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
// This file is part of Substrate.

// Copyright (C) 2020-2022 Parity Technologies (UK) Ltd.
// SPDX-License-Identifier: Apache-2.0

// 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.

//! Balancing algorithm implementation.
//!
//! Given a committee `A` and an edge weight vector `w`, a balanced solution is one that
//!
//! 1.  it maximizes the sum of member supports, i.e `Argmax { sum(support(c)) }`. for all `c` in
//! `A`.
//! 2. it minimizes the sum of supports squared, i.e `Argmin { sum(support(c).pow(2)) }` for all `c`
//! in `A`.
//!
//! See [`balance`] for more information.

use crate::{BalancingConfig, Edge, ExtendedBalance, IdentifierT, Voter};
use sp_arithmetic::traits::Zero;
use sp_std::prelude::*;

/// Balance the weight distribution of a given `voters` at most `iterations` times, or up until the
/// point where the biggest difference created per iteration of all stakes is `tolerance`. If this
/// is called with `tolerance = 0`, then exactly `iterations` rounds will be executed, except if no
/// change has been made (`difference = 0`). `tolerance` and `iterations` are part of the
/// [`BalancingConfig`] struct.
///
/// In almost all cases, a balanced solution will have a better score than an unbalanced solution,
/// yet this is not 100% guaranteed because the first element of a [`crate::ElectionScore`] does not
/// directly relate to balancing.
///
/// Note that some reference implementation adopt an approach in which voters are balanced randomly
/// per round. To advocate determinism, we don't do this. In each round, all voters are exactly
/// balanced once, in the same order.
///
/// Also, note that due to re-distribution of weights, the outcome of this function might contain
/// edges with weight zero. The call site should filter such weight if desirable. Moreover, the
/// outcome might need balance re-normalization, see `Voter::try_normalize`.
///
/// ### References
///
/// - [A new approach to the maximum flow problem](https://dl.acm.org/doi/10.1145/48014.61051).
/// - [Validator election in nominated proof-of-stake](https://arxiv.org/abs/2004.12990) (Appendix
///   A.)
/// - [Computing a balanced solution](https://research.web3.foundation/en/latest/polkadot/NPoS/3.%20Balancing.html),
///   which contains the details of the algorithm implementation.
pub fn balance<AccountId: IdentifierT>(
	voters: &mut Vec<Voter<AccountId>>,
	config: &BalancingConfig,
) -> usize {
	if config.iterations == 0 {
		return 0
	}

	let mut iter = 0;
	loop {
		let mut max_diff = 0;
		for voter in voters.iter_mut() {
			let diff = balance_voter(voter, config.tolerance);
			if diff > max_diff {
				max_diff = diff;
			}
		}

		iter += 1;
		if max_diff <= config.tolerance || iter >= config.iterations {
			break iter
		}
	}
}

/// Internal implementation of balancing for one voter.
pub(crate) fn balance_voter<AccountId: IdentifierT>(
	voter: &mut Voter<AccountId>,
	tolerance: ExtendedBalance,
) -> ExtendedBalance {
	// create a shallow copy of the elected ones. The original one will not be used henceforth.
	let mut elected_edges = voter
		.edges
		.iter_mut()
		.filter(|e| e.candidate.borrow().elected)
		.collect::<Vec<&mut Edge<AccountId>>>();

	// Either empty, or a self vote. Not much to do in either case.
	if elected_edges.len() <= 1 {
		return Zero::zero()
	}

	// amount of stake from this voter that is used in edges.
	let stake_used =
		elected_edges.iter().fold(0, |a: ExtendedBalance, e| a.saturating_add(e.weight));

	// backed stake of each of the elected edges.
	let backed_stakes = elected_edges
		.iter()
		.map(|e| e.candidate.borrow().backed_stake)
		.collect::<Vec<_>>();

	// backed stake of all the edges for whom we've spent some stake.
	let backing_backed_stake = elected_edges
		.iter()
		.filter_map(|e| if e.weight > 0 { Some(e.candidate.borrow().backed_stake) } else { None })
		.collect::<Vec<_>>();

	let difference = if backing_backed_stake.len() > 0 {
		let max_stake = backing_backed_stake
			.iter()
			.max()
			.expect("vector with positive length will have a max; qed");
		let min_stake = backed_stakes
			.iter()
			.min()
			.expect("iterator with positive length will have a min; qed");
		let mut difference = max_stake.saturating_sub(*min_stake);
		difference = difference.saturating_add(voter.budget.saturating_sub(stake_used));
		if difference < tolerance {
			return difference
		}
		difference
	} else {
		voter.budget
	};

	// remove all backings.
	for edge in elected_edges.iter_mut() {
		let mut candidate = edge.candidate.borrow_mut();
		candidate.backed_stake = candidate.backed_stake.saturating_sub(edge.weight);
		edge.weight = 0;
	}

	elected_edges.sort_by_key(|e| e.candidate.borrow().backed_stake);

	let mut cumulative_backed_stake = Zero::zero();
	let mut last_index = elected_edges.len() - 1;

	for (index, edge) in elected_edges.iter().enumerate() {
		let index = index as ExtendedBalance;
		let backed_stake = edge.candidate.borrow().backed_stake;
		let temp = backed_stake.saturating_mul(index);
		if temp.saturating_sub(cumulative_backed_stake) > voter.budget {
			// defensive only. length of elected_edges is checked to be above 1.
			last_index = index.saturating_sub(1) as usize;
			break
		}
		cumulative_backed_stake = cumulative_backed_stake.saturating_add(backed_stake);
	}

	let last_stake = elected_edges
		.get(last_index)
		.expect(
			"length of elected_edges is greater than or equal 2; last_index index is at the \
 			 minimum elected_edges.len() - 1; index is within range; qed",
		)
		.candidate
		.borrow()
		.backed_stake;
	let ways_to_split = last_index + 1;
	let excess = voter
		.budget
		.saturating_add(cumulative_backed_stake)
		.saturating_sub(last_stake.saturating_mul(ways_to_split as ExtendedBalance));

	// Do the final update.
	for edge in elected_edges.into_iter().take(ways_to_split) {
		// first, do one scoped borrow to get the previous candidate stake.
		let candidate_backed_stake = {
			let candidate = edge.candidate.borrow();
			candidate.backed_stake
		};

		let new_edge_weight = (excess / ways_to_split as ExtendedBalance)
			.saturating_add(last_stake)
			.saturating_sub(candidate_backed_stake);

		// write the new edge weight
		edge.weight = new_edge_weight;

		// write the new candidate stake
		let mut candidate = edge.candidate.borrow_mut();
		candidate.backed_stake = candidate.backed_stake.saturating_add(new_edge_weight);
	}

	// excess / ways_to_split can cause a small un-normalized voters to be created.
	// We won't `expect` here because even a result which is not normalized is not corrupt;
	let _ = voter.try_normalize_elected();

	difference
}