2 releases
new 0.12.4 | Nov 2, 2024 |
---|---|
0.12.3 | Nov 1, 2024 |
#157 in Concurrency
173 downloads per month
180KB
3.5K
SLoC
parking_lot
Documentation (synchronization primitives)
Documentation (core parking lot API)
Documentation (type-safe lock API)
This library provides implementations of Mutex
, RwLock
, Condvar
and
Once
that are smaller, faster and more flexible than those in the Rust
standard library, as well as a ReentrantMutex
type which supports recursive
locking. It also exposes a low-level API for creating your own efficient
synchronization primitives.
When tested on x86_64 Linux, parking_lot::Mutex
was found to be 1.5x
faster than std::sync::Mutex
when uncontended, and up to 5x faster when
contended from multiple threads. The numbers for RwLock
vary depending on
the number of reader and writer threads, but are almost always faster than
the standard library RwLock
, and even up to 50x faster in some cases.
Features
The primitives provided by this library have several advantages over those in the Rust standard library:
Mutex
andOnce
only require 1 byte of storage space, whileCondvar
andRwLock
only require 1 word of storage space. On the other hand on some platforms (macOS and a few others) the standard library primitives require a dynamically allocatedBox
to hold OS-specific synchronization primitives. The small size ofMutex
in particular encourages the use of fine-grained locks to increase parallelism.- Uncontended lock acquisition and release is done through fast inline paths which only require a single atomic operation.
- Microcontention (a contended lock with a short critical section) is efficiently handled by spinning a few times while trying to acquire a lock.
- The locks are adaptive and will suspend a thread after a few failed spin attempts. This makes the locks suitable for both long and short critical sections.
Condvar
,RwLock
andOnce
work on Windows XP, unlike the standard library versions of those types.RwLock
takes advantage of hardware lock elision on processors that support it, which can lead to huge performance wins with many readers. This must be enabled with thehardware-lock-elision
feature.RwLock
uses a task-fair locking policy, which avoids reader and writer starvation, whereas the standard library version makes no guarantees.Condvar
is guaranteed not to produce spurious wakeups. A thread will only be woken up if it timed out or it was woken up by a notification.Condvar::notify_all
will only wake up a single thread and requeue the rest to wait on the associatedMutex
. This avoids a thundering herd problem where all threads try to acquire the lock at the same time.RwLock
supports atomically downgrading a write lock into a read lock.Mutex
andRwLock
allow raw unlocking without a RAII guard object.Mutex<()>
andRwLock<()>
allow raw locking without a RAII guard object.Mutex
andRwLock
support eventual fairness which allows them to be fair on average without sacrificing performance.- A
ReentrantMutex
type which supports recursive locking. - An experimental deadlock detector that works for
Mutex
,RwLock
andReentrantMutex
. This feature is disabled by default and can be enabled via thedeadlock_detection
feature. RwLock
supports atomically upgrading an "upgradable" read lock into a write lock.- Optional support for serde. Enable via the
feature
serde
. NOTE! this support is forMutex
,ReentrantMutex
, andRwLock
only;Condvar
andOnce
are not currently supported. - Lock guards can be sent to other threads when the
send_guard
feature is enabled.
The parking lot
To keep these primitives small, all thread queuing and suspending
functionality is offloaded to the parking lot. The idea behind this is
based on the Webkit WTF::ParkingLot
class, which essentially consists of a hash table mapping of lock addresses
to queues of parked (sleeping) threads. The Webkit parking lot was itself
inspired by Linux futexes,
but it is more powerful since it allows invoking callbacks while holding a queue
lock.
Nightly vs stable
There are a few restrictions when using this library on stable Rust:
- The
wasm32-unknown-unknown
target is only fully supported on nightly with-C target-feature=+atomics
inRUSTFLAGS
and-Zbuild-std=panic_abort,std
passed to cargo. parking_lot will work mostly fine on stable, the only difference is it will panic instead of block forever if you hit a deadlock. Just make sure not to enable-C target-feature=+atomics
on stable as that will allow wasm to run with multiple threads which will completely break parking_lot's concurrency guarantees.
To enable nightly-only functionality, you need to enable the nightly
feature
in Cargo (see below).
Usage
Add this to your Cargo.toml
:
[dependencies]
parking_lot = "0.12"
To enable nightly-only features, add this to your Cargo.toml
instead:
[dependencies]
parking_lot = { version = "0.12", features = ["nightly"] }
The experimental deadlock detector can be enabled with the
deadlock_detection
Cargo feature.
To allow sending MutexGuard
s and RwLock*Guard
s to other threads, enable the
send_guard
option.
Note that the deadlock_detection
and send_guard
features are incompatible
and cannot be used together.
Hardware lock elision support for x86 can be enabled with the
hardware-lock-elision
feature. This requires Rust 1.59 due to the use of
inline assembly.
The core parking lot API is provided by the parking_lot_core
crate. It is
separate from the synchronization primitives in the parking_lot
crate so that
changes to the core API do not cause breaking changes for users of parking_lot
.
Minimum Rust version
The current minimum required Rust version is 1.56. Any change to this is considered a breaking change and will require a major version bump.
License
Licensed under either of
- Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)
at your option.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.
Dependencies
~0.3–6MB
~26K SLoC