#circuit-breaker #ring-buffer #call #implemented #len #failure-rate

recloser

A concurrent circuit breaker implemented with ring buffers

6 releases (3 stable)

1.1.1 Feb 11, 2024
1.1.0 Jul 27, 2022
1.0.0 Aug 9, 2021
0.3.0 Sep 22, 2020
0.1.0 Jul 3, 2019

#76 in Concurrency

Download history 2413/week @ 2023-12-04 1441/week @ 2023-12-11 1065/week @ 2023-12-18 430/week @ 2023-12-25 615/week @ 2024-01-01 1919/week @ 2024-01-08 1498/week @ 2024-01-15 1371/week @ 2024-01-22 1278/week @ 2024-01-29 1189/week @ 2024-02-05 1709/week @ 2024-02-12 2096/week @ 2024-02-19 1626/week @ 2024-02-26 1255/week @ 2024-03-04 1243/week @ 2024-03-11 943/week @ 2024-03-18

5,115 downloads per month

MIT license

25KB
529 lines

recloser   latest doc

A concurrent circuit breaker implemented with ring buffers.

The Recloser struct provides a call(...) method to wrap function calls that may fail, it will eagerly reject them when some failure_rate is reached, and it will allow them again after some time. A future aware version of call(...) is also available through an AsyncRecloser wrapper.

The API is largely based on failsafe and the ring buffer implementation on resilient4j.

Usage

The Recloser can be in three states:

  • State::Closed(RingBuffer(len)): The initial Recloser's state. At least len calls will be performed before calculating a failure_rate based on which transitions to State::Open(_) state may happen.
  • State::Open(duration): All calls will return Err(Error::Rejected) until duration has elapsed, then transition to State::HalfOpen(_) state will happen.
  • State::HalfOpen(RingBuffer(len)): At least len calls will be performed before calculating a failure_rate based on which transitions to either State::Closed(_) or State::Open(_) states will happen.

The state transition settings can be customized as follows:

use std::time::Duration;
use recloser::Recloser;

// Equivalent to Recloser::default()
let recloser = Recloser::custom()
   .error_rate(0.5)
   .closed_len(100)
   .half_open_len(10)
   .open_wait(Duration::from_secs(30))
   .build();

Wrapping dangerous function calls in order to control failure propagation:

use recloser::{Recloser, Error};

// Performs 1 call before calculating failure_rate
let recloser = Recloser::custom().closed_len(1).build();

let f1 = || Err::<(), usize>(1);

// First call, just recorded as an error
let res = recloser.call(f1);
assert!(matches!(res, Err(Error::Inner(1))));

// Now also computes failure_rate, that is 100% here
// Will transition to State::Open afterward
let res = recloser.call(f1);
assert!(matches!(res, Err(Error::Inner(1))));

let f2 = || Err::<(), i64>(-1);

// All calls are rejected (while in State::Open)
let res = recloser.call(f2);
assert!(matches!(res, Err(Error::Rejected)));

It is also possible to discard some errors on a per call basis. This behavior is controlled by the ErrorPredicate<E>trait, which is already implemented for all Fn(&E) -> bool.

use recloser::{Recloser, Error};

let recloser = Recloser::default();

let f = || Err::<(), usize>(1);

// Custom predicate that doesn't consider usize values as errors
let p = |_: &usize| false;

// Will not record resulting Err(1) as an error
let res = recloser.call_with(p, f);
assert!(matches!(res, Err(Error::Inner(1))));

Wrapping functions that return Futures requires to use an AsyncRecloser that just wraps a regular Recloser.

use std::future;
use recloser::{Recloser, Error, AsyncRecloser};

let recloser = AsyncRecloser::from(Recloser::default());

let future = future::ready::<Result<(), usize>>(Err(1));
let future = recloser.call(future);

Performances

Benchmarks for Recloser and failsafe::CircuitBreaker

  • Single threaded workload: same performances
  • Multi threaded workload: Recloser has 10x better performances
recloser_simple         time:   [355.17 us 358.67 us 362.52 us]
failsafe_simple         time:   [403.47 us 406.90 us 410.29 us]
recloser_concurrent     time:   [668.44 us 674.26 us 680.48 us]
failsafe_concurrent     time:   [11.523 ms 11.613 ms 11.694 ms]

These benchmarks were run on a Intel Core i7-6700HQ @ 8x 3.5GHz CPU.

Dependencies

~0.6–1.1MB
~23K SLoC