-
Notifications
You must be signed in to change notification settings - Fork 451
/
annotator.go
421 lines (377 loc) · 14.9 KB
/
annotator.go
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
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
// Copyright 2024 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package manifest
import (
"sort"
"github.com/cockroachdb/pebble/internal/base"
)
// The Annotator type defined below is used by other packages to lazily
// compute a value over a B-Tree. Each node of the B-Tree stores one
// `annotation` per annotator, containing the result of the computation over
// the node's subtree.
//
// An annotation is marked as valid if it's current with the current subtree
// state. Annotations are marked as invalid whenever a node will be mutated
// (in mut). Annotators may also return `false` from `Accumulate` to signal
// that a computation for a file is not stable and may change in the future.
// Annotations that include these unstable values are also marked as invalid
// on the node, ensuring that future queries for the annotation will recompute
// the value.
// An Annotator defines a computation over a level's FileMetadata. If the
// computation is stable and uses inputs that are fixed for the lifetime of
// a FileMetadata, the LevelMetadata's internal data structures are annotated
// with the intermediary computations. This allows the computation to be
// computed incrementally as edits are applied to a level.
type Annotator[T any] struct {
Aggregator AnnotationAggregator[T]
// scratch is used to hold the aggregated annotation value when computing
// range annotations in order to avoid additional allocations.
scratch *T
}
// An AnnotationAggregator defines how an annotation should be accumulated
// from a single FileMetadata and merged with other annotated values.
type AnnotationAggregator[T any] interface {
// Zero returns the zero value of an annotation. This value is returned
// when a LevelMetadata is empty. The dst argument, if non-nil, is an
// obsolete value previously returned by this Annotator and may be
// overwritten and reused to avoid a memory allocation.
Zero(dst *T) *T
// Accumulate computes the annotation for a single file in a level's
// metadata. It merges the file's value into dst and returns a bool flag
// indicating whether or not the value is stable and okay to cache as an
// annotation. If the file's value may change over the life of the file,
// the annotator must return false.
//
// Implementations may modify dst and return it to avoid an allocation.
Accumulate(f *FileMetadata, dst *T) (v *T, cacheOK bool)
// Merge combines two values src and dst, returning the result.
// Implementations may modify dst and return it to avoid an allocation.
Merge(src *T, dst *T) *T
}
// A PartialOverlapAnnotationAggregator is an extension of AnnotationAggregator
// that allows for custom accumulation of range annotations for files that only
// partially overlap with the range.
type PartialOverlapAnnotationAggregator[T any] interface {
AnnotationAggregator[T]
AccumulatePartialOverlap(f *FileMetadata, dst *T, bounds base.UserKeyBounds) *T
}
type annotation struct {
// annotator is a pointer to the Annotator that computed this annotation.
// NB: This is untyped to allow AnnotationAggregator to use Go generics,
// since annotations are stored in a slice on each node and a single
// slice cannot contain elements with different type parameters.
annotator interface{}
// v is contains the annotation value, the output of either
// AnnotationAggregator.Accumulate or AnnotationAggregator.Merge.
// NB: This is untyped for the same reason as annotator above.
v interface{}
// valid indicates whether future reads of the annotation may use the
// value as-is. If false, v will be zeroed and recalculated.
valid bool
}
// findAnnotation finds this Annotator's annotation on a node, creating
// one if it doesn't already exist.
func (a *Annotator[T]) findAnnotation(n *node) *annotation {
for i := range n.annot {
if n.annot[i].annotator == a {
return &n.annot[i]
}
}
// This node has never been annotated by a. Create a new annotation.
n.annot = append(n.annot, annotation{
annotator: a,
v: a.Aggregator.Zero(nil),
})
return &n.annot[len(n.annot)-1]
}
// nodeAnnotation computes this annotator's annotation of this node across all
// files in the node's subtree. The second return value indicates whether the
// annotation is stable and thus cacheable.
func (a *Annotator[T]) nodeAnnotation(n *node) (_ *T, cacheOK bool) {
annot := a.findAnnotation(n)
t := annot.v.(*T)
// If the annotation is already marked as valid, we can return it without
// recomputing anything.
if annot.valid {
return t, true
}
t = a.Aggregator.Zero(t)
valid := true
for i := int16(0); i <= n.count; i++ {
if !n.leaf {
v, ok := a.nodeAnnotation(n.children[i])
t = a.Aggregator.Merge(v, t)
valid = valid && ok
}
if i < n.count {
var ok bool
t, ok = a.Aggregator.Accumulate(n.items[i], t)
valid = valid && ok
}
}
annot.v = t
annot.valid = valid
return t, annot.valid
}
// accumulateRangeAnnotation computes this annotator's annotation across all
// files in the node's subtree which overlap with the range defined by bounds.
// The computed annotation is accumulated into a.scratch.
func (a *Annotator[T]) accumulateRangeAnnotation(
n *node,
cmp base.Compare,
bounds base.UserKeyBounds,
// fullyWithinLowerBound and fullyWithinUpperBound indicate whether this
// node's subtree is already known to be within each bound.
fullyWithinLowerBound bool,
fullyWithinUpperBound bool,
) {
// If this node's subtree is fully within the bounds, compute a regular
// annotation.
if fullyWithinLowerBound && fullyWithinUpperBound {
v, _ := a.nodeAnnotation(n)
a.scratch = a.Aggregator.Merge(v, a.scratch)
return
}
// We will accumulate annotations from each item in the end-exclusive
// range [leftItem, rightItem).
leftItem, rightItem := 0, int(n.count)
if !fullyWithinLowerBound {
// leftItem is the index of the first item that overlaps the lower bound.
leftItem = sort.Search(int(n.count), func(i int) bool {
return cmp(bounds.Start, n.items[i].Largest.UserKey) <= 0
})
}
if !fullyWithinUpperBound {
// rightItem is the index of the first item that does not overlap the
// upper bound.
rightItem = sort.Search(int(n.count), func(i int) bool {
return !bounds.End.IsUpperBoundFor(cmp, n.items[i].Smallest.UserKey)
})
}
// Accumulate annotations from every item that overlaps the bounds.
for i := leftItem; i < rightItem; i++ {
if i == leftItem || i == rightItem-1 {
if agg, ok := a.Aggregator.(PartialOverlapAnnotationAggregator[T]); ok {
fb := n.items[i].UserKeyBounds()
if cmp(bounds.Start, fb.Start) > 0 || bounds.End.CompareUpperBounds(cmp, fb.End) < 0 {
a.scratch = agg.AccumulatePartialOverlap(n.items[i], a.scratch, bounds)
continue
}
}
}
v, _ := a.Aggregator.Accumulate(n.items[i], a.scratch)
a.scratch = v
}
if !n.leaf {
// We will accumulate annotations from each child in the end-inclusive
// range [leftChild, rightChild].
leftChild, rightChild := leftItem, rightItem
// If the lower bound overlaps with the child at leftItem, there is no
// need to accumulate annotations from the child to its left.
if leftItem < int(n.count) && cmp(bounds.Start, n.items[leftItem].Smallest.UserKey) >= 0 {
leftChild++
}
// If the upper bound spans beyond the child at rightItem, we must also
// accumulate annotations from the child to its right.
if rightItem < int(n.count) && bounds.End.IsUpperBoundFor(cmp, n.items[rightItem].Largest.UserKey) {
rightChild++
}
for i := leftChild; i <= rightChild; i++ {
a.accumulateRangeAnnotation(
n.children[i],
cmp,
bounds,
// If this child is to the right of leftItem, then its entire
// subtree is within the lower bound.
fullyWithinLowerBound || i > leftItem,
// If this child is to the left of rightItem, then its entire
// subtree is within the upper bound.
fullyWithinUpperBound || i < rightItem,
)
}
}
}
// InvalidateAnnotation removes any existing cached annotations from this
// annotator from a node's subtree.
func (a *Annotator[T]) invalidateNodeAnnotation(n *node) {
annot := a.findAnnotation(n)
annot.valid = false
if !n.leaf {
for i := int16(0); i <= n.count; i++ {
a.invalidateNodeAnnotation(n.children[i])
}
}
}
// LevelAnnotation calculates the annotation defined by this Annotator for all
// files in the given LevelMetadata. A pointer to the Annotator is used as the
// key for pre-calculated values, so the same Annotator must be used to avoid
// duplicate computation. Annotation must not be called concurrently, and in
// practice this is achieved by requiring callers to hold DB.mu.
func (a *Annotator[T]) LevelAnnotation(lm LevelMetadata) *T {
if lm.Empty() {
return a.Aggregator.Zero(nil)
}
v, _ := a.nodeAnnotation(lm.tree.root)
return v
}
// MultiLevelAnnotation calculates the annotation defined by this Annotator for
// all files across the given levels. A pointer to the Annotator is used as the
// key for pre-calculated values, so the same Annotator must be used to avoid
// duplicate computation. Annotation must not be called concurrently, and in
// practice this is achieved by requiring callers to hold DB.mu.
func (a *Annotator[T]) MultiLevelAnnotation(lms []LevelMetadata) *T {
aggregated := a.Aggregator.Zero(nil)
for l := 0; l < len(lms); l++ {
if !lms[l].Empty() {
v := a.LevelAnnotation(lms[l])
aggregated = a.Aggregator.Merge(v, aggregated)
}
}
return aggregated
}
// LevelRangeAnnotation calculates the annotation defined by this Annotator for
// the files within LevelMetadata which are within the range
// [lowerBound, upperBound). A pointer to the Annotator is used as the key for
// pre-calculated values, so the same Annotator must be used to avoid duplicate
// computation. Annotation must not be called concurrently, and in practice this
// is achieved by requiring callers to hold DB.mu.
func (a *Annotator[T]) LevelRangeAnnotation(lm LevelMetadata, bounds base.UserKeyBounds) *T {
if lm.Empty() {
return a.Aggregator.Zero(nil)
}
a.scratch = a.Aggregator.Zero(a.scratch)
a.accumulateRangeAnnotation(lm.tree.root, lm.tree.cmp, bounds, false, false)
return a.scratch
}
// VersionRangeAnnotation calculates the annotation defined by this Annotator
// for all files within the given Version which are within the range
// defined by bounds.
func (a *Annotator[T]) VersionRangeAnnotation(v *Version, bounds base.UserKeyBounds) *T {
accumulateSlice := func(ls LevelSlice) {
if ls.Empty() {
return
}
a.accumulateRangeAnnotation(ls.iter.r, v.cmp.Compare, bounds, false, false)
}
a.scratch = a.Aggregator.Zero(a.scratch)
for _, ls := range v.L0SublevelFiles {
accumulateSlice(ls)
}
for _, lm := range v.Levels[1:] {
accumulateSlice(lm.Slice())
}
return a.scratch
}
// InvalidateAnnotation clears any cached annotations defined by Annotator. A
// pointer to the Annotator is used as the key for pre-calculated values, so
// the same Annotator must be used to clear the appropriate cached annotation.
// InvalidateAnnotation must not be called concurrently, and in practice this
// is achieved by requiring callers to hold DB.mu.
func (a *Annotator[T]) InvalidateLevelAnnotation(lm LevelMetadata) {
if lm.Empty() {
return
}
a.invalidateNodeAnnotation(lm.tree.root)
}
// SumAggregator defines an Aggregator which sums together a uint64 value
// across files.
type SumAggregator struct {
AccumulateFunc func(f *FileMetadata) (v uint64, cacheOK bool)
AccumulatePartialOverlapFunc func(f *FileMetadata, bounds base.UserKeyBounds) uint64
}
// Zero implements AnnotationAggregator.Zero, returning a new uint64 set to 0.
func (sa SumAggregator) Zero(dst *uint64) *uint64 {
if dst == nil {
return new(uint64)
}
*dst = 0
return dst
}
// Accumulate implements AnnotationAggregator.Accumulate, accumulating a single
// file's uint64 value.
func (sa SumAggregator) Accumulate(f *FileMetadata, dst *uint64) (v *uint64, cacheOK bool) {
accumulated, ok := sa.AccumulateFunc(f)
*dst += accumulated
return dst, ok
}
// AccumulatePartialOverlap implements
// PartialOverlapAnnotationAggregator.AccumulatePartialOverlap, accumulating a
// single file's uint64 value for a file which only partially overlaps with the
// range defined by bounds.
func (sa SumAggregator) AccumulatePartialOverlap(
f *FileMetadata, dst *uint64, bounds base.UserKeyBounds,
) *uint64 {
if sa.AccumulatePartialOverlapFunc == nil {
v, _ := sa.Accumulate(f, dst)
return v
}
*dst += sa.AccumulatePartialOverlapFunc(f, bounds)
return dst
}
// Merge implements AnnotationAggregator.Merge by summing two uint64 values.
func (sa SumAggregator) Merge(src *uint64, dst *uint64) *uint64 {
*dst += *src
return dst
}
// SumAnnotator takes a function that computes a uint64 value from a single
// FileMetadata and returns an Annotator that sums together the values across
// files.
func SumAnnotator(accumulate func(f *FileMetadata) (v uint64, cacheOK bool)) *Annotator[uint64] {
return &Annotator[uint64]{
Aggregator: SumAggregator{
AccumulateFunc: accumulate,
},
}
}
// NumFilesAnnotator is an Annotator which computes an annotation value
// equal to the number of files included in the annotation. Particularly, it
// can be used to efficiently calculate the number of files in a given key
// range using range annotations.
var NumFilesAnnotator = SumAnnotator(func(f *FileMetadata) (uint64, bool) {
return 1, true
})
// PickFileAggregator implements the AnnotationAggregator interface. It defines
// an aggregator that picks a single file from a set of eligible files.
type PickFileAggregator struct {
// Filter takes a FileMetadata and returns whether it is eligible to be
// picked by this PickFileAggregator. The second return value indicates
// whether this eligibility is stable and thus cacheable.
Filter func(f *FileMetadata) (eligible bool, cacheOK bool)
// Compare compares two instances of FileMetadata and returns true if
// the first one should be picked over the second one. It may assume
// that both arguments are non-nil.
Compare func(f1 *FileMetadata, f2 *FileMetadata) bool
}
// Zero implements AnnotationAggregator.Zero, returning nil as the zero value.
func (fa PickFileAggregator) Zero(dst *FileMetadata) *FileMetadata {
return nil
}
func (fa PickFileAggregator) mergePickedFiles(src *FileMetadata, dst *FileMetadata) *FileMetadata {
switch {
case src == nil:
return dst
case dst == nil:
return src
case fa.Compare(src, dst):
return src
default:
return dst
}
}
// Accumulate implements AnnotationAggregator.Accumulate, accumulating a single
// file as long as it is eligible to be picked.
func (fa PickFileAggregator) Accumulate(
f *FileMetadata, dst *FileMetadata,
) (v *FileMetadata, cacheOK bool) {
eligible, ok := fa.Filter(f)
if eligible {
return fa.mergePickedFiles(f, dst), ok
}
return dst, ok
}
// Merge implements AnnotationAggregator.Merge by picking a single file based
// on the output of PickFileAggregator.Compare.
func (fa PickFileAggregator) Merge(src *FileMetadata, dst *FileMetadata) *FileMetadata {
return fa.mergePickedFiles(src, dst)
}