1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
//! # Leaky Bucket Rate-Limiting (as a meter) in Rust
//!
//! This crate provides generic rate-limiting interfaces and implements a
//! few rate-limiting algorithms. The generic rate-limiting interfaces
//! are extendable to persistence-based rate limiting schemes such as
//! databases.
//!
//! This crate currently provides in-memory implementations of a by-key
//! (limits enforced per key, e.g. an IP address or a customer ID) and a
//! simple (one limit per object) state tracker.
//!
//! ## Interface
//!
//! This crate implements two "serious" rate-limiting/traffic-shaping
//! algorithms:
//! [GCRA](https://en.wikipedia.org/wiki/Generic_cell_rate_algorithm)
//! and a [Leaky
//! Bucket](https://en.wikipedia.org/wiki/Leaky_bucket#As_a_meter). An
//! "unserious" implementation is provided also: The
//! [`Allower`](example_algorithms/struct.Allower.html), which returns
//! "Yes" to all rate-limiting queries.
//!
//! The Generic Cell Rate Algorithm can be used by in an in-memory
//! rate limiter like so:
//!
//! ``` rust
//! use std::num::NonZeroU32;
//! use ratelimit_meter::{DirectRateLimiter, GCRA};
//!
//! # #[macro_use] extern crate nonzero_ext;
//! # extern crate ratelimit_meter;
//! # fn main () {
//! let mut lim = DirectRateLimiter::<GCRA>::per_second(nonzero!(50u32)); // Allow 50 units per second
//! assert_eq!(Ok(()), lim.check());
//! # }
//! ```
//!
//! The rate-limiter interface is intentionally geared towards only
//! providing callers with the information they need to make decisions
//! about what to do with each cell. Deciders return additional
//! information about why a cell should be denied alongside the
//! decision. This allows callers to e.g. provide better error
//! messages to users.
//!
//! As a consequence, the `ratelimit_meter` crate does not provide any
//! facility to wait until a cell would be allowed - if you require
//! this, you should use the
//! [`NonConformance`](struct.NonConformance.html) returned with
//! negative decisions and have the program wait using the method best
//! suited for this, e.g. an event loop.
//!
//! ## Using this crate effectively
//!
//! Many of the parameters in use by this crate are `NonZeroU32` -
//! since they are not very ergonomic to construct from constants
//! using stdlib means, I recommend using the
//! [nonzero_ext](https://crates.io/crates/nonzero_ext) crate, which
//! comes with a macro `nonzero!()`. This macro makes it far easier to
//! construct rate limiters without cluttering your code.
//!
//! ## Rate-limiting Algorithms
//!
//! ### Design and implementation of GCRA
//!
//! The GCRA limits the rate of cells by determining when the "next"
//! cell is expected to arrive; any cells that arrive before that time
//! are classified as non-conforming; the methods for checking cells
//! also return an expected arrival time for these cells, so that
//! callers can choose to wait (adding jitter), or reject the cell.
//!
//! Since using the GCRA results in a much smoother usage pattern, it
//! appears to be very useful for "outgoing" traffic behaviors,
//! e.g. throttling API call rates, or emails sent to a person in a
//! period of time.
//!
//! Unlike token or leaky bucket algorithms, the GCRA assumes that all
//! units of work are of the same "weight", and so allows some
//! optimizations which result in much more concise and fast code (it
//! does not even use multiplication or division in the "hot" path).
//!
//! See [the documentation of the GCRA type](algorithms/gcra/struct.GCRA.html) for
//! more details on its implementation and on trade-offs that apply to
//! it.
//!
//! ### Design and implementation of the leaky bucket
//!
//! In contrast to the GCRA, the leaky bucket algorithm does not place
//! any constraints on the next cell's arrival time: Whenever there is
//! capacity left in the bucket, it can be used. This means that the
//! distribution of "yes" decisions from heavy usage on the leaky
//! bucket rate-limiter will be clustered together. On average, the
//! cell rates of both the GCRA and the leaky bucket will be the same,
//! but in terms of observable behavior, the leaky bucket will appear
//! to allow requests at a more predictable rate.
//!
//! This kind of behavior is usually what people of online APIs expect
//! these days, which makes the leaky bucket a very popular technique
//! for rate-limiting on these kinds of services.
//!
//! The leaky bucket algorithm implemented in this crate is fairly
//! standard: It only updates the bucket fill gauge when a cell is
//! checked, and supports checking "batches" of cells in a single call
//! with no problems.
//!
//! ## Thread-safe operation
//!
//! The in-memory implementations in this crate use parking_lot
//! mutexes to ensure rate-limiting operations can happen safely
//! across threads.
//!
//! Example:
//!
//! ```
//! use std::thread;
//! use std::num::NonZeroU32;
//! use std::time::Duration;
//! use ratelimit_meter::{DirectRateLimiter, GCRA};
//!
//! # #[macro_use] extern crate nonzero_ext;
//! # extern crate ratelimit_meter;
//! # fn main () {
//! // Allow 50 units/second across all threads:
//! let mut lim = DirectRateLimiter::<GCRA>::per_second(nonzero!(50u32));
//! let mut thread_lim = lim.clone();
//! thread::spawn(move || { assert_eq!(Ok(()), thread_lim.check());});
//! assert_eq!(Ok(()), lim.check());
//! # }
//! ```
#![cfg_attr(feature = "cargo-clippy", deny(warnings))]
pub mod algorithms;
pub mod example_algorithms;
pub mod state;
pub mod test_utilities;
mod thread_safety;
extern crate failure;
#[macro_use]
extern crate failure_derive;
extern crate evmap;
#[macro_use]
extern crate nonzero_ext;
extern crate parking_lot;
use failure::Fail;
use std::num::NonZeroU32;
pub use self::algorithms::LeakyBucket;
pub use self::algorithms::NonConformance;
pub use self::algorithms::GCRA;
pub use self::state::DirectRateLimiter;
pub use self::state::KeyedRateLimiter;
/// Gives additional information about the negative outcome of a batch
/// cell decision.
///
/// Since batch queries can be made for batch sizes bigger than the
/// rate limiter parameter could accomodate, there are now two
/// possible negative outcomes:
///
/// * `BatchNonConforming` - the query is valid but the Decider can
/// not accomodate them.
///
/// * `InsufficientCapacity` - the query was invalid as the rate
/// limite parameters can never accomodate the number of cells
/// queried for.
#[derive(Fail, Debug, PartialEq)]
pub enum NegativeMultiDecision<E: Fail> {
/// A batch of cells (the first argument) is non-conforming and
/// can not be let through at this time. The second argument gives
/// information about when that batch of cells might be let
/// through again (not accounting for thundering herds and other,
/// simultaneous decisions).
#[fail(display = "{} cells: {}", _0, _1)]
BatchNonConforming(u32, E),
/// The number of cells tested (the first argument) is larger than
/// the bucket's capacity, which means the decision can never have
/// a conforming result.
#[fail(
display = "bucket does not have enough capacity to accomodate {} cells",
_0
)]
InsufficientCapacity(u32),
}
/// An error that is returned when initializing a rate limiter that is
/// too small to let a single cell through.
#[derive(Fail, Debug)]
#[fail(
display = "bucket capacity {} too small for a single cell with weight {}",
capacity,
cell_weight
)]
pub struct InconsistentCapacity {
capacity: NonZeroU32,
cell_weight: NonZeroU32,
}