Trait BackoffStrategy 

pub trait BackoffStrategy:
    Default
    + Send
    + Sync
    + 'static {
    const BACKOFF: bool = true;

    // Required method
    fn backoff(&mut self) -> RetryStrategy;

    // Provided methods
    fn will_reload(&self) -> bool { ... }
    fn backoff_reload<T: PartialEq, F: FnMut() -> T>(
        &mut self,
        current: T,
        reload: F,
    ) -> T { ... }
    fn backoff_until<C: BackoffUntilCondition, F: FnMut() -> C>(
        &mut self,
        f: F,
    ) -> C::Result { ... }
}

Backoff strategy to be used after an atomic compare-and-swap (CAS) failure and in spin loops.

Waiting before retrying a failed CAS can greatly reduce the contention on the atomic’s cache line, and improve the performance of CAS loops.

Spin loops also benefit from backoff as it avoids keeping the CPU 100% busy while waiting, at little latency cost.

Provided Associated Constants

const BACKOFF: bool = true

Whether the strategy does backoff or not.

Some algorithms may have a different behavior depending on whether backoff is used; for example switching between an unbounded spin loop or a thread parking algorithm. This constant can be used for this purpose.

It should be set to false only for NoBackoff.

Required Methods

fn backoff(&mut self) -> RetryStrategy

Performs backoff and returns how the CAS should be retried.

will_reload should also be implemented accordingly.

In spin loops, the returned value can simply be ignored.

Provided Methods

fn will_reload(&self) -> bool

Returns true if the next call to backoff will not return RetryStrategy::NoReload.

This hint can be used to downgrade the failure ordering of the CAS to Relaxed when the returned value is overwritten anyway.

fn backoff_reload<T: PartialEq, F: FnMut() -> T>( &mut self, current: T, reload: F, ) -> T

Performs backoff after a failed CAS and reloads the atomic value according to the returned RetryStrategy.

ReloadUntilUnchanged causes this function to loop until the value stops changing. If the value returned by the failed CAS must be tested between reloads, use BackoffState instead.

Example

fn update_with_backoff(
    atomic: &AtomicUsize,
    set_order: Ordering,
    fetch_order: Ordering,
    mut f: impl FnMut(usize) -> usize,
    mut strategy: impl BackoffStrategy,
) -> usize {
    let mut current = atomic.load(fetch_order);
    loop {
        let failure_order = if strategy.will_reload() {
            Ordering::Relaxed
        } else {
            fetch_order
        };
        match atomic.compare_exchange_weak(current, f(current), set_order, failure_order) {
            Ok(x) => return x,
            Err(cur) => current = strategy.backoff_reload(cur, || atomic.load(fetch_order)),
        }
    }
}

fn backoff_until<C: BackoffUntilCondition, F: FnMut() -> C>( &mut self, f: F, ) -> C::Result

Loops until a condition is fulfilled, performing backoff at each iteration.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors

impl BackoffStrategy for NoBackoff

const BACKOFF: bool = false

impl BackoffStrategy for SpinBackoff

impl<const SPIN_LIMIT: usize, const UNTIL_UNCHANGED_LIMIT: usize, const YIELD_AFTER: usize> BackoffStrategy for ExponentialBackoff<SPIN_LIMIT, UNTIL_UNCHANGED_LIMIT, YIELD_AFTER>