grin/core/src/consensus.rs

358 lines
12 KiB
Rust
Raw Normal View History

2016-11-11 03:02:47 +03:00
// Copyright 2016 The Grin Developers
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//! All the rules required for a cryptocurrency to have reach consensus across
//! the whole network are complex and hard to completely isolate. Some can be
//! simple parameters (like block reward), others complex algorithms (like
//! Merkle sum trees or reorg rules). However, as long as they're simple
2016-11-14 22:21:18 +03:00
//! enough, consensus-relevant constants and short functions should be kept
2016-11-11 03:02:47 +03:00
//! here.
use std::fmt;
use std::cmp::max;
use ser;
use core::target::Difficulty;
/// A grin is divisible to 10^9, a nanogrin
pub const GRIN_BASE: u64 = 1_000_000_000;
2016-11-11 03:02:47 +03:00
/// The block subsidy amount
pub const REWARD: u64 = 50 * GRIN_BASE;
2016-11-11 03:02:47 +03:00
/// Actual block reward for a given total fee amount
pub fn reward(fee: u64) -> u64 {
REWARD + fee / 2
}
/// Number of blocks before a coinbase matures and can be spent
pub const COINBASE_MATURITY: u64 = 1_000;
/// Block interval, in seconds, the network will tune its next_target for. Note
/// that we may reduce this value in the future as we get more data on mining
/// with Cuckoo Cycle, networks improve and block propagation is optimized
/// (adjusting the reward accordingly).
pub const BLOCK_TIME_SEC: u64 = 60;
2016-11-11 03:02:47 +03:00
/// Cuckoo-cycle proof size (cycle length)
pub const PROOFSIZE: usize = 42;
/// Default Cuckoo Cycle size shift used for mining and validating.
pub const DEFAULT_SIZESHIFT: u8 = 30;
2016-11-11 03:02:47 +03:00
/// Default Cuckoo Cycle easiness, high enough to have good likeliness to find
/// a solution.
pub const EASINESS: u32 = 50;
/// Default number of blocks in the past when cross-block cut-through will start
/// happening. Needs to be long enough to not overlap with a long reorg.
/// Rational
/// behind the value is the longest bitcoin fork was about 30 blocks, so 5h. We
/// add an order of magnitude to be safe and round to 48h of blocks to make it
/// easier to reason about.
pub const CUT_THROUGH_HORIZON: u32 = 48 * 3600 / (BLOCK_TIME_SEC as u32);
/// The maximum size we're willing to accept for any message. Enforced by the
/// peer-to-peer networking layer only for DoS protection.
pub const MAX_MSG_LEN: u64 = 20_000_000;
2017-10-06 06:34:35 +03:00
/// Weight of an input when counted against the max block weigth capacity
pub const BLOCK_INPUT_WEIGHT: usize = 1;
/// Weight of an output when counted against the max block weight capacity
pub const BLOCK_OUTPUT_WEIGHT: usize = 10;
/// Weight of a kernel when counted against the max block weight capacity
pub const BLOCK_KERNEL_WEIGHT: usize = 2;
/// Total maximum block weight
pub const MAX_BLOCK_WEIGHT: usize = 80_000;
/// Whether a block exceeds the maximum acceptable weight
pub fn exceeds_weight(input_len: usize, output_len: usize, kernel_len: usize) -> bool {
input_len * BLOCK_INPUT_WEIGHT + output_len * BLOCK_OUTPUT_WEIGHT
+ kernel_len * BLOCK_KERNEL_WEIGHT > MAX_BLOCK_WEIGHT
2017-10-06 06:34:35 +03:00
}
2017-10-10 03:08:17 +03:00
/// Fork every 250,000 blocks for first 2 years, simple number and just a
/// little less than 6 months.
pub const HARD_FORK_INTERVAL: u64 = 250_000;
/// Check whether the block version is valid at a given height, implements
/// 6 months interval scheduled hard forks for the first 2 years.
pub fn valid_header_version(height: u64, version: u16) -> bool {
// uncomment below as we go from hard fork to hard fork
if height <= HARD_FORK_INTERVAL && version == 1 {
true
/* } else if height <= 2 * HARD_FORK_INTERVAL && version == 2 {
true */
/* } else if height <= 3 * HARD_FORK_INTERVAL && version == 3 {
true */
/* } else if height <= 4 * HARD_FORK_INTERVAL && version == 4 {
true */
/* } else if height > 4 * HARD_FORK_INTERVAL && version > 4 {
true */
} else {
false
}
}
/// The minimum mining difficulty we'll allow
pub const MINIMUM_DIFFICULTY: u64 = 10;
/// Time window in blocks to calculate block time median
pub const MEDIAN_TIME_WINDOW: u64 = 11;
/// Number of blocks used to calculate difficulty adjustments
pub const DIFFICULTY_ADJUST_WINDOW: u64 = 23;
/// Average time span of the difficulty adjustment window
pub const BLOCK_TIME_WINDOW: u64 = DIFFICULTY_ADJUST_WINDOW * BLOCK_TIME_SEC;
/// Maximum size time window used for difficulty adjustments
pub const UPPER_TIME_BOUND: u64 = BLOCK_TIME_WINDOW * 4 / 3;
/// Minimum size time window used for difficulty adjustments
pub const LOWER_TIME_BOUND: u64 = BLOCK_TIME_WINDOW * 5 / 6;
/// Error when computing the next difficulty adjustment.
#[derive(Debug, Clone)]
pub struct TargetError(pub String);
impl fmt::Display for TargetError {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "Error computing new difficulty: {}", self.0)
}
}
/// Computes the proof-of-work difficulty that the next block should comply
/// with. Takes an iterator over past blocks, from latest (highest height) to
/// oldest (lowest height). The iterator produces pairs of timestamp and
/// difficulty for each block.
///
/// The difficulty calculation is based on both Digishield and GravityWave
/// family of difficulty computation, coming to something very close to Zcash.
/// The refence difficulty is an average of the difficulty over a window of
/// 23 blocks. The corresponding timespan is calculated by using the
/// difference between the median timestamps at the beginning and the end
/// of the window.
pub fn next_difficulty<T>(cursor: T) -> Result<Difficulty, TargetError>
2017-09-29 21:44:25 +03:00
where
T: IntoIterator<Item = Result<(u64, Difficulty), TargetError>>,
{
// Block times at the begining and end of the adjustment window, used to
// calculate medians later.
let mut window_begin = vec![];
let mut window_end = vec![];
// Sum of difficulties in the window, used to calculate the average later.
let mut diff_sum = Difficulty::zero();
// Enumerating backward over blocks
for (n, head_info) in cursor.into_iter().enumerate() {
let m = n as u64;
let (ts, diff) = head_info?;
// Sum each element in the adjustment window. In addition, retain
// timestamps within median windows (at ]start;start-11] and ]end;end-11]
// to later calculate medians.
if m < DIFFICULTY_ADJUST_WINDOW {
diff_sum = diff_sum + diff;
if m < MEDIAN_TIME_WINDOW {
window_begin.push(ts);
}
} else if m < DIFFICULTY_ADJUST_WINDOW + MEDIAN_TIME_WINDOW {
window_end.push(ts);
} else {
break;
}
}
// Check we have enough blocks
if window_end.len() < (MEDIAN_TIME_WINDOW as usize) {
return Ok(Difficulty::minimum());
}
// Calculating time medians at the beginning and end of the window.
window_begin.sort();
window_end.sort();
let begin_ts = window_begin[window_begin.len() / 2];
let end_ts = window_end[window_end.len() / 2];
// Average difficulty and dampened average time
let diff_avg = diff_sum.clone() / Difficulty::from_num(DIFFICULTY_ADJUST_WINDOW);
let ts_damp = (3 * BLOCK_TIME_WINDOW + (begin_ts - end_ts)) / 4;
// Apply time bounds
let adj_ts = if ts_damp < LOWER_TIME_BOUND {
LOWER_TIME_BOUND
} else if ts_damp > UPPER_TIME_BOUND {
UPPER_TIME_BOUND
} else {
ts_damp
};
let difficulty =
diff_avg * Difficulty::from_num(BLOCK_TIME_WINDOW) / Difficulty::from_num(adj_ts);
Ok(max(difficulty, Difficulty::minimum()))
}
/// Consensus rule that collections of items are sorted lexicographically over the wire.
pub trait VerifySortOrder<T> {
/// Verify a collection of items is sorted as required.
fn verify_sort_order(&self) -> Result<(), ser::Error>;
}
#[cfg(test)]
use std;
#[cfg(test)]
mod test {
use core::target::Difficulty;
use super::*;
// Builds an iterator for next difficulty calculation with the provided
// constant time interval, difficulty and total length.
fn repeat(interval: u64, diff: u64, len: u64) -> Vec<Result<(u64, Difficulty), TargetError>> {
2017-09-29 21:44:25 +03:00
// watch overflow here, length shouldn't be ridiculous anyhow
assert!(len < std::usize::MAX as u64);
let diffs = vec![Difficulty::from_num(diff); len as usize];
let times = (0..(len as usize)).map(|n| n * interval as usize).rev();
let pairs = times.zip(diffs.iter());
2017-09-29 21:44:25 +03:00
pairs
.map(|(t, d)| Ok((t as u64, d.clone())))
.collect::<Vec<_>>()
}
2017-09-29 21:44:25 +03:00
fn repeat_offs(
from: u64,
interval: u64,
diff: u64,
len: u64,
) -> Vec<Result<(u64, Difficulty), TargetError>> {
map_vec!(repeat(interval, diff, len), |e| match e.clone() {
Err(e) => Err(e),
Ok((t, d)) => Ok((t + from, d)),
})
}
/// Checks different next_target adjustments and difficulty boundaries
#[test]
fn next_target_adjustment() {
// not enough data
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(vec![]).unwrap(),
Difficulty::from_num(MINIMUM_DIFFICULTY)
);
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(vec![Ok((60, Difficulty::one()))]).unwrap(),
Difficulty::from_num(MINIMUM_DIFFICULTY)
);
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(repeat(60, 10, DIFFICULTY_ADJUST_WINDOW)).unwrap(),
Difficulty::from_num(MINIMUM_DIFFICULTY)
);
// just enough data, right interval, should stay constant
let just_enough = DIFFICULTY_ADJUST_WINDOW + MEDIAN_TIME_WINDOW;
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(repeat(60, 1000, just_enough)).unwrap(),
Difficulty::from_num(1000)
);
// checking averaging works, window length is odd so need to compensate a little
let sec = DIFFICULTY_ADJUST_WINDOW / 2 + 1 + MEDIAN_TIME_WINDOW;
let mut s1 = repeat(60, 500, sec);
let mut s2 = repeat_offs((sec * 60) as u64, 60, 1545, DIFFICULTY_ADJUST_WINDOW / 2);
s2.append(&mut s1);
assert_eq!(next_difficulty(s2).unwrap(), Difficulty::from_num(999));
// too slow, diff goes down
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(repeat(90, 1000, just_enough)).unwrap(),
Difficulty::from_num(889)
);
assert_eq!(
next_difficulty(repeat(120, 1000, just_enough)).unwrap(),
Difficulty::from_num(800)
);
// too fast, diff goes up
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(repeat(55, 1000, just_enough)).unwrap(),
Difficulty::from_num(1021)
);
assert_eq!(
next_difficulty(repeat(45, 1000, just_enough)).unwrap(),
Difficulty::from_num(1067)
);
// hitting lower time bound, should always get the same result below
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(repeat(20, 1000, just_enough)).unwrap(),
Difficulty::from_num(1200)
);
assert_eq!(
next_difficulty(repeat(10, 1000, just_enough)).unwrap(),
Difficulty::from_num(1200)
);
// hitting higher time bound, should always get the same result above
2017-09-29 21:44:25 +03:00
assert_eq!(
next_difficulty(repeat(160, 1000, just_enough)).unwrap(),
Difficulty::from_num(750)
);
assert_eq!(
next_difficulty(repeat(200, 1000, just_enough)).unwrap(),
Difficulty::from_num(750)
);
// We should never drop below MINIMUM_DIFFICULTY (10)
assert_eq!(
next_difficulty(repeat(90, 10, just_enough)).unwrap(),
Difficulty::from_num(10)
);
}
2017-10-10 03:08:17 +03:00
#[test]
fn hard_fork_1() {
assert!(valid_header_version(0, 1));
assert!(valid_header_version(10, 1));
assert!(!valid_header_version(10, 2));
assert!(valid_header_version(250_000, 1));
assert!(!valid_header_version(250_001, 1));
assert!(!valid_header_version(500_000, 1));
assert!(!valid_header_version(250_001, 2));
}
// #[test]
// fn hard_fork_2() {
// assert!(valid_header_version(0, 1));
// assert!(valid_header_version(10, 1));
// assert!(valid_header_version(10, 2));
// assert!(valid_header_version(250_000, 1));
// assert!(!valid_header_version(250_001, 1));
// assert!(!valid_header_version(500_000, 1));
// assert!(valid_header_version(250_001, 2));
// assert!(valid_header_version(500_000, 2));
// assert!(!valid_header_version(500_001, 2));
// }
}