Line data Source code
1 : use std::ops::Range;
2 :
3 : // NOTE the `im` crate has 20x more downloads and also has
4 : // persistent/immutable BTree. But it's bugged so rpds is a
5 : // better choice <https://github.com/neondatabase/neon/issues/3395>
6 : use rpds::RedBlackTreeMapSync;
7 :
8 : /// Data structure that can efficiently:
9 : /// - find the latest layer by lsn.end at a given key
10 : /// - iterate the latest layers in a key range
11 : /// - insert layers in non-decreasing lsn.start order
12 : ///
13 : /// For a detailed explanation and justification of this approach, see:
14 : /// <https://neon.tech/blog/persistent-structures-in-neons-wal-indexing>
15 : ///
16 : /// NOTE The struct is parameterized over Value for easier
17 : /// testing, but in practice it's some sort of layer.
18 : pub struct LayerCoverage<Value> {
19 : /// For every change in coverage (as we sweep the key space)
20 : /// we store (lsn.end, value).
21 : ///
22 : /// NOTE We use an immutable/persistent tree so that we can keep historic
23 : /// versions of this coverage without cloning the whole thing and
24 : /// incurring quadratic memory cost. See HistoricLayerCoverage.
25 : ///
26 : /// NOTE We use the Sync version of the map because we want Self to
27 : /// be Sync. Using nonsync might be faster, if we can work with
28 : /// that.
29 : nodes: RedBlackTreeMapSync<i128, Option<(u64, Value)>>,
30 : }
31 :
32 : impl<T: Clone> Default for LayerCoverage<T> {
33 1652 : fn default() -> Self {
34 1652 : Self::new()
35 1652 : }
36 : }
37 :
38 : impl<Value: Clone> LayerCoverage<Value> {
39 1652 : pub fn new() -> Self {
40 1652 : Self {
41 1652 : nodes: RedBlackTreeMapSync::default(),
42 1652 : }
43 1652 : }
44 :
45 : /// Helper function to subdivide the key range without changing any values
46 : ///
47 : /// This operation has no semantic effect by itself. It only helps us pin in
48 : /// place the part of the coverage we don't want to change when inserting.
49 : ///
50 : /// As an analogy, think of a polygon. If you add a vertex along one of the
51 : /// segments, the polygon is still the same, but it behaves differently when
52 : /// we move or delete one of the other points.
53 : ///
54 : /// Complexity: O(log N)
55 10136 : fn add_node(&mut self, key: i128) {
56 10136 : let value = match self.nodes.range(..=key).last() {
57 6862 : Some((_, Some(v))) => Some(v.clone()),
58 2682 : Some((_, None)) => None,
59 592 : None => None,
60 : };
61 10136 : self.nodes.insert_mut(key, value);
62 10136 : }
63 :
64 : /// Insert a layer.
65 : ///
66 : /// Complexity: worst case O(N), in practice O(log N). See NOTE in implementation.
67 5068 : pub fn insert(&mut self, key: Range<i128>, lsn: Range<u64>, value: Value) {
68 5068 : // Add nodes at endpoints
69 5068 : //
70 5068 : // NOTE The order of lines is important. We add nodes at the start
71 5068 : // and end of the key range **before updating any nodes** in order
72 5068 : // to pin down the current coverage outside of the relevant key range.
73 5068 : // Only the coverage inside the layer's key range should change.
74 5068 : self.add_node(key.start);
75 5068 : self.add_node(key.end);
76 5068 :
77 5068 : // Raise the height where necessary
78 5068 : //
79 5068 : // NOTE This loop is worst case O(N), but amortized O(log N) in the special
80 5068 : // case when rectangles have no height. In practice I don't think we'll see
81 5068 : // the kind of layer intersections needed to trigger O(N) behavior. The worst
82 5068 : // case is N/2 horizontal layers overlapped with N/2 vertical layers in a
83 5068 : // grid pattern.
84 5068 : let mut to_update = Vec::new();
85 5068 : let mut to_remove = Vec::new();
86 5068 : let mut prev_covered = false;
87 6830 : for (k, node) in self.nodes.range(key) {
88 6830 : let needs_cover = match node {
89 1108 : None => true,
90 5722 : Some((h, _)) => h < &lsn.end,
91 : };
92 6830 : if needs_cover {
93 6816 : match prev_covered {
94 1742 : true => to_remove.push(*k),
95 5074 : false => to_update.push(*k),
96 : }
97 14 : }
98 6830 : prev_covered = needs_cover;
99 : }
100 : // TODO check if the nodes inserted at key.start and key.end are safe
101 : // to remove. It's fine to keep them but they could be redundant.
102 10142 : for k in to_update {
103 5074 : self.nodes.insert_mut(k, Some((lsn.end, value.clone())));
104 5074 : }
105 6810 : for k in to_remove {
106 1742 : self.nodes.remove_mut(&k);
107 1742 : }
108 5068 : }
109 :
110 : /// Get the latest (by lsn.end) layer at a given key
111 : ///
112 : /// Complexity: O(log N)
113 635152 : pub fn query(&self, key: i128) -> Option<Value> {
114 635152 : self.nodes
115 635152 : .range(..=key)
116 635152 : .next_back()?
117 : .1
118 541794 : .as_ref()
119 541794 : .map(|(_, v)| v.clone())
120 635152 : }
121 :
122 : /// Iterate the changes in layer coverage in a given range. You will likely
123 : /// want to start with self.query(key.start), and then follow up with self.range
124 : ///
125 : /// Complexity: O(log N + result_size)
126 491074 : pub fn range(&self, key: Range<i128>) -> impl '_ + Iterator<Item = (i128, Option<Value>)> {
127 491074 : self.nodes
128 491074 : .range(key)
129 491074 : .map(|(k, v)| (*k, v.as_ref().map(|x| x.1.clone())))
130 491074 : }
131 :
132 : /// Returns an iterator which includes all coverage changes for layers that intersect
133 : /// with the provided range.
134 491044 : pub fn range_overlaps(
135 491044 : &self,
136 491044 : key_range: &Range<i128>,
137 491044 : ) -> impl Iterator<Item = (i128, Option<Value>)> + '_
138 491044 : where
139 491044 : Value: Eq,
140 491044 : {
141 491044 : let first_change = self.query(key_range.start);
142 491044 : match first_change {
143 411894 : Some(change) => {
144 411894 : // If the start of the range is covered, we have to deal with two cases:
145 411894 : // 1. Start of the range is aligned with the start of a layer.
146 411894 : // In this case the return of `self.range` will contain the layer which aligns with the start of the key range.
147 411894 : // We advance said iterator to avoid duplicating the first change.
148 411894 : // 2. Start of the range is not aligned with the start of a layer.
149 411894 : let range = key_range.start..key_range.end;
150 411894 : let mut range_coverage = self.range(range).peekable();
151 411894 : if range_coverage
152 411894 : .peek()
153 411894 : .is_some_and(|c| c.1.as_ref() == Some(&change))
154 3786 : {
155 3786 : range_coverage.next();
156 408108 : }
157 411894 : itertools::Either::Left(
158 411894 : std::iter::once((key_range.start, Some(change))).chain(range_coverage),
159 411894 : )
160 : }
161 : None => {
162 79150 : let range = key_range.start..key_range.end;
163 79150 : let coverage = self.range(range);
164 79150 : itertools::Either::Right(coverage)
165 : }
166 : }
167 491044 : }
168 : /// O(1) clone
169 12096 : pub fn clone(&self) -> Self {
170 12096 : Self {
171 12096 : nodes: self.nodes.clone(),
172 12096 : }
173 12096 : }
174 : }
175 :
176 : /// Image and delta coverage at a specific LSN.
177 : pub struct LayerCoverageTuple<Value> {
178 : pub image_coverage: LayerCoverage<Value>,
179 : pub delta_coverage: LayerCoverage<Value>,
180 : }
181 :
182 : impl<T: Clone> Default for LayerCoverageTuple<T> {
183 826 : fn default() -> Self {
184 826 : Self {
185 826 : image_coverage: LayerCoverage::default(),
186 826 : delta_coverage: LayerCoverage::default(),
187 826 : }
188 826 : }
189 : }
190 :
191 : impl<Value: Clone> LayerCoverageTuple<Value> {
192 6048 : pub fn clone(&self) -> Self {
193 6048 : Self {
194 6048 : image_coverage: self.image_coverage.clone(),
195 6048 : delta_coverage: self.delta_coverage.clone(),
196 6048 : }
197 6048 : }
198 : }
|
