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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
//! Multi-probe consistent hashing implementation based on [this paper](https://arxiv.org/pdf/1505.00062.pdf).
//!
//! # Overview
//!
//! Multi-probe consistent hashing is a variant of consistent hashing that
//! doesn't require virtual nodes to achieve the same load balancing properties.
//!
//! Nodes are assigned randomly to positions on the ring, however the key is
//! not assigned to the next clockwise node, but instead multiple probes (using
//! double-hashing) are made, and attempt/position with the closest distance to
//! some node wins -- that node is considered owning the key space for the key.
//!
//! This means that nodes that happen to control wide range of the key space
//! still do not get an increased chance of being selected, as most probes for
//! such nodes will happen to possess not the smallest distance to a key.
//!
//! # Main structures
//!
//! [`HashRing`] is the main structure that holds the ring state. For cases
//! where ring of the `u64` size is not big enough, you will need to define your
//! own [`Partitioner`].
//!
//! # Usage
//! ```
//! use {
//! mpchash::{HashRing, RingDirection::Clockwise},
//! rand::Rng,
//! };
//!
//! // Define a node type.
//! #[derive(Hash, Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd)]
//! struct Node {
//! id: u64,
//! }
//! impl Node {
//! fn random() -> Self {
//! Self {
//! id: rand::thread_rng().gen(),
//! }
//! }
//! }
//!
//! // Create a new ring.
//! let mut ring = HashRing::new();
//!
//! // Populate the ring with some nodes.
//! let node1 = Node::random();
//! let node2 = Node::random();
//! ring.add(node1);
//! ring.add(node2);
//!
//! // Get the primary key space owning node for a key.
//! // This will return the first node when moving clockwise from the key's position.
//! let node = ring.primary_node(&42).unwrap();
//!
//! // Remove a node from the ring.
//! ring.remove(&node1);
//!
//! // Iterate over the ring.
//! for (position, node) in ring.tokens(0, Clockwise) {
//! println!("{}: {:?}", position, node);
//! }
//! ```
mod iter;
mod partitioner;
mod range;
use {
crate::{
iter::HashRingIter,
RingDirection::{Clockwise, CounterClockwise},
},
std::{
collections::BTreeMap,
fmt::Debug,
hash::Hash,
ops::Bound::{Excluded, Unbounded},
},
};
pub use {partitioner::*, range::*};
/// Number of probing attempts before selecting key's position on the ring.
///
/// The probe with minimal distance to some assigned node is selected. Then the
/// first node when moving clockwise from the selected probe is deemed to be key
/// owner.
pub const DEFAULT_PROBE_COUNT: u16 = 23;
/// Position on the ring.
pub type RingPosition = u64;
/// Node that can be assigned a position on the ring.
pub trait RingNode: Hash + Clone + Copy + Debug + Eq + PartialEq + Ord + PartialOrd {}
/// Blanket implementation of `RingNode` for all types that implement the
/// necessary traits.
impl<T> RingNode for T where T: Hash + Clone + Copy + Debug + Eq + PartialEq + Ord + PartialOrd {}
/// An ownership over a position on the ring (by the object of type `T`,
/// normally, `RingNode`).
pub type RingToken<'a, T> = (&'a RingPosition, &'a T);
/// Defines the direction in which the ring is traversed.
#[derive(Clone, Copy)]
pub enum RingDirection {
Clockwise,
CounterClockwise,
}
/// Consistent hash ring.
///
/// Nodes are assigned positions on the ring, effectively becoming responsible
/// for a range of keys: from the previous node (counter-clockwise) up to and
/// not including the node's position.
#[derive(Clone)]
pub struct HashRing<N: RingNode, P = DefaultPartitioner> {
/// Partitioner used to compute ring positions.
partitioner: P,
/// The ring positions assigned to nodes (sorted in ascending order).
positions: BTreeMap<RingPosition, N>,
/// The number of positions to probe for a given key.
probe_count: u16,
}
impl<N: RingNode> Default for HashRing<N> {
fn default() -> Self {
Self {
partitioner: DefaultPartitioner::new(),
positions: BTreeMap::new(),
probe_count: DEFAULT_PROBE_COUNT,
}
}
}
impl<N: RingNode> Debug for HashRing<N> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("HashRing")
.field("positions", &self.positions)
.field("probe_count", &self.probe_count)
.finish_non_exhaustive()
}
}
impl<N: RingNode> HashRing<N> {
/// Creates a new hash ring.
///
/// Any type implementing [`RingNode`] can be used as a node type.
///
/// # Examples
///
/// Create ring with `u64` nodes:
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// ring.add(0);
/// ring.add(2)
/// ```
///
/// Create ring with custom node type:
/// ```
/// use mpchash::{HashRing, RingNode};
///
/// #[derive(Hash, Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd)]
/// struct Node {
/// id: u64,
/// }
///
/// let mut ring = HashRing::<Node>::new();
/// ring.add(Node { id: 0 });
/// ring.add(Node { id: 2 });
/// ```
pub fn new() -> Self {
Self::default()
}
/// Adds a new node to the ring.
///
/// The position is computed deterministically using keyspace partitioner.
///
/// # Examples
///
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// ring.add(0);
/// ```
pub fn add(&mut self, node: N) {
let pos = self.partitioner.position(&node);
self.positions.insert(pos, node);
}
/// Inserts a node to a given ring position.
///
/// Mostly useful for testing and simulation, use `add` in all other cases.
///
/// # Examples
///
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// // Insert node "15" at position 0.
/// ring.insert(0, 15);
/// // Insert node "16" at position 1.
/// ring.insert(1, 16);
/// ```
pub fn insert(&mut self, pos: RingPosition, node: N) {
self.positions.insert(pos, node);
}
/// Removes a node from the ring.
///
/// # Examples
///
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// ring.add(42);
/// ring.remove(&42);
/// ```
pub fn remove(&mut self, node: &N) {
let pos = self.partitioner.position(node);
self.positions.remove(&pos);
}
/// Returns the primary node responsible for the given key.
///
/// Due to replication, a key may land on several nodes, but the primary
/// destination is the node controlling ring position coming immediately
/// after the key.
///
/// # Examples
///
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// for i in 0..6 {
/// ring.add(i);
/// }
/// for i in 0..100 {
/// println!(
/// "key {i} should go to node {}",
/// ring.primary_node(&i).expect("no node found for key")
/// );
/// }
/// ```
pub fn primary_node<K: Hash>(&self, key: &K) -> Option<&N> {
self.primary_token(key).map(|token| token.1)
}
/// Returns the token of a node that owns a range for the given key.
///
/// A token is a pair of a ring position of a node and a node itself.
///
/// In replicated setting a single range is owned by multiple nodes (which
/// are basically the first `n` nodes when moving clockwise from the
/// selected probe), but the first node is considered as primary.
///
/// Double hashing is used to avoid non-uniform distribution of keys across
/// the ring. From the multiple produced positions, the one with the
/// minimal distance to the next node is selected.
///
/// # Examples
///
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// for i in 0..6 {
/// ring.add(i);
/// }
/// for i in 0..100 {
/// let (pos, node) = ring
/// .primary_token(&i)
/// .expect("no primary token found for key");
/// println!("key {i} should go to node {node} at position {pos}");
/// }
/// ```
pub fn primary_token<K: Hash>(&self, key: &K) -> Option<RingToken<N>> {
let mut min_distance = RingPosition::MAX;
let mut min_token = None;
let h1 = self.partitioner.position_seeded(key, DEFAULT_SEED1);
let h2 = self.partitioner.position_seeded(key, DEFAULT_SEED2);
// Calculate several positions for the given key and select the one with the
// minimal distance to the owner.
for i in 0..self.probe_count {
// pos = h1 + i * h2
let pos = h1.wrapping_add((i as RingPosition).wrapping_mul(h2));
// Find the peer that owns the position, and calculate the distance to it.
match self.tokens(pos, Clockwise).next() {
Some((next_pos, next_peer_id)) => {
let distance = distance(pos, *next_pos);
if distance < min_distance {
min_distance = distance;
min_token = Some((next_pos, next_peer_id));
}
}
None => {
return None;
}
};
}
min_token
}
/// Returns assigned node positions (tokens) starting from the given
/// location on the ring.
///
/// One can go in both directions, clockwise and counter-clockwise, allowing
/// to see both the next assigned positions and the previous ones. Since we
/// position nodes on a ring, when maximum position is reached, the next
/// position is the minimum one (positions wrap around). Hence, we chain
/// another iterator, to account for this semantics.
///
/// # Examples
///
/// ```
/// let mut ring = mpchash::HashRing::<u64>::new();
/// for i in 0..6 {
/// ring.add(i);
/// }
/// for (pos, node) in ring.tokens(0, mpchash::RingDirection::Clockwise) {
/// println!("node {} is at position {}", node, pos);
/// }
/// // We can move in both directions.
/// for (pos, node) in ring.tokens(0, mpchash::RingDirection::CounterClockwise) {
/// println!("node {} is at position {}", node, pos);
/// }
/// ```
#[must_use]
pub fn tokens(
&self,
start: RingPosition,
dir: RingDirection,
) -> impl DoubleEndedIterator<Item = RingToken<N>> {
match dir {
Clockwise => HashRingIter::Clockwise(
self.positions
.range(start..)
.chain(self.positions.range(0..start)),
),
CounterClockwise => HashRingIter::CounterClockwise(
self.positions
.range(..=start)
.rev()
// We must exclude start position i.e. `(start..)`.
.chain(self.positions.range((Excluded(start), Unbounded)).rev()),
),
}
}
/// Returns the key space range owned by a node, if it was located at given
/// position.
///
/// If range is available, it always ends at the given position, and starts
/// at the position to the left (counter-clockwise) of the provided `pos`.
/// If range is not available, on an empty ring, for example, `None` is
/// returned.
///
/// Note: since we semantically treat the ordered set as a ring, the key
/// range wraps around.
pub fn key_range(&self, pos: RingPosition) -> Option<KeyRange<RingPosition>> {
if self.positions.is_empty() {
return None;
}
let prev_pos = self.tokens(pos, Clockwise).next_back();
let start = prev_pos.map_or(0, |token| *token.0);
Some(KeyRange::new(start, pos))
}
/// Returns ring position to which a given key will be assigned.
pub fn position<K: Hash>(&self, key: &K) -> RingPosition {
self.partitioner.position(key)
}
/// Returns size of the ring, i.e. number of contained tokens.
pub fn len(&self) -> usize {
self.positions.len()
}
/// Returns `true` if the ring is empty.
pub fn is_empty(&self) -> bool {
self.positions.is_empty()
}
}
/// Calculates distance between two ring positions.
fn distance(pos1: RingPosition, pos2: RingPosition) -> RingPosition {
if pos1 > pos2 {
RingPosition::MAX - pos1 + pos2
} else {
pos2 - pos1
}
}