Struct thingbuf::ThingBuf

source ·
pub struct ThingBuf<T, R = DefaultRecycle> { /* private fields */ }
Available on crate feature alloc only.
Expand description

A fixed-size, lock-free, multi-producer multi-consumer (MPMC) queue.

This is a fixed-capacity, first-in, first-out data structure. Elements are enqueued at the front of the queue by calling the push method, and dequeued from the end of the queue by calling the pop method. Elements may be enqueued and dequeued from a ThingBuf concurrently by any number of threads or asynchronous tasks.

This queue is an implementation of a design by Dmitry Vyukov of 1024cores.net.

§Examples

use thingbuf::ThingBuf;

let q = ThingBuf::new(2);

// Push some values to the queue.
q.push(1).unwrap();
q.push(2).unwrap();

// Now, the queue is at capacity.
assert!(q.push(3).is_err());

§Allocation

A ThingBuf is a fixed-size, array-based queue. Elements in the queue are stored in a single array whose capacity is specified when the ThingBuf is constructed. This means that a ThingBuf requires only a single heap allocation over its entire lifespan. Calling ThingBuf::new will allocate an array to store capacity elements. Subsequent calls to push and pop will never allocate or deallocate memory.

If the size of the queue is known at compile-time, the StaticThingBuf type, which requires no heap allocations at all, can be used instead.

§Reusing Allocations

Of course, if the elements in the queue are themselves heap-allocated (such as Strings or Vecs), heap allocations and deallocations may still occur when those types are created or dropped. However, ThingBuf also provides an API for enqueueing and dequeueing elements by reference. In some use cases, this API can be used to reduce allocations for queue elements.

As an example, consider the case where multiple threads in a program format log messages and send them to a dedicated worker thread that writes those messages to a file. A naive implementation might look something like this:

use thingbuf::ThingBuf;
use std::{sync::Arc, fmt, thread, fs::File, error::Error, io::Write};

// Called by application threads to log a message.
fn log_event(q: &Arc<ThingBuf<String>>, message: &dyn fmt::Debug) {
    // Format the log line to a `String`.
    let line = format!("{:?}\n", message);
    // Send the string to the worker thread.
    let _ = q.push(line);
    // If the queue was full, ignore the error and drop the log line.
}

fn main() -> Result<(), Box<dyn Error>> {
    let log_queue = Arc::new(ThingBuf::<String>::new(1024));

    // Spawn the background worker thread.
    let q = log_queue.clone();
    let mut file = File::create("myapp.log")?;
    thread::spawn(move || {
        use std::io::Write;
        loop {
            // Pop from the queue, and write each log line to the file.
            while let Some(line) = q.pop() {
                file.write_all(line.as_bytes()).unwrap();
            }

            // No more messages in the queue!
            file.flush().unwrap();
            thread::yield_now();
        }
    });

    // ...
}

With this design, however, new Strings are allocated for every message that’s logged, and then are immediately deallocated once they are written to the file. This can have a negative performance impact.

Using ThingBuf’s push_ref and pop_ref methods, this code can be redesigned to reuse String allocations in place. With these methods, rather than moving an element by-value into the queue when enqueueing, we instead reserve the rights to mutate a slot in the queue in place, returning a Ref. Similarly, when dequeueing, we also recieve a Ref that allows reading from (and mutating) the dequeued element. This allows the queue to own the set of Strings used for formatting log messages, which are cleared in place and the existing allocation reused.

The rewritten code might look like this:

use thingbuf::ThingBuf;
use std::{sync::Arc, fmt, thread, fs::File, error::Error};

// Called by application threads to log a message.
fn log_event(q: &Arc<ThingBuf<String>>, message: &dyn fmt::Debug) {
    use std::fmt::Write;

    // Reserve a slot in the queue to write to.
    if let Ok(mut slot) = q.push_ref() {
        // Clear the string in place, retaining the allocated capacity.
        slot.clear();
        // Write the log message to the string.
        write!(&mut *slot, "{:?}\n", message);
    }
    // Otherwise, if `push_ref` returns an error, the queue is full;
    // ignore this log line.
}

fn main() -> Result<(), Box<dyn Error>> {
    let log_queue = Arc::new(ThingBuf::<String>::new(1024));

    // Spawn the background worker thread.
    let q = log_queue.clone();
    let mut file = File::create("myapp.log")?;
    thread::spawn(move || {
        use std::io::Write;
        loop {
            // Pop from the queue, and write each log line to the file.
            while let Some(line) = q.pop_ref() {
                file.write_all(line.as_bytes()).unwrap();
            }

            // No more messages in the queue!
            file.flush().unwrap();
            thread::yield_now();
        }
    });

    // ...
}

In this implementation, the strings will only be reallocated if their current capacity is not large enough to store the formatted representation of the log message.

When using a ThingBuf in this manner, it can be thought of as a combination of a concurrent queue and an object pool.

Implementations§

source§

impl<T: Default + Clone> ThingBuf<T>

source

pub fn new(capacity: usize) -> Self

Returns a new ThingBuf with space for capacity elements.

source§

impl<T, R> ThingBuf<T, R>

source

pub fn capacity(&self) -> usize

Returns the total capacity of this queue. This includes both occupied and unoccupied entries.

To determine the queue’s remaining unoccupied capacity, use remaining instead.

§Examples
use thingbuf::ThingBuf;

let q = ThingBuf::<usize>::new(100);
assert_eq!(q.capacity(), 100);

Even after pushing several messages to the queue, the capacity remains the same:


let q = ThingBuf::<usize>::new(100);

*q.push_ref().unwrap() = 1;
*q.push_ref().unwrap() = 2;
*q.push_ref().unwrap() = 3;

assert_eq!(q.capacity(), 100);
source

pub fn remaining(&self) -> usize

Returns the unoccupied capacity of the queue (i.e., how many additional elements can be enqueued before the queue will be full).

This is equivalent to subtracting the queue’s len from its capacity.

source

pub fn len(&self) -> usize

Returns the number of elements in the queue

To determine the queue’s remaining unoccupied capacity, use remaining instead.

§Examples
use thingbuf::ThingBuf;

let q = ThingBuf::new(100);
assert_eq!(q.len(), 0);

*q.push_ref().unwrap() = 1;
*q.push_ref().unwrap() = 2;
*q.push_ref().unwrap() = 3;
assert_eq!(q.len(), 3);

let _ = q.pop_ref();
assert_eq!(q.len(), 2);
source

pub fn is_empty(&self) -> bool

Returns true if there are currently no elements in this ThingBuf.

§Examples
use thingbuf::ThingBuf;

let q = ThingBuf::new(100);
assert!(q.is_empty());

*q.push_ref().unwrap() = 1;
assert!(!q.is_empty());

let _ = q.pop_ref();
assert!(q.is_empty());
source§

impl<T, R> ThingBuf<T, R>
where R: Recycle<T>,

source

pub fn with_recycle(capacity: usize, recycle: R) -> Self

Returns a new ThingBuf with space for capacity elements and the provided recycling policy.

§Panics

Panics if the capacity exceeds usize::MAX & !(1 << (usize::BITS - 1)). This value represents the highest power of two that can be expressed by a usize, excluding the most significant bit.

source

pub fn push_ref(&self) -> Result<Ref<'_, T>, Full>

Reserves a slot to push an element into the queue, returning a Ref that can be used to write to that slot.

This can be used to reuse allocations for queue elements in place, by clearing previous data prior to writing. In order to ensure allocations can be reused in place, elements should be dequeued using pop_ref rather than pop. If values are expensive to produce, push_ref can also be used to avoid producing a value if there is no capacity for it in the queue.

For values that don’t own heap allocations, or heap allocated values that cannot be reused in place, push can also be used.

§Returns
  • Ok(Ref) if there is space for a new element
  • Err(Full), if there is no capacity remaining in the queue
§Examples
use thingbuf::ThingBuf;

let q = ThingBuf::new(1);

// Reserve a `Ref` and enqueue an element.
*q.push_ref().expect("queue should have capacity") = 10;

// Now that the queue has one element in it, a subsequent `push_ref`
// call will fail.
assert!(q.push_ref().is_err());

Avoiding producing an expensive element when the queue is at capacity:

use thingbuf::ThingBuf;

#[derive(Clone, Default)]
struct Message {
    // ...
}

fn make_expensive_message() -> Message {
    // Imagine this function performs some costly operation, or acquires
    // a limited resource...
}

fn enqueue_message(q: &ThingBuf<Message>) {
    loop {
        match q.push_ref() {
            // the queue for our message, so it's okay to generate
            // the expensive message.
            Ok(mut slot) => {
                *slot = make_expensive_message();
                return;
            },

            // If there's no space in the queue, avoid generating
            // an expensive message that won't be sent.
            Err(_) => {
                // Wait until the queue has capacity...
                std::thread::yield_now();
            }
        }
    }
}
source

pub fn push(&self, val: T) -> Result<(), Full<T>>

Attempt to enqueue an element by value.

If the queue is full, the element is returned in the Full error.

Unlike push_ref, this method does not enable the reuse of previously allocated elements. If allocations are being reused by using push_ref and pop_ref, this method should not be used, as it will drop pooled allocations.

§Returns
  • Ok(()) if the element was enqueued
  • Err(Full), containing the value, if there is no capacity remaining in the queue
source

pub fn push_with<U>(&self, f: impl FnOnce(&mut T) -> U) -> Result<U, Full>

Reserves a slot to push an element into the queue, and invokes the provided function f with a mutable reference to that element.

§Returns
  • Ok(U) containing the return value of the provided function, if the element was enqueued
  • Err(Full), if there is no capacity remaining in the queue
source

pub fn pop_ref(&self) -> Option<Ref<'_, T>>

Dequeue the first element in the queue, returning a Ref that can be used to read from (or mutate) the element.

Note: A ThingBuf is not a “broadcast”-style queue. Each element is dequeued a single time. Once a thread has dequeued a given element, it is no longer the head of the queue.

This can be used to reuse allocations for queue elements in place, by clearing previous data prior to writing. In order to ensure allocations can be reused in place, elements should be enqueued using push_ref rather than push.

For values that don’t own heap allocations, or heap allocated values that cannot be reused in place, pop can also be used.

§Returns
  • Some(Ref<T>) if an element was dequeued
  • None if there are no elements in the queue
source

pub fn pop(&self) -> Option<T>

Dequeue the first element in the queue by value, moving it out of the queue.

Note: A ThingBuf is not a “broadcast”-style queue. Each element is dequeued a single time. Once a thread has dequeued a given element, it is no longer the head of the queue.

§Returns
  • Some(T) if an element was dequeued
  • None if there are no elements in the queue
source

pub fn pop_with<U>(&self, f: impl FnOnce(&mut T) -> U) -> Option<U>

Dequeue the first element in the queue by reference, and invoke the provided function f with a mutable reference to the dequeued element.

§Returns
  • Some(U) containing the return value of the provided function, if the element was dequeued
  • None if the queue is empty

Trait Implementations§

source§

impl<T, R: Debug> Debug for ThingBuf<T, R>

source§

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter. Read more
source§

impl<T, R> Drop for ThingBuf<T, R>

source§

fn drop(&mut self)

Executes the destructor for this type. Read more

Auto Trait Implementations§

§

impl<T, R = DefaultRecycle> !Freeze for ThingBuf<T, R>

§

impl<T, R = DefaultRecycle> !RefUnwindSafe for ThingBuf<T, R>

§

impl<T, R> Send for ThingBuf<T, R>
where R: Send, T: Send,

§

impl<T, R> Sync for ThingBuf<T, R>
where R: Sync, T: Sync,

§

impl<T, R> Unpin for ThingBuf<T, R>
where R: Unpin,

§

impl<T, R> UnwindSafe for ThingBuf<T, R>
where R: UnwindSafe, T: UnwindSafe,

Blanket Implementations§

source§

impl<T> Any for T
where T: 'static + ?Sized,

source§

fn type_id(&self) -> TypeId

Gets the TypeId of self. Read more
source§

impl<T> Borrow<T> for T
where T: ?Sized,

source§

fn borrow(&self) -> &T

Immutably borrows from an owned value. Read more
source§

impl<T> BorrowMut<T> for T
where T: ?Sized,

source§

fn borrow_mut(&mut self) -> &mut T

Mutably borrows from an owned value. Read more
source§

impl<T> From<T> for T

source§

fn from(t: T) -> T

Returns the argument unchanged.

source§

impl<T, U> Into<U> for T
where U: From<T>,

source§

fn into(self) -> U

Calls U::from(self).

That is, this conversion is whatever the implementation of From<T> for U chooses to do.

source§

impl<T, U> TryFrom<U> for T
where U: Into<T>,

§

type Error = Infallible

The type returned in the event of a conversion error.
source§

fn try_from(value: U) -> Result<T, <T as TryFrom<U>>::Error>

Performs the conversion.
source§

impl<T, U> TryInto<U> for T
where U: TryFrom<T>,

§

type Error = <U as TryFrom<T>>::Error

The type returned in the event of a conversion error.
source§

fn try_into(self) -> Result<U, <U as TryFrom<T>>::Error>

Performs the conversion.