);
last_value = *cc_rotxn;
}
}
})
}).collect();
let _writers: Vec<_> = (0..3)
.map(|_| {
scope.spawn(move || {
let mut last_value: i64 = 0;
while last_value < MAX_TARGET {
let mut cc_wrtxn = cc_ref.write();
{
let mut_ptr = cc_wrtxn.get_mut();
assert!(*mut_ptr >= last_value);
last_value = *mut_ptr;
*mut_ptr = *mut_ptr + 1;
}
cc_wrtxn.commit();
}
})
}).collect();
});
let end = time::now();
print!("Ebr MT create :{} ", end - start);
}
static GC_COUNT: AtomicUsize = AtomicUsize::new(0);
#[derive(Debug, Clone)]
struct TestGcWrapper<T> {
data: T,
}
impl<T> Drop for TestGcWrapper<T> {
fn drop(&mut self) {
// Add to the atomic counter ...
GC_COUNT.fetch_add(1, Ordering::Release);
}
}
fn test_gc_operation_thread(cc: &EbrCell<TestGcWrapper<i64>>) {
while GC_COUNT.load(Ordering::Acquire) < 50 {
// thread::sleep(std::time::Duration::from_millis(200));
{
let mut cc_wrtxn = cc.write();
{
let mut_ptr = cc_wrtxn.get_mut();
mut_ptr.data = mut_ptr.data + 1;
}
cc_wrtxn.commit();
}
}
}
#[test]
fn test_gc_operation() {
GC_COUNT.store(0, Ordering::Release);
let data = TestGcWrapper { data: 0 };
let cc = EbrCell::new(data);
scope(|scope| {
let cc_ref = &cc;
let _writers: Vec<_> = (0..3)
.map(|_| {
scope.spawn(move || {
test_gc_operation_thread(cc_ref);
})
}).collect();
});
assert!(GC_COUNT.load(Ordering::Acquire) >= 50);
}
}
#[cfg(test)]
mod tests_linear {
use std::sync::atomic::{AtomicUsize, Ordering};
use super::EbrCell;
static GC_COUNT: AtomicUsize = AtomicUsize::new(0);
#[derive(Debug, Clone)]
struct TestGcWrapper<T> {
data: T,
}
impl<T> Drop for TestGcWrapper<T> {
fn drop(&mut self) {
// Add to the atomic counter ...
GC_COUNT.fetch_add(1, Ordering::Release);
}
}
#[test]
fn test_gc_operation_linear() {
/*
* Test if epoch drops in order (or ordered enough).
* A property required for b+tree with cow is that txn's
* are dropped in order so that tree states are not invalidated.
*
* A -> B -> C
*
* If B is dropped, it invalidates nodes copied from A
* causing the tree to corrupt txn A (and maybe C).
*
* EBR due to it's design while it won't drop in order,
* it drops generationally, in blocks. This is probably
* good enough. This means that:
*
* A -> B -> C .. -> X -> Y
*
* EBR will drop in blocks such as:
*
* | g1 | g2 | live |
* A -> B -> C .. -> X -> Y
*
* This test is "small" but asserts a basic sanity of drop
* ordering, but it's not conclusive for b+tree. More testing
* (likely multi-thread strees test) is needed, or analysis from
* other EBR developers.
*/
GC_COUNT.store(0, Ordering::Release);
let data = TestGcWrapper { data: 0 };
let cc = EbrCell::new(data);
// Open a read A.
let cc_rotxn_a = cc.read();
// open a write, change and commit
{
let mut cc_wrtxn = cc.write();
{
let mut_ptr = cc_wrtxn.get_mut();
mut_ptr.data = mut_ptr.data + 1;
}
cc_wrtxn.commit();
}
// open a read B.
let cc_rotxn_b = cc.read();
// open a write, change and commit
{
let mut cc_wrtxn = cc.write();
{
let mut_ptr = cc_wrtxn.get_mut();
mut_ptr.data = mut_ptr.data + 1;
}
cc_wrtxn.commit();
}
// open a read C
let cc_rotxn_c = cc.read();
assert!(GC_COUNT.load(Ordering::Acquire) == 0);
// drop B
drop(cc_rotxn_b);
// gc count should be 0.
assert!(GC_COUNT.load(Ordering::Acquire) == 0);
// drop C
drop(cc_rotxn_c);
// gc count should be 0
assert!(GC_COUNT.load(Ordering::Acquire) == 0);
// drop A
drop(cc_rotxn_a);
// gc count should be 2 (A + B, C is still live)
assert!(GC_COUNT.load(Ordering::Acquire) <= 2);
}
}