pub struct Cache<A, T> { /* private fields */ }
Expand description
Caching handle for ArcSwapAny
.
Instead of loading the Arc
on every request from the shared storage, this keeps
another copy inside itself. Upon request it only cheaply revalidates it is up to
date. If it is, access is significantly faster. If it is stale, the load_full is done and the
cache value is replaced. Under a read-heavy loads, the measured speedup are 10-25 times,
depending on the architecture.
There are, however, downsides:
- The handle needs to be kept around by the caller (usually, one per thread). This is fine if
there’s one global
ArcSwapAny
, but starts being tricky with eg. data structures build from them. - As it keeps a copy of the Arc inside the cache, the old value may be kept alive for longer period of time ‒ it is replaced by the new value on load. You may not want to use this if dropping the old value in timely manner is important (possibly because of releasing large amount of RAM or because of closing file handles).
Examples
use std::sync::Arc;
use std::sync::atomic::{AtomicBool, Ordering};
use arc_swap::{ArcSwap, Cache};
let shared = Arc::new(ArcSwap::from_pointee(42));
let terminate = Arc::new(AtomicBool::new(false));
// Start 10 worker threads...
for _ in 0..10 {
let mut cache = Cache::new(Arc::clone(&shared));
let terminate = Arc::clone(&terminate);
std::thread::spawn(move || {
// Keep loading it like mad..
while !terminate.load(Ordering::Relaxed) {
let value = cache.load();
do_something(value);
}
});
}
shared.store(Arc::new(12));
Another one with using a thread local storage and explicit types:
static CURRENT_CONFIG: Lazy<ArcSwap<Config>> = Lazy::new(|| ArcSwap::from_pointee(Config::default()));
thread_local! {
static CACHE: RefCell<Cache<&'static ArcSwap<Config>, Arc<Config>>> = RefCell::new(Cache::from(CURRENT_CONFIG.deref()));
}
CACHE.with(|c| {
// * RefCell needed, because load on cache is `&mut`.
// * You want to operate inside the `with` ‒ cloning the Arc is comparably expensive as
// ArcSwap::load itself and whatever you'd save by the cache would be lost on that.
println!("{:?}", c.borrow_mut().load());
});
Implementations
sourceimpl<A, T, S> Cache<A, T> where
A: Deref<Target = ArcSwapAny<T, S>>,
T: RefCnt,
S: Strategy<T>,
impl<A, T, S> Cache<A, T> where
A: Deref<Target = ArcSwapAny<T, S>>,
T: RefCnt,
S: Strategy<T>,
sourcepub fn new(arc_swap: A) -> Self
pub fn new(arc_swap: A) -> Self
Creates a new caching handle.
The parameter is something dereferencing into an ArcSwapAny
(eg. either to ArcSwap
or ArcSwapOption
). That can be ArcSwapAny
itself, but that’s not very useful. But
it also can be a reference to it or Arc
, which makes it possible to share the
ArcSwapAny
with multiple caches or access it in non-cached way too.
sourcepub fn arc_swap(&self) -> &A::Target
pub fn arc_swap(&self) -> &A::Target
Gives access to the (possibly shared) cached ArcSwapAny
.
sourcepub fn load(&mut self) -> &T
pub fn load(&mut self) -> &T
Loads the currently held value.
This first checks if the cached value is up to date. This check is very cheap.
If it is up to date, the cached value is simply returned without additional costs. If it is outdated, a load is done on the underlying shared storage. The newly loaded value is then stored in the cache and returned.
sourcepub fn map<F, U>(self, f: F) -> MapCache<A, T, F> where
F: FnMut(&T) -> &U,
pub fn map<F, U>(self, f: F) -> MapCache<A, T, F> where
F: FnMut(&T) -> &U,
Turns this cache into a cache with a projection inside the cached value.
You’d use this in case when some part of code needs access to fresh values of U
, however
a bigger structure containing U
is provided by this cache. The possibility of giving the
whole structure to the part of the code falls short in terms of reusability (the part of
the code could be used within multiple contexts, each with a bigger different structure
containing U
) and code separation (the code shouldn’t needs to know about the big
structure).
Warning
As the provided f
is called inside every load
, this one should be
cheap. Most often it is expected to be just a closure taking reference of some inner field.
For the same reasons, it should not have side effects and should never panic (these will not break Rust’s safety rules, but might produce behaviour you don’t expect).
Examples
use arc_swap::ArcSwap;
use arc_swap::cache::{Access, Cache};
struct InnerCfg {
answer: usize,
}
struct FullCfg {
inner: InnerCfg,
}
fn use_inner<A: Access<InnerCfg>>(cache: &mut A) {
let value = cache.load();
println!("The answer is: {}", value.answer);
}
let full_cfg = ArcSwap::from_pointee(FullCfg {
inner: InnerCfg {
answer: 42,
}
});
let cache = Cache::new(&full_cfg);
use_inner(&mut cache.map(|full| &full.inner));
let inner_cfg = ArcSwap::from_pointee(InnerCfg { answer: 24 });
let mut inner_cache = Cache::new(&inner_cfg);
use_inner(&mut inner_cache);
Trait Implementations
sourceimpl<A, T, S> Access<<T as Deref>::Target> for Cache<A, T> where
A: Deref<Target = ArcSwapAny<T, S>>,
T: Deref<Target = <T as RefCnt>::Base> + RefCnt,
S: Strategy<T>,
impl<A, T, S> Access<<T as Deref>::Target> for Cache<A, T> where
A: Deref<Target = ArcSwapAny<T, S>>,
T: Deref<Target = <T as RefCnt>::Base> + RefCnt,
S: Strategy<T>,
Auto Trait Implementations
impl<A, T> RefUnwindSafe for Cache<A, T> where
A: RefUnwindSafe,
T: RefUnwindSafe,
impl<A, T> Send for Cache<A, T> where
A: Send,
T: Send,
impl<A, T> Sync for Cache<A, T> where
A: Sync,
T: Sync,
impl<A, T> Unpin for Cache<A, T> where
A: Unpin,
T: Unpin,
impl<A, T> UnwindSafe for Cache<A, T> where
A: UnwindSafe,
T: UnwindSafe,
Blanket Implementations
sourceimpl<T> BorrowMut<T> for T where
T: ?Sized,
impl<T> BorrowMut<T> for T where
T: ?Sized,
const: unstable · sourcefn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more