Line data Source code
1 : // Copyright 2018 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "bytes"
9 : "context"
10 : "fmt"
11 : "math"
12 : "sort"
13 : "strings"
14 :
15 : "github.com/cockroachdb/errors"
16 : "github.com/cockroachdb/pebble/internal/base"
17 : "github.com/cockroachdb/pebble/internal/humanize"
18 : "github.com/cockroachdb/pebble/internal/invariants"
19 : "github.com/cockroachdb/pebble/internal/manifest"
20 : )
21 :
22 : // The minimum count for an intra-L0 compaction. This matches the RocksDB
23 : // heuristic.
24 : const minIntraL0Count = 4
25 :
26 : type compactionEnv struct {
27 : // diskAvailBytes holds a statistic on the number of bytes available on
28 : // disk, as reported by the filesystem. It's used to be more restrictive in
29 : // expanding compactions if available disk space is limited.
30 : //
31 : // The cached value (d.diskAvailBytes) is updated whenever a file is deleted
32 : // and whenever a compaction or flush completes. Since file removal is the
33 : // primary means of reclaiming space, there is a rough bound on the
34 : // statistic's staleness when available bytes is growing. Compactions and
35 : // flushes are longer, slower operations and provide a much looser bound
36 : // when available bytes is decreasing.
37 : diskAvailBytes uint64
38 : earliestUnflushedSeqNum base.SeqNum
39 : earliestSnapshotSeqNum base.SeqNum
40 : inProgressCompactions []compactionInfo
41 : readCompactionEnv readCompactionEnv
42 : }
43 :
44 : type compactionPicker interface {
45 : getScores([]compactionInfo) [numLevels]float64
46 : getBaseLevel() int
47 : estimatedCompactionDebt(l0ExtraSize uint64) uint64
48 : pickAuto(env compactionEnv) (pc *pickedCompaction)
49 : pickElisionOnlyCompaction(env compactionEnv) (pc *pickedCompaction)
50 : pickRewriteCompaction(env compactionEnv) (pc *pickedCompaction)
51 : pickReadTriggeredCompaction(env compactionEnv) (pc *pickedCompaction)
52 : forceBaseLevel1()
53 : }
54 :
55 : // readCompactionEnv is used to hold data required to perform read compactions
56 : type readCompactionEnv struct {
57 : rescheduleReadCompaction *bool
58 : readCompactions *readCompactionQueue
59 : flushing bool
60 : }
61 :
62 : // Information about in-progress compactions provided to the compaction picker.
63 : // These are used to constrain the new compactions that will be picked.
64 : type compactionInfo struct {
65 : // versionEditApplied is true if this compaction's version edit has already
66 : // been committed. The compaction may still be in-progress deleting newly
67 : // obsolete files.
68 : versionEditApplied bool
69 : inputs []compactionLevel
70 : outputLevel int
71 : smallest InternalKey
72 : largest InternalKey
73 : }
74 :
75 1 : func (info compactionInfo) String() string {
76 1 : var buf bytes.Buffer
77 1 : var largest int
78 1 : for i, in := range info.inputs {
79 1 : if i > 0 {
80 1 : fmt.Fprintf(&buf, " -> ")
81 1 : }
82 1 : fmt.Fprintf(&buf, "L%d", in.level)
83 1 : in.files.Each(func(m *fileMetadata) {
84 1 : fmt.Fprintf(&buf, " %s", m.FileNum)
85 1 : })
86 1 : if largest < in.level {
87 1 : largest = in.level
88 1 : }
89 : }
90 1 : if largest != info.outputLevel || len(info.inputs) == 1 {
91 1 : fmt.Fprintf(&buf, " -> L%d", info.outputLevel)
92 1 : }
93 1 : return buf.String()
94 : }
95 :
96 : type sortCompactionLevelsByPriority []candidateLevelInfo
97 :
98 1 : func (s sortCompactionLevelsByPriority) Len() int {
99 1 : return len(s)
100 1 : }
101 :
102 : // A level should be picked for compaction if the compensatedScoreRatio is >= the
103 : // compactionScoreThreshold.
104 : const compactionScoreThreshold = 1
105 :
106 : // Less should return true if s[i] must be placed earlier than s[j] in the final
107 : // sorted list. The candidateLevelInfo for the level placed earlier is more likely
108 : // to be picked for a compaction.
109 1 : func (s sortCompactionLevelsByPriority) Less(i, j int) bool {
110 1 : iShouldCompact := s[i].compensatedScoreRatio >= compactionScoreThreshold
111 1 : jShouldCompact := s[j].compensatedScoreRatio >= compactionScoreThreshold
112 1 : // Ordering is defined as decreasing on (shouldCompact, uncompensatedScoreRatio)
113 1 : // where shouldCompact is 1 for true and 0 for false.
114 1 : if iShouldCompact && !jShouldCompact {
115 1 : return true
116 1 : }
117 1 : if !iShouldCompact && jShouldCompact {
118 1 : return false
119 1 : }
120 :
121 1 : if s[i].uncompensatedScoreRatio != s[j].uncompensatedScoreRatio {
122 1 : return s[i].uncompensatedScoreRatio > s[j].uncompensatedScoreRatio
123 1 : }
124 1 : return s[i].level < s[j].level
125 : }
126 :
127 1 : func (s sortCompactionLevelsByPriority) Swap(i, j int) {
128 1 : s[i], s[j] = s[j], s[i]
129 1 : }
130 :
131 : // sublevelInfo is used to tag a LevelSlice for an L0 sublevel with the
132 : // sublevel.
133 : type sublevelInfo struct {
134 : manifest.LevelSlice
135 : sublevel manifest.Layer
136 : }
137 :
138 1 : func (cl sublevelInfo) Clone() sublevelInfo {
139 1 : return sublevelInfo{
140 1 : sublevel: cl.sublevel,
141 1 : LevelSlice: cl.LevelSlice,
142 1 : }
143 1 : }
144 1 : func (cl sublevelInfo) String() string {
145 1 : return fmt.Sprintf(`Sublevel %s; Levels %s`, cl.sublevel, cl.LevelSlice)
146 1 : }
147 :
148 : // generateSublevelInfo will generate the level slices for each of the sublevels
149 : // from the level slice for all of L0.
150 1 : func generateSublevelInfo(cmp base.Compare, levelFiles manifest.LevelSlice) []sublevelInfo {
151 1 : sublevelMap := make(map[uint64][]*fileMetadata)
152 1 : it := levelFiles.Iter()
153 1 : for f := it.First(); f != nil; f = it.Next() {
154 1 : sublevelMap[uint64(f.SubLevel)] = append(sublevelMap[uint64(f.SubLevel)], f)
155 1 : }
156 :
157 1 : var sublevels []int
158 1 : for level := range sublevelMap {
159 1 : sublevels = append(sublevels, int(level))
160 1 : }
161 1 : sort.Ints(sublevels)
162 1 :
163 1 : var levelSlices []sublevelInfo
164 1 : for _, sublevel := range sublevels {
165 1 : metas := sublevelMap[uint64(sublevel)]
166 1 : levelSlices = append(
167 1 : levelSlices,
168 1 : sublevelInfo{
169 1 : manifest.NewLevelSliceKeySorted(cmp, metas),
170 1 : manifest.L0Sublevel(sublevel),
171 1 : },
172 1 : )
173 1 : }
174 1 : return levelSlices
175 : }
176 :
177 : // compactionPickerMetrics holds metrics related to the compaction picking process
178 : type compactionPickerMetrics struct {
179 : // scores contains the compensatedScoreRatio from the candidateLevelInfo.
180 : scores []float64
181 : singleLevelOverlappingRatio float64
182 : multiLevelOverlappingRatio float64
183 : }
184 :
185 : // pickedCompaction contains information about a compaction that has already
186 : // been chosen, and is being constructed. Compaction construction info lives in
187 : // this struct, and is copied over into the compaction struct when that's
188 : // created.
189 : type pickedCompaction struct {
190 : cmp Compare
191 : // score of the chosen compaction. This is the same as the
192 : // compensatedScoreRatio in the candidateLevelInfo.
193 : score float64
194 : // kind indicates the kind of compaction.
195 : kind compactionKind
196 : // startLevel is the level that is being compacted. Inputs from startLevel
197 : // and outputLevel will be merged to produce a set of outputLevel files.
198 : startLevel *compactionLevel
199 : // outputLevel is the level that files are being produced in. outputLevel is
200 : // equal to startLevel+1 except when:
201 : // - if startLevel is 0, the output level equals compactionPicker.baseLevel().
202 : // - in multilevel compaction, the output level is the lowest level involved in
203 : // the compaction
204 : outputLevel *compactionLevel
205 : // extraLevels contain additional levels in between the input and output
206 : // levels that get compacted in multi level compactions
207 : extraLevels []*compactionLevel
208 : inputs []compactionLevel
209 : // LBase at the time of compaction picking.
210 : baseLevel int
211 : // L0-specific compaction info. Set to a non-nil value for all compactions
212 : // where startLevel == 0 that were generated by L0Sublevels.
213 : lcf *manifest.L0CompactionFiles
214 : // maxOutputFileSize is the maximum size of an individual table created
215 : // during compaction.
216 : maxOutputFileSize uint64
217 : // maxOverlapBytes is the maximum number of bytes of overlap allowed for a
218 : // single output table with the tables in the grandparent level.
219 : maxOverlapBytes uint64
220 : // maxReadCompactionBytes is the maximum bytes a read compaction is allowed to
221 : // overlap in its output level with. If the overlap is greater than
222 : // maxReadCompaction bytes, then we don't proceed with the compaction.
223 : maxReadCompactionBytes uint64
224 :
225 : // slot is the compaction slot used up by this compaction. Encapsulates any
226 : // limiting/pacing logic.
227 : slot base.CompactionSlot
228 : // The boundaries of the input data.
229 : smallest InternalKey
230 : largest InternalKey
231 : version *version
232 : pickerMetrics compactionPickerMetrics
233 : }
234 :
235 1 : func (pc *pickedCompaction) userKeyBounds() base.UserKeyBounds {
236 1 : return base.UserKeyBoundsFromInternal(pc.smallest, pc.largest)
237 1 : }
238 :
239 1 : func defaultOutputLevel(startLevel, baseLevel int) int {
240 1 : outputLevel := startLevel + 1
241 1 : if startLevel == 0 {
242 1 : outputLevel = baseLevel
243 1 : }
244 1 : if outputLevel >= numLevels-1 {
245 1 : outputLevel = numLevels - 1
246 1 : }
247 1 : return outputLevel
248 : }
249 :
250 : func newPickedCompaction(
251 : opts *Options, cur *version, startLevel, outputLevel, baseLevel int,
252 1 : ) *pickedCompaction {
253 1 : if startLevel > 0 && startLevel < baseLevel {
254 1 : panic(fmt.Sprintf("invalid compaction: start level %d should not be empty (base level %d)",
255 1 : startLevel, baseLevel))
256 : }
257 :
258 1 : adjustedLevel := adjustedOutputLevel(outputLevel, baseLevel)
259 1 : pc := &pickedCompaction{
260 1 : cmp: opts.Comparer.Compare,
261 1 : version: cur,
262 1 : baseLevel: baseLevel,
263 1 : inputs: []compactionLevel{{level: startLevel}, {level: outputLevel}},
264 1 : maxOutputFileSize: uint64(opts.Level(adjustedLevel).TargetFileSize),
265 1 : maxOverlapBytes: maxGrandparentOverlapBytes(opts, adjustedLevel),
266 1 : maxReadCompactionBytes: maxReadCompactionBytes(opts, adjustedLevel),
267 1 : }
268 1 : pc.startLevel = &pc.inputs[0]
269 1 : pc.outputLevel = &pc.inputs[1]
270 1 : return pc
271 : }
272 :
273 : // adjustedOutputLevel is the output level used for the purpose of
274 : // determining the target output file size, overlap bytes, and expanded
275 : // bytes, taking into account the base level.
276 1 : func adjustedOutputLevel(outputLevel int, baseLevel int) int {
277 1 : adjustedOutputLevel := outputLevel
278 1 : if adjustedOutputLevel > 0 {
279 1 : // Output level is in the range [baseLevel, numLevels]. For the purpose of
280 1 : // determining the target output file size, overlap bytes, and expanded
281 1 : // bytes, we want to adjust the range to [1,numLevels].
282 1 : adjustedOutputLevel = 1 + outputLevel - baseLevel
283 1 : }
284 1 : return adjustedOutputLevel
285 : }
286 :
287 : func newPickedCompactionFromL0(
288 : lcf *manifest.L0CompactionFiles, opts *Options, vers *version, baseLevel int, isBase bool,
289 1 : ) *pickedCompaction {
290 1 : outputLevel := baseLevel
291 1 : if !isBase {
292 1 : outputLevel = 0 // Intra L0
293 1 : }
294 :
295 1 : pc := newPickedCompaction(opts, vers, 0, outputLevel, baseLevel)
296 1 : pc.lcf = lcf
297 1 : pc.outputLevel.level = outputLevel
298 1 :
299 1 : // Manually build the compaction as opposed to calling
300 1 : // pickAutoHelper. This is because L0Sublevels has already added
301 1 : // any overlapping L0 SSTables that need to be added, and
302 1 : // because compactions built by L0SSTables do not necessarily
303 1 : // pick contiguous sequences of files in pc.version.Levels[0].
304 1 : files := make([]*manifest.FileMetadata, 0, len(lcf.Files))
305 1 : iter := vers.Levels[0].Iter()
306 1 : for f := iter.First(); f != nil; f = iter.Next() {
307 1 : if lcf.FilesIncluded[f.L0Index] {
308 1 : files = append(files, f)
309 1 : }
310 : }
311 1 : pc.startLevel.files = manifest.NewLevelSliceSeqSorted(files)
312 1 : return pc
313 : }
314 :
315 1 : func (pc *pickedCompaction) String() string {
316 1 : var builder strings.Builder
317 1 : builder.WriteString(fmt.Sprintf(`Score=%f, `, pc.score))
318 1 : builder.WriteString(fmt.Sprintf(`Kind=%s, `, pc.kind))
319 1 : builder.WriteString(fmt.Sprintf(`AdjustedOutputLevel=%d, `, adjustedOutputLevel(pc.outputLevel.level, pc.baseLevel)))
320 1 : builder.WriteString(fmt.Sprintf(`maxOutputFileSize=%d, `, pc.maxOutputFileSize))
321 1 : builder.WriteString(fmt.Sprintf(`maxReadCompactionBytes=%d, `, pc.maxReadCompactionBytes))
322 1 : builder.WriteString(fmt.Sprintf(`smallest=%s, `, pc.smallest))
323 1 : builder.WriteString(fmt.Sprintf(`largest=%s, `, pc.largest))
324 1 : builder.WriteString(fmt.Sprintf(`version=%s, `, pc.version))
325 1 : builder.WriteString(fmt.Sprintf(`inputs=%s, `, pc.inputs))
326 1 : builder.WriteString(fmt.Sprintf(`startlevel=%s, `, pc.startLevel))
327 1 : builder.WriteString(fmt.Sprintf(`outputLevel=%s, `, pc.outputLevel))
328 1 : builder.WriteString(fmt.Sprintf(`extraLevels=%s, `, pc.extraLevels))
329 1 : builder.WriteString(fmt.Sprintf(`l0SublevelInfo=%s, `, pc.startLevel.l0SublevelInfo))
330 1 : builder.WriteString(fmt.Sprintf(`lcf=%s`, pc.lcf))
331 1 : return builder.String()
332 1 : }
333 :
334 : // Clone creates a deep copy of the pickedCompaction
335 1 : func (pc *pickedCompaction) clone() *pickedCompaction {
336 1 :
337 1 : // Quickly copy over fields that do not require special deep copy care, and
338 1 : // set all fields that will require a deep copy to nil.
339 1 : newPC := &pickedCompaction{
340 1 : cmp: pc.cmp,
341 1 : score: pc.score,
342 1 : kind: pc.kind,
343 1 : baseLevel: pc.baseLevel,
344 1 : maxOutputFileSize: pc.maxOutputFileSize,
345 1 : maxOverlapBytes: pc.maxOverlapBytes,
346 1 : maxReadCompactionBytes: pc.maxReadCompactionBytes,
347 1 : smallest: pc.smallest.Clone(),
348 1 : largest: pc.largest.Clone(),
349 1 :
350 1 : // TODO(msbutler): properly clone picker metrics
351 1 : pickerMetrics: pc.pickerMetrics,
352 1 :
353 1 : // Both copies see the same manifest, therefore, it's ok for them to se
354 1 : // share the same pc. version.
355 1 : version: pc.version,
356 1 : }
357 1 :
358 1 : newPC.inputs = make([]compactionLevel, len(pc.inputs))
359 1 : newPC.extraLevels = make([]*compactionLevel, 0, len(pc.extraLevels))
360 1 : for i := range pc.inputs {
361 1 : newPC.inputs[i] = pc.inputs[i].Clone()
362 1 : if i == 0 {
363 1 : newPC.startLevel = &newPC.inputs[i]
364 1 : } else if i == len(pc.inputs)-1 {
365 1 : newPC.outputLevel = &newPC.inputs[i]
366 1 : } else {
367 1 : newPC.extraLevels = append(newPC.extraLevels, &newPC.inputs[i])
368 1 : }
369 : }
370 :
371 1 : if len(pc.startLevel.l0SublevelInfo) > 0 {
372 1 : newPC.startLevel.l0SublevelInfo = make([]sublevelInfo, len(pc.startLevel.l0SublevelInfo))
373 1 : for i := range pc.startLevel.l0SublevelInfo {
374 1 : newPC.startLevel.l0SublevelInfo[i] = pc.startLevel.l0SublevelInfo[i].Clone()
375 1 : }
376 : }
377 1 : if pc.lcf != nil {
378 1 : newPC.lcf = pc.lcf.Clone()
379 1 : }
380 1 : return newPC
381 : }
382 :
383 : // maybeExpandBounds is a helper function for setupInputs which ensures the
384 : // pickedCompaction's smallest and largest internal keys are updated iff
385 : // the candidate keys expand the key span. This avoids a bug for multi-level
386 : // compactions: during the second call to setupInputs, the picked compaction's
387 : // smallest and largest keys should not decrease the key span.
388 1 : func (pc *pickedCompaction) maybeExpandBounds(smallest InternalKey, largest InternalKey) {
389 1 : if len(smallest.UserKey) == 0 && len(largest.UserKey) == 0 {
390 1 : return
391 1 : }
392 1 : if len(pc.smallest.UserKey) == 0 && len(pc.largest.UserKey) == 0 {
393 1 : pc.smallest = smallest
394 1 : pc.largest = largest
395 1 : return
396 1 : }
397 1 : if base.InternalCompare(pc.cmp, pc.smallest, smallest) >= 0 {
398 1 : pc.smallest = smallest
399 1 : }
400 1 : if base.InternalCompare(pc.cmp, pc.largest, largest) <= 0 {
401 1 : pc.largest = largest
402 1 : }
403 : }
404 :
405 : // setupInputs returns true if a compaction has been set up. It returns false if
406 : // a concurrent compaction is occurring on the start or output level files.
407 : func (pc *pickedCompaction) setupInputs(
408 : opts *Options, diskAvailBytes uint64, startLevel *compactionLevel,
409 1 : ) bool {
410 1 : // maxExpandedBytes is the maximum size of an expanded compaction. If
411 1 : // growing a compaction results in a larger size, the original compaction
412 1 : // is used instead.
413 1 : maxExpandedBytes := expandedCompactionByteSizeLimit(
414 1 : opts, adjustedOutputLevel(pc.outputLevel.level, pc.baseLevel), diskAvailBytes,
415 1 : )
416 1 :
417 1 : if anyTablesCompacting(startLevel.files) {
418 1 : return false
419 1 : }
420 :
421 1 : pc.maybeExpandBounds(manifest.KeyRange(pc.cmp, startLevel.files.Iter()))
422 1 :
423 1 : // Determine the sstables in the output level which overlap with the input
424 1 : // sstables. No need to do this for intra-L0 compactions; outputLevel.files is
425 1 : // left empty for those.
426 1 : if startLevel.level != pc.outputLevel.level {
427 1 : pc.outputLevel.files = pc.version.Overlaps(pc.outputLevel.level, pc.userKeyBounds())
428 1 : if anyTablesCompacting(pc.outputLevel.files) {
429 1 : return false
430 1 : }
431 :
432 1 : pc.maybeExpandBounds(manifest.KeyRange(pc.cmp,
433 1 : startLevel.files.Iter(), pc.outputLevel.files.Iter()))
434 : }
435 :
436 : // Grow the sstables in startLevel.level as long as it doesn't affect the number
437 : // of sstables included from pc.outputLevel.level.
438 1 : if pc.lcf != nil && startLevel.level == 0 && pc.outputLevel.level != 0 {
439 1 : // Call the L0-specific compaction extension method. Similar logic as
440 1 : // pc.grow. Additional L0 files are optionally added to the compaction at
441 1 : // this step. Note that the bounds passed in are not the bounds of the
442 1 : // compaction, but rather the smallest and largest internal keys that
443 1 : // the compaction cannot include from L0 without pulling in more Lbase
444 1 : // files. Consider this example:
445 1 : //
446 1 : // L0: c-d e+f g-h
447 1 : // Lbase: a-b e+f i-j
448 1 : // a b c d e f g h i j
449 1 : //
450 1 : // The e-f files have already been chosen in the compaction. As pulling
451 1 : // in more LBase files is undesirable, the logic below will pass in
452 1 : // smallest = b and largest = i to ExtendL0ForBaseCompactionTo, which
453 1 : // will expand the compaction to include c-d and g-h from L0. The
454 1 : // bounds passed in are exclusive; the compaction cannot be expanded
455 1 : // to include files that "touch" it.
456 1 : smallestBaseKey := base.InvalidInternalKey
457 1 : largestBaseKey := base.InvalidInternalKey
458 1 : if pc.outputLevel.files.Empty() {
459 1 : baseIter := pc.version.Levels[pc.outputLevel.level].Iter()
460 1 : if sm := baseIter.SeekLT(pc.cmp, pc.smallest.UserKey); sm != nil {
461 1 : smallestBaseKey = sm.Largest
462 1 : }
463 1 : if la := baseIter.SeekGE(pc.cmp, pc.largest.UserKey); la != nil {
464 1 : largestBaseKey = la.Smallest
465 1 : }
466 1 : } else {
467 1 : // NB: We use Reslice to access the underlying level's files, but
468 1 : // we discard the returned slice. The pc.outputLevel.files slice
469 1 : // is not modified.
470 1 : _ = pc.outputLevel.files.Reslice(func(start, end *manifest.LevelIterator) {
471 1 : if sm := start.Prev(); sm != nil {
472 1 : smallestBaseKey = sm.Largest
473 1 : }
474 1 : if la := end.Next(); la != nil {
475 1 : largestBaseKey = la.Smallest
476 1 : }
477 : })
478 : }
479 1 : oldLcf := pc.lcf.Clone()
480 1 : if pc.version.L0Sublevels.ExtendL0ForBaseCompactionTo(smallestBaseKey, largestBaseKey, pc.lcf) {
481 1 : var newStartLevelFiles []*fileMetadata
482 1 : iter := pc.version.Levels[0].Iter()
483 1 : var sizeSum uint64
484 1 : for j, f := 0, iter.First(); f != nil; j, f = j+1, iter.Next() {
485 1 : if pc.lcf.FilesIncluded[f.L0Index] {
486 1 : newStartLevelFiles = append(newStartLevelFiles, f)
487 1 : sizeSum += f.Size
488 1 : }
489 : }
490 1 : if sizeSum+pc.outputLevel.files.SizeSum() < maxExpandedBytes {
491 1 : startLevel.files = manifest.NewLevelSliceSeqSorted(newStartLevelFiles)
492 1 : pc.smallest, pc.largest = manifest.KeyRange(pc.cmp,
493 1 : startLevel.files.Iter(), pc.outputLevel.files.Iter())
494 1 : } else {
495 1 : *pc.lcf = *oldLcf
496 1 : }
497 : }
498 1 : } else if pc.grow(pc.smallest, pc.largest, maxExpandedBytes, startLevel) {
499 1 : pc.maybeExpandBounds(manifest.KeyRange(pc.cmp,
500 1 : startLevel.files.Iter(), pc.outputLevel.files.Iter()))
501 1 : }
502 :
503 1 : if pc.startLevel.level == 0 {
504 1 : // We don't change the input files for the compaction beyond this point.
505 1 : pc.startLevel.l0SublevelInfo = generateSublevelInfo(pc.cmp, pc.startLevel.files)
506 1 : }
507 :
508 1 : return true
509 : }
510 :
511 : // grow grows the number of inputs at c.level without changing the number of
512 : // c.level+1 files in the compaction, and returns whether the inputs grew. sm
513 : // and la are the smallest and largest InternalKeys in all of the inputs.
514 : func (pc *pickedCompaction) grow(
515 : sm, la InternalKey, maxExpandedBytes uint64, startLevel *compactionLevel,
516 1 : ) bool {
517 1 : if pc.outputLevel.files.Empty() {
518 1 : return false
519 1 : }
520 1 : grow0 := pc.version.Overlaps(startLevel.level, base.UserKeyBoundsFromInternal(sm, la))
521 1 : if anyTablesCompacting(grow0) {
522 0 : return false
523 0 : }
524 1 : if grow0.Len() <= startLevel.files.Len() {
525 1 : return false
526 1 : }
527 1 : if grow0.SizeSum()+pc.outputLevel.files.SizeSum() >= maxExpandedBytes {
528 1 : return false
529 1 : }
530 : // We need to include the outputLevel iter because without it, in a multiLevel scenario,
531 : // sm1 and la1 could shift the output level keyspace when pc.outputLevel.files is set to grow1.
532 1 : sm1, la1 := manifest.KeyRange(pc.cmp, grow0.Iter(), pc.outputLevel.files.Iter())
533 1 : grow1 := pc.version.Overlaps(pc.outputLevel.level, base.UserKeyBoundsFromInternal(sm1, la1))
534 1 : if anyTablesCompacting(grow1) {
535 0 : return false
536 0 : }
537 1 : if grow1.Len() != pc.outputLevel.files.Len() {
538 1 : return false
539 1 : }
540 1 : startLevel.files = grow0
541 1 : pc.outputLevel.files = grow1
542 1 : return true
543 : }
544 :
545 1 : func (pc *pickedCompaction) compactionSize() uint64 {
546 1 : var bytesToCompact uint64
547 1 : for i := range pc.inputs {
548 1 : bytesToCompact += pc.inputs[i].files.SizeSum()
549 1 : }
550 1 : return bytesToCompact
551 : }
552 :
553 : // setupMultiLevelCandidated returns true if it successfully added another level
554 : // to the compaction.
555 1 : func (pc *pickedCompaction) setupMultiLevelCandidate(opts *Options, diskAvailBytes uint64) bool {
556 1 : pc.inputs = append(pc.inputs, compactionLevel{level: pc.outputLevel.level + 1})
557 1 :
558 1 : // Recalibrate startLevel and outputLevel:
559 1 : // - startLevel and outputLevel pointers may be obsolete after appending to pc.inputs.
560 1 : // - push outputLevel to extraLevels and move the new level to outputLevel
561 1 : pc.startLevel = &pc.inputs[0]
562 1 : pc.extraLevels = []*compactionLevel{&pc.inputs[1]}
563 1 : pc.outputLevel = &pc.inputs[2]
564 1 : return pc.setupInputs(opts, diskAvailBytes, pc.extraLevels[len(pc.extraLevels)-1])
565 1 : }
566 :
567 : // anyTablesCompacting returns true if any tables in the level slice are
568 : // compacting.
569 1 : func anyTablesCompacting(inputs manifest.LevelSlice) bool {
570 1 : it := inputs.Iter()
571 1 : for f := it.First(); f != nil; f = it.Next() {
572 1 : if f.IsCompacting() {
573 1 : return true
574 1 : }
575 : }
576 1 : return false
577 : }
578 :
579 : // newCompactionPickerByScore creates a compactionPickerByScore associated with
580 : // the newest version. The picker is used under logLock (until a new version is
581 : // installed).
582 : func newCompactionPickerByScore(
583 : v *version,
584 : virtualBackings *manifest.VirtualBackings,
585 : opts *Options,
586 : inProgressCompactions []compactionInfo,
587 1 : ) *compactionPickerByScore {
588 1 : p := &compactionPickerByScore{
589 1 : opts: opts,
590 1 : vers: v,
591 1 : virtualBackings: virtualBackings,
592 1 : }
593 1 : p.initLevelMaxBytes(inProgressCompactions)
594 1 : return p
595 1 : }
596 :
597 : // Information about a candidate compaction level that has been identified by
598 : // the compaction picker.
599 : type candidateLevelInfo struct {
600 : // The compensatedScore of the level after adjusting according to the other
601 : // levels' sizes. For L0, the compensatedScoreRatio is equivalent to the
602 : // uncompensatedScoreRatio as we don't account for level size compensation in
603 : // L0.
604 : compensatedScoreRatio float64
605 : // The score of the level after accounting for level size compensation before
606 : // adjusting according to other levels' sizes. For L0, the compensatedScore
607 : // is equivalent to the uncompensatedScore as we don't account for level
608 : // size compensation in L0.
609 : compensatedScore float64
610 : // The score of the level to be compacted, calculated using uncompensated file
611 : // sizes and without any adjustments.
612 : uncompensatedScore float64
613 : // uncompensatedScoreRatio is the uncompensatedScore adjusted according to
614 : // the other levels' sizes.
615 : uncompensatedScoreRatio float64
616 : level int
617 : // The level to compact to.
618 : outputLevel int
619 : // The file in level that will be compacted. Additional files may be
620 : // picked by the compaction, and a pickedCompaction created for the
621 : // compaction.
622 : file manifest.LevelFile
623 : }
624 :
625 1 : func (c *candidateLevelInfo) shouldCompact() bool {
626 1 : return c.compensatedScoreRatio >= compactionScoreThreshold
627 1 : }
628 :
629 1 : func fileCompensation(f *fileMetadata) uint64 {
630 1 : return uint64(f.Stats.PointDeletionsBytesEstimate) + f.Stats.RangeDeletionsBytesEstimate
631 1 : }
632 :
633 : // compensatedSize returns f's file size, inflated according to compaction
634 : // priorities.
635 1 : func compensatedSize(f *fileMetadata) uint64 {
636 1 : // Add in the estimate of disk space that may be reclaimed by compacting the
637 1 : // file's tombstones.
638 1 : return f.Size + fileCompensation(f)
639 1 : }
640 :
641 : // compensatedSizeAnnotator is a manifest.Annotator that annotates B-Tree
642 : // nodes with the sum of the files' compensated sizes. Compensated sizes may
643 : // change once a table's stats are loaded asynchronously, so its values are
644 : // marked as cacheable only if a file's stats have been loaded.
645 1 : var compensatedSizeAnnotator = manifest.SumAnnotator(func(f *fileMetadata) (uint64, bool) {
646 1 : return compensatedSize(f), f.StatsValid()
647 1 : })
648 :
649 : // totalCompensatedSize computes the compensated size over a file metadata
650 : // iterator. Note that this function is linear in the files available to the
651 : // iterator. Use the compensatedSizeAnnotator if querying the total
652 : // compensated size of a level.
653 1 : func totalCompensatedSize(iter manifest.LevelIterator) uint64 {
654 1 : var sz uint64
655 1 : for f := iter.First(); f != nil; f = iter.Next() {
656 1 : sz += compensatedSize(f)
657 1 : }
658 1 : return sz
659 : }
660 :
661 : // compactionPickerByScore holds the state and logic for picking a compaction. A
662 : // compaction picker is associated with a single version. A new compaction
663 : // picker is created and initialized every time a new version is installed.
664 : type compactionPickerByScore struct {
665 : opts *Options
666 : vers *version
667 : virtualBackings *manifest.VirtualBackings
668 : // The level to target for L0 compactions. Levels L1 to baseLevel must be
669 : // empty.
670 : baseLevel int
671 : // levelMaxBytes holds the dynamically adjusted max bytes setting for each
672 : // level.
673 : levelMaxBytes [numLevels]int64
674 : }
675 :
676 : var _ compactionPicker = &compactionPickerByScore{}
677 :
678 1 : func (p *compactionPickerByScore) getScores(inProgress []compactionInfo) [numLevels]float64 {
679 1 : var scores [numLevels]float64
680 1 : for _, info := range p.calculateLevelScores(inProgress) {
681 1 : scores[info.level] = info.compensatedScoreRatio
682 1 : }
683 1 : return scores
684 : }
685 :
686 1 : func (p *compactionPickerByScore) getBaseLevel() int {
687 1 : if p == nil {
688 0 : return 1
689 0 : }
690 1 : return p.baseLevel
691 : }
692 :
693 : // estimatedCompactionDebt estimates the number of bytes which need to be
694 : // compacted before the LSM tree becomes stable.
695 1 : func (p *compactionPickerByScore) estimatedCompactionDebt(l0ExtraSize uint64) uint64 {
696 1 : if p == nil {
697 0 : return 0
698 0 : }
699 :
700 : // We assume that all the bytes in L0 need to be compacted to Lbase. This is
701 : // unlike the RocksDB logic that figures out whether L0 needs compaction.
702 1 : bytesAddedToNextLevel := l0ExtraSize + p.vers.Levels[0].Size()
703 1 : lbaseSize := p.vers.Levels[p.baseLevel].Size()
704 1 :
705 1 : var compactionDebt uint64
706 1 : if bytesAddedToNextLevel > 0 && lbaseSize > 0 {
707 1 : // We only incur compaction debt if both L0 and Lbase contain data. If L0
708 1 : // is empty, no compaction is necessary. If Lbase is empty, a move-based
709 1 : // compaction from L0 would occur.
710 1 : compactionDebt += bytesAddedToNextLevel + lbaseSize
711 1 : }
712 :
713 : // loop invariant: At the beginning of the loop, bytesAddedToNextLevel is the
714 : // bytes added to `level` in the loop.
715 1 : for level := p.baseLevel; level < numLevels-1; level++ {
716 1 : levelSize := p.vers.Levels[level].Size() + bytesAddedToNextLevel
717 1 : nextLevelSize := p.vers.Levels[level+1].Size()
718 1 : if levelSize > uint64(p.levelMaxBytes[level]) {
719 1 : bytesAddedToNextLevel = levelSize - uint64(p.levelMaxBytes[level])
720 1 : if nextLevelSize > 0 {
721 1 : // We only incur compaction debt if the next level contains data. If the
722 1 : // next level is empty, a move-based compaction would be used.
723 1 : levelRatio := float64(nextLevelSize) / float64(levelSize)
724 1 : // The current level contributes bytesAddedToNextLevel to compactions.
725 1 : // The next level contributes levelRatio * bytesAddedToNextLevel.
726 1 : compactionDebt += uint64(float64(bytesAddedToNextLevel) * (levelRatio + 1))
727 1 : }
728 1 : } else {
729 1 : // We're not moving any bytes to the next level.
730 1 : bytesAddedToNextLevel = 0
731 1 : }
732 : }
733 1 : return compactionDebt
734 : }
735 :
736 1 : func (p *compactionPickerByScore) initLevelMaxBytes(inProgressCompactions []compactionInfo) {
737 1 : // The levelMaxBytes calculations here differ from RocksDB in two ways:
738 1 : //
739 1 : // 1. The use of dbSize vs maxLevelSize. RocksDB uses the size of the maximum
740 1 : // level in L1-L6, rather than determining the size of the bottom level
741 1 : // based on the total amount of data in the dB. The RocksDB calculation is
742 1 : // problematic if L0 contains a significant fraction of data, or if the
743 1 : // level sizes are roughly equal and thus there is a significant fraction
744 1 : // of data outside of the largest level.
745 1 : //
746 1 : // 2. Not adjusting the size of Lbase based on L0. RocksDB computes
747 1 : // baseBytesMax as the maximum of the configured LBaseMaxBytes and the
748 1 : // size of L0. This is problematic because baseBytesMax is used to compute
749 1 : // the max size of lower levels. A very large baseBytesMax will result in
750 1 : // an overly large value for the size of lower levels which will caused
751 1 : // those levels not to be compacted even when they should be
752 1 : // compacted. This often results in "inverted" LSM shapes where Ln is
753 1 : // larger than Ln+1.
754 1 :
755 1 : // Determine the first non-empty level and the total DB size.
756 1 : firstNonEmptyLevel := -1
757 1 : var dbSize uint64
758 1 : for level := 1; level < numLevels; level++ {
759 1 : if p.vers.Levels[level].Size() > 0 {
760 1 : if firstNonEmptyLevel == -1 {
761 1 : firstNonEmptyLevel = level
762 1 : }
763 1 : dbSize += p.vers.Levels[level].Size()
764 : }
765 : }
766 1 : for _, c := range inProgressCompactions {
767 1 : if c.outputLevel == 0 || c.outputLevel == -1 {
768 1 : continue
769 : }
770 1 : if c.inputs[0].level == 0 && (firstNonEmptyLevel == -1 || c.outputLevel < firstNonEmptyLevel) {
771 1 : firstNonEmptyLevel = c.outputLevel
772 1 : }
773 : }
774 :
775 : // Initialize the max-bytes setting for each level to "infinity" which will
776 : // disallow compaction for that level. We'll fill in the actual value below
777 : // for levels we want to allow compactions from.
778 1 : for level := 0; level < numLevels; level++ {
779 1 : p.levelMaxBytes[level] = math.MaxInt64
780 1 : }
781 :
782 1 : if dbSize == 0 {
783 1 : // No levels for L1 and up contain any data. Target L0 compactions for the
784 1 : // last level or to the level to which there is an ongoing L0 compaction.
785 1 : p.baseLevel = numLevels - 1
786 1 : if firstNonEmptyLevel >= 0 {
787 1 : p.baseLevel = firstNonEmptyLevel
788 1 : }
789 1 : return
790 : }
791 :
792 1 : dbSize += p.vers.Levels[0].Size()
793 1 : bottomLevelSize := dbSize - dbSize/uint64(p.opts.Experimental.LevelMultiplier)
794 1 :
795 1 : curLevelSize := bottomLevelSize
796 1 : for level := numLevels - 2; level >= firstNonEmptyLevel; level-- {
797 1 : curLevelSize = uint64(float64(curLevelSize) / float64(p.opts.Experimental.LevelMultiplier))
798 1 : }
799 :
800 : // Compute base level (where L0 data is compacted to).
801 1 : baseBytesMax := uint64(p.opts.LBaseMaxBytes)
802 1 : p.baseLevel = firstNonEmptyLevel
803 1 : for p.baseLevel > 1 && curLevelSize > baseBytesMax {
804 1 : p.baseLevel--
805 1 : curLevelSize = uint64(float64(curLevelSize) / float64(p.opts.Experimental.LevelMultiplier))
806 1 : }
807 :
808 1 : smoothedLevelMultiplier := 1.0
809 1 : if p.baseLevel < numLevels-1 {
810 1 : smoothedLevelMultiplier = math.Pow(
811 1 : float64(bottomLevelSize)/float64(baseBytesMax),
812 1 : 1.0/float64(numLevels-p.baseLevel-1))
813 1 : }
814 :
815 1 : levelSize := float64(baseBytesMax)
816 1 : for level := p.baseLevel; level < numLevels; level++ {
817 1 : if level > p.baseLevel && levelSize > 0 {
818 1 : levelSize *= smoothedLevelMultiplier
819 1 : }
820 : // Round the result since test cases use small target level sizes, which
821 : // can be impacted by floating-point imprecision + integer truncation.
822 1 : roundedLevelSize := math.Round(levelSize)
823 1 : if roundedLevelSize > float64(math.MaxInt64) {
824 0 : p.levelMaxBytes[level] = math.MaxInt64
825 1 : } else {
826 1 : p.levelMaxBytes[level] = int64(roundedLevelSize)
827 1 : }
828 : }
829 : }
830 :
831 : type levelSizeAdjust struct {
832 : incomingActualBytes uint64
833 : outgoingActualBytes uint64
834 : outgoingCompensatedBytes uint64
835 : }
836 :
837 1 : func (a levelSizeAdjust) compensated() uint64 {
838 1 : return a.incomingActualBytes - a.outgoingCompensatedBytes
839 1 : }
840 :
841 1 : func (a levelSizeAdjust) actual() uint64 {
842 1 : return a.incomingActualBytes - a.outgoingActualBytes
843 1 : }
844 :
845 1 : func calculateSizeAdjust(inProgressCompactions []compactionInfo) [numLevels]levelSizeAdjust {
846 1 : // Compute size adjustments for each level based on the in-progress
847 1 : // compactions. We sum the file sizes of all files leaving and entering each
848 1 : // level in in-progress compactions. For outgoing files, we also sum a
849 1 : // separate sum of 'compensated file sizes', which are inflated according
850 1 : // to deletion estimates.
851 1 : //
852 1 : // When we adjust a level's size according to these values during score
853 1 : // calculation, we subtract the compensated size of start level inputs to
854 1 : // account for the fact that score calculation uses compensated sizes.
855 1 : //
856 1 : // Since compensated file sizes may be compensated because they reclaim
857 1 : // space from the output level's files, we only add the real file size to
858 1 : // the output level.
859 1 : //
860 1 : // This is slightly different from RocksDB's behavior, which simply elides
861 1 : // compacting files from the level size calculation.
862 1 : var sizeAdjust [numLevels]levelSizeAdjust
863 1 : for i := range inProgressCompactions {
864 1 : c := &inProgressCompactions[i]
865 1 : // If this compaction's version edit has already been applied, there's
866 1 : // no need to adjust: The LSM we'll examine will already reflect the
867 1 : // new LSM state.
868 1 : if c.versionEditApplied {
869 1 : continue
870 : }
871 :
872 1 : for _, input := range c.inputs {
873 1 : actualSize := input.files.SizeSum()
874 1 : compensatedSize := totalCompensatedSize(input.files.Iter())
875 1 :
876 1 : if input.level != c.outputLevel {
877 1 : sizeAdjust[input.level].outgoingCompensatedBytes += compensatedSize
878 1 : sizeAdjust[input.level].outgoingActualBytes += actualSize
879 1 : if c.outputLevel != -1 {
880 1 : sizeAdjust[c.outputLevel].incomingActualBytes += actualSize
881 1 : }
882 : }
883 : }
884 : }
885 1 : return sizeAdjust
886 : }
887 :
888 : func (p *compactionPickerByScore) calculateLevelScores(
889 : inProgressCompactions []compactionInfo,
890 1 : ) [numLevels]candidateLevelInfo {
891 1 : var scores [numLevels]candidateLevelInfo
892 1 : for i := range scores {
893 1 : scores[i].level = i
894 1 : scores[i].outputLevel = i + 1
895 1 : }
896 1 : l0UncompensatedScore := calculateL0UncompensatedScore(p.vers, p.opts, inProgressCompactions)
897 1 : scores[0] = candidateLevelInfo{
898 1 : outputLevel: p.baseLevel,
899 1 : uncompensatedScore: l0UncompensatedScore,
900 1 : compensatedScore: l0UncompensatedScore, /* No level size compensation for L0 */
901 1 : }
902 1 : sizeAdjust := calculateSizeAdjust(inProgressCompactions)
903 1 : for level := 1; level < numLevels; level++ {
904 1 : compensatedLevelSize := *compensatedSizeAnnotator.LevelAnnotation(p.vers.Levels[level]) + sizeAdjust[level].compensated()
905 1 : scores[level].compensatedScore = float64(compensatedLevelSize) / float64(p.levelMaxBytes[level])
906 1 : scores[level].uncompensatedScore = float64(p.vers.Levels[level].Size()+sizeAdjust[level].actual()) / float64(p.levelMaxBytes[level])
907 1 : }
908 :
909 : // Adjust each level's {compensated, uncompensated}Score by the uncompensatedScore
910 : // of the next level to get a {compensated, uncompensated}ScoreRatio. If the
911 : // next level has a high uncompensatedScore, and is thus a priority for compaction,
912 : // this reduces the priority for compacting the current level. If the next level
913 : // has a low uncompensatedScore (i.e. it is below its target size), this increases
914 : // the priority for compacting the current level.
915 : //
916 : // The effect of this adjustment is to help prioritize compactions in lower
917 : // levels. The following example shows the compensatedScoreRatio and the
918 : // compensatedScore. In this scenario, L0 has 68 sublevels. L3 (a.k.a. Lbase)
919 : // is significantly above its target size. The original score prioritizes
920 : // compactions from those two levels, but doing so ends up causing a future
921 : // problem: data piles up in the higher levels, starving L5->L6 compactions,
922 : // and to a lesser degree starving L4->L5 compactions.
923 : //
924 : // Note that in the example shown there is no level size compensation so the
925 : // compensatedScore and the uncompensatedScore is the same for each level.
926 : //
927 : // compensatedScoreRatio compensatedScore uncompensatedScore size max-size
928 : // L0 3.2 68.0 68.0 2.2 G -
929 : // L3 3.2 21.1 21.1 1.3 G 64 M
930 : // L4 3.4 6.7 6.7 3.1 G 467 M
931 : // L5 3.4 2.0 2.0 6.6 G 3.3 G
932 : // L6 0.6 0.6 0.6 14 G 24 G
933 1 : var prevLevel int
934 1 : for level := p.baseLevel; level < numLevels; level++ {
935 1 : // The compensated scores, and uncompensated scores will be turned into
936 1 : // ratios as they're adjusted according to other levels' sizes.
937 1 : scores[prevLevel].compensatedScoreRatio = scores[prevLevel].compensatedScore
938 1 : scores[prevLevel].uncompensatedScoreRatio = scores[prevLevel].uncompensatedScore
939 1 :
940 1 : // Avoid absurdly large scores by placing a floor on the score that we'll
941 1 : // adjust a level by. The value of 0.01 was chosen somewhat arbitrarily.
942 1 : const minScore = 0.01
943 1 : if scores[prevLevel].compensatedScoreRatio >= compactionScoreThreshold {
944 1 : if scores[level].uncompensatedScore >= minScore {
945 1 : scores[prevLevel].compensatedScoreRatio /= scores[level].uncompensatedScore
946 1 : } else {
947 1 : scores[prevLevel].compensatedScoreRatio /= minScore
948 1 : }
949 : }
950 1 : if scores[prevLevel].uncompensatedScoreRatio >= compactionScoreThreshold {
951 1 : if scores[level].uncompensatedScore >= minScore {
952 1 : scores[prevLevel].uncompensatedScoreRatio /= scores[level].uncompensatedScore
953 1 : } else {
954 1 : scores[prevLevel].uncompensatedScoreRatio /= minScore
955 1 : }
956 : }
957 1 : prevLevel = level
958 : }
959 : // Set the score ratios for the lowest level.
960 : // INVARIANT: prevLevel == numLevels-1
961 1 : scores[prevLevel].compensatedScoreRatio = scores[prevLevel].compensatedScore
962 1 : scores[prevLevel].uncompensatedScoreRatio = scores[prevLevel].uncompensatedScore
963 1 :
964 1 : sort.Sort(sortCompactionLevelsByPriority(scores[:]))
965 1 : return scores
966 : }
967 :
968 : // calculateL0UncompensatedScore calculates a float score representing the
969 : // relative priority of compacting L0. Level L0 is special in that files within
970 : // L0 may overlap one another, so a different set of heuristics that take into
971 : // account read amplification apply.
972 : func calculateL0UncompensatedScore(
973 : vers *version, opts *Options, inProgressCompactions []compactionInfo,
974 1 : ) float64 {
975 1 : // Use the sublevel count to calculate the score. The base vs intra-L0
976 1 : // compaction determination happens in pickAuto, not here.
977 1 : score := float64(2*vers.L0Sublevels.MaxDepthAfterOngoingCompactions()) /
978 1 : float64(opts.L0CompactionThreshold)
979 1 :
980 1 : // Also calculate a score based on the file count but use it only if it
981 1 : // produces a higher score than the sublevel-based one. This heuristic is
982 1 : // designed to accommodate cases where L0 is accumulating non-overlapping
983 1 : // files in L0. Letting too many non-overlapping files accumulate in few
984 1 : // sublevels is undesirable, because:
985 1 : // 1) we can produce a massive backlog to compact once files do overlap.
986 1 : // 2) constructing L0 sublevels has a runtime that grows superlinearly with
987 1 : // the number of files in L0 and must be done while holding D.mu.
988 1 : noncompactingFiles := vers.Levels[0].Len()
989 1 : for _, c := range inProgressCompactions {
990 1 : for _, cl := range c.inputs {
991 1 : if cl.level == 0 {
992 1 : noncompactingFiles -= cl.files.Len()
993 1 : }
994 : }
995 : }
996 1 : fileScore := float64(noncompactingFiles) / float64(opts.L0CompactionFileThreshold)
997 1 : if score < fileScore {
998 1 : score = fileScore
999 1 : }
1000 1 : return score
1001 : }
1002 :
1003 : // pickCompactionSeedFile picks a file from `level` in the `vers` to build a
1004 : // compaction around. Currently, this function implements a heuristic similar to
1005 : // RocksDB's kMinOverlappingRatio, seeking to minimize write amplification. This
1006 : // function is linear with respect to the number of files in `level` and
1007 : // `outputLevel`.
1008 : func pickCompactionSeedFile(
1009 : vers *version,
1010 : virtualBackings *manifest.VirtualBackings,
1011 : opts *Options,
1012 : level, outputLevel int,
1013 : earliestSnapshotSeqNum base.SeqNum,
1014 1 : ) (manifest.LevelFile, bool) {
1015 1 : // Select the file within the level to compact. We want to minimize write
1016 1 : // amplification, but also ensure that (a) deletes are propagated to the
1017 1 : // bottom level in a timely fashion, and (b) virtual sstables that are
1018 1 : // pinning backing sstables where most of the data is garbage are compacted
1019 1 : // away. Doing (a) and (b) reclaims disk space. A table's smallest sequence
1020 1 : // number provides a measure of its age. The ratio of overlapping-bytes /
1021 1 : // table-size gives an indication of write amplification (a smaller ratio is
1022 1 : // preferrable).
1023 1 : //
1024 1 : // The current heuristic is based off the RocksDB kMinOverlappingRatio
1025 1 : // heuristic. It chooses the file with the minimum overlapping ratio with
1026 1 : // the target level, which minimizes write amplification.
1027 1 : //
1028 1 : // The heuristic uses a "compensated size" for the denominator, which is the
1029 1 : // file size inflated by (a) an estimate of the space that may be reclaimed
1030 1 : // through compaction, and (b) a fraction of the amount of garbage in the
1031 1 : // backing sstable pinned by this (virtual) sstable.
1032 1 : //
1033 1 : // TODO(peter): For concurrent compactions, we may want to try harder to
1034 1 : // pick a seed file whose resulting compaction bounds do not overlap with
1035 1 : // an in-progress compaction.
1036 1 :
1037 1 : cmp := opts.Comparer.Compare
1038 1 : startIter := vers.Levels[level].Iter()
1039 1 : outputIter := vers.Levels[outputLevel].Iter()
1040 1 :
1041 1 : var file manifest.LevelFile
1042 1 : smallestRatio := uint64(math.MaxUint64)
1043 1 :
1044 1 : outputFile := outputIter.First()
1045 1 :
1046 1 : for f := startIter.First(); f != nil; f = startIter.Next() {
1047 1 : var overlappingBytes uint64
1048 1 : compacting := f.IsCompacting()
1049 1 : if compacting {
1050 1 : // Move on if this file is already being compacted. We'll likely
1051 1 : // still need to move past the overlapping output files regardless,
1052 1 : // but in cases where all start-level files are compacting we won't.
1053 1 : continue
1054 : }
1055 :
1056 : // Trim any output-level files smaller than f.
1057 1 : for outputFile != nil && sstableKeyCompare(cmp, outputFile.Largest, f.Smallest) < 0 {
1058 1 : outputFile = outputIter.Next()
1059 1 : }
1060 :
1061 1 : for outputFile != nil && sstableKeyCompare(cmp, outputFile.Smallest, f.Largest) <= 0 && !compacting {
1062 1 : overlappingBytes += outputFile.Size
1063 1 : compacting = compacting || outputFile.IsCompacting()
1064 1 :
1065 1 : // For files in the bottommost level of the LSM, the
1066 1 : // Stats.RangeDeletionsBytesEstimate field is set to the estimate
1067 1 : // of bytes /within/ the file itself that may be dropped by
1068 1 : // recompacting the file. These bytes from obsolete keys would not
1069 1 : // need to be rewritten if we compacted `f` into `outputFile`, so
1070 1 : // they don't contribute to write amplification. Subtracting them
1071 1 : // out of the overlapping bytes helps prioritize these compactions
1072 1 : // that are cheaper than their file sizes suggest.
1073 1 : if outputLevel == numLevels-1 && outputFile.LargestSeqNum < earliestSnapshotSeqNum {
1074 1 : overlappingBytes -= outputFile.Stats.RangeDeletionsBytesEstimate
1075 1 : }
1076 :
1077 : // If the file in the next level extends beyond f's largest key,
1078 : // break out and don't advance outputIter because f's successor
1079 : // might also overlap.
1080 : //
1081 : // Note, we stop as soon as we encounter an output-level file with a
1082 : // largest key beyond the input-level file's largest bound. We
1083 : // perform a simple user key comparison here using sstableKeyCompare
1084 : // which handles the potential for exclusive largest key bounds.
1085 : // There's some subtlety when the bounds are equal (eg, equal and
1086 : // inclusive, or equal and exclusive). Current Pebble doesn't split
1087 : // user keys across sstables within a level (and in format versions
1088 : // FormatSplitUserKeysMarkedCompacted and later we guarantee no
1089 : // split user keys exist within the entire LSM). In that case, we're
1090 : // assured that neither the input level nor the output level's next
1091 : // file shares the same user key, so compaction expansion will not
1092 : // include them in any compaction compacting `f`.
1093 : //
1094 : // NB: If we /did/ allow split user keys, or we're running on an
1095 : // old database with an earlier format major version where there are
1096 : // existing split user keys, this logic would be incorrect. Consider
1097 : // L1: [a#120,a#100] [a#80,a#60]
1098 : // L2: [a#55,a#45] [a#35,a#25] [a#15,a#5]
1099 : // While considering the first file in L1, [a#120,a#100], we'd skip
1100 : // past all of the files in L2. When considering the second file in
1101 : // L1, we'd improperly conclude that the second file overlaps
1102 : // nothing in the second level and is cheap to compact, when in
1103 : // reality we'd need to expand the compaction to include all 5
1104 : // files.
1105 1 : if sstableKeyCompare(cmp, outputFile.Largest, f.Largest) > 0 {
1106 1 : break
1107 : }
1108 1 : outputFile = outputIter.Next()
1109 : }
1110 :
1111 : // If the input level file or one of the overlapping files is
1112 : // compacting, we're not going to be able to compact this file
1113 : // anyways, so skip it.
1114 1 : if compacting {
1115 1 : continue
1116 : }
1117 :
1118 1 : compSz := compensatedSize(f) + responsibleForGarbageBytes(virtualBackings, f)
1119 1 : scaledRatio := overlappingBytes * 1024 / compSz
1120 1 : if scaledRatio < smallestRatio {
1121 1 : smallestRatio = scaledRatio
1122 1 : file = startIter.Take()
1123 1 : }
1124 : }
1125 1 : return file, file.FileMetadata != nil
1126 : }
1127 :
1128 : // responsibleForGarbageBytes returns the amount of garbage in the backing
1129 : // sstable that we consider the responsibility of this virtual sstable. For
1130 : // non-virtual sstables, this is of course 0. For virtual sstables, we equally
1131 : // distribute the responsibility of the garbage across all the virtual
1132 : // sstables that are referencing the same backing sstable. One could
1133 : // alternatively distribute this in proportion to the virtual sst sizes, but
1134 : // it isn't clear that more sophisticated heuristics are worth it, given that
1135 : // the garbage cannot be reclaimed until all the referencing virtual sstables
1136 : // are compacted.
1137 1 : func responsibleForGarbageBytes(virtualBackings *manifest.VirtualBackings, m *fileMetadata) uint64 {
1138 1 : if !m.Virtual {
1139 1 : return 0
1140 1 : }
1141 1 : useCount, virtualizedSize := virtualBackings.Usage(m.FileBacking.DiskFileNum)
1142 1 : // Since virtualizedSize is the sum of the estimated size of all virtual
1143 1 : // ssts, we allow for the possibility that virtualizedSize could exceed
1144 1 : // m.FileBacking.Size.
1145 1 : totalGarbage := int64(m.FileBacking.Size) - int64(virtualizedSize)
1146 1 : if totalGarbage <= 0 {
1147 0 : return 0
1148 0 : }
1149 1 : if useCount == 0 {
1150 0 : // This cannot happen if m exists in the latest version. The call to
1151 0 : // ResponsibleForGarbageBytes during compaction picking ensures that m
1152 0 : // exists in the latest version by holding versionSet.logLock.
1153 0 : panic(errors.AssertionFailedf("%s has zero useCount", m.String()))
1154 : }
1155 1 : return uint64(totalGarbage) / uint64(useCount)
1156 : }
1157 :
1158 : // pickAuto picks the best compaction, if any.
1159 : //
1160 : // On each call, pickAuto computes per-level size adjustments based on
1161 : // in-progress compactions, and computes a per-level score. The levels are
1162 : // iterated over in decreasing score order trying to find a valid compaction
1163 : // anchored at that level.
1164 : //
1165 : // If a score-based compaction cannot be found, pickAuto falls back to looking
1166 : // for an elision-only compaction to remove obsolete keys.
1167 1 : func (p *compactionPickerByScore) pickAuto(env compactionEnv) (pc *pickedCompaction) {
1168 1 : // Compaction concurrency is controlled by L0 read-amp. We allow one
1169 1 : // additional compaction per L0CompactionConcurrency sublevels, as well as
1170 1 : // one additional compaction per CompactionDebtConcurrency bytes of
1171 1 : // compaction debt. Compaction concurrency is tied to L0 sublevels as that
1172 1 : // signal is independent of the database size. We tack on the compaction
1173 1 : // debt as a second signal to prevent compaction concurrency from dropping
1174 1 : // significantly right after a base compaction finishes, and before those
1175 1 : // bytes have been compacted further down the LSM.
1176 1 : if n := len(env.inProgressCompactions); n > 0 {
1177 1 : l0ReadAmp := p.vers.L0Sublevels.MaxDepthAfterOngoingCompactions()
1178 1 : compactionDebt := p.estimatedCompactionDebt(0)
1179 1 : ccSignal1 := n * p.opts.Experimental.L0CompactionConcurrency
1180 1 : ccSignal2 := uint64(n) * p.opts.Experimental.CompactionDebtConcurrency
1181 1 : if l0ReadAmp < ccSignal1 && compactionDebt < ccSignal2 {
1182 1 : return nil
1183 1 : }
1184 : }
1185 :
1186 : // NB: CompactionLimiter defaults to a no-op limiter unless one is implemented
1187 : // and passed-in as an option during Open.
1188 1 : limiter := p.opts.Experimental.CompactionLimiter
1189 1 : var slot base.CompactionSlot
1190 1 : if n := len(env.inProgressCompactions); n == 0 {
1191 1 : // We are not running a compaction at the moment. We should take a compaction slot
1192 1 : // without permission.
1193 1 : slot = limiter.TookWithoutPermission(context.TODO())
1194 1 : } else {
1195 1 : var err error
1196 1 : slot, err = limiter.RequestSlot(context.TODO())
1197 1 : if err != nil {
1198 0 : p.opts.EventListener.BackgroundError(err)
1199 0 : return nil
1200 0 : }
1201 1 : if slot == nil {
1202 0 : // The limiter is denying us a compaction slot. Yield to other work.
1203 0 : return nil
1204 0 : }
1205 : }
1206 1 : defer func() {
1207 1 : if pc != nil {
1208 1 : var inputSize uint64
1209 1 : for i := range pc.inputs {
1210 1 : inputSize += pc.inputs[i].files.SizeSum()
1211 1 : }
1212 1 : slot.CompactionSelected(pc.startLevel.level, pc.outputLevel.level, inputSize)
1213 1 : pc.slot = slot
1214 : }
1215 : }()
1216 :
1217 1 : scores := p.calculateLevelScores(env.inProgressCompactions)
1218 1 :
1219 1 : // TODO(bananabrick): Either remove, or change this into an event sent to the
1220 1 : // EventListener.
1221 1 : logCompaction := func(pc *pickedCompaction) {
1222 0 : var buf bytes.Buffer
1223 0 : for i := 0; i < numLevels; i++ {
1224 0 : if i != 0 && i < p.baseLevel {
1225 0 : continue
1226 : }
1227 :
1228 0 : var info *candidateLevelInfo
1229 0 : for j := range scores {
1230 0 : if scores[j].level == i {
1231 0 : info = &scores[j]
1232 0 : break
1233 : }
1234 : }
1235 :
1236 0 : marker := " "
1237 0 : if pc.startLevel.level == info.level {
1238 0 : marker = "*"
1239 0 : }
1240 0 : fmt.Fprintf(&buf, " %sL%d: %5.1f %5.1f %5.1f %5.1f %8s %8s",
1241 0 : marker, info.level, info.compensatedScoreRatio, info.compensatedScore,
1242 0 : info.uncompensatedScoreRatio, info.uncompensatedScore,
1243 0 : humanize.Bytes.Int64(int64(totalCompensatedSize(
1244 0 : p.vers.Levels[info.level].Iter(),
1245 0 : ))),
1246 0 : humanize.Bytes.Int64(p.levelMaxBytes[info.level]),
1247 0 : )
1248 0 :
1249 0 : count := 0
1250 0 : for i := range env.inProgressCompactions {
1251 0 : c := &env.inProgressCompactions[i]
1252 0 : if c.inputs[0].level != info.level {
1253 0 : continue
1254 : }
1255 0 : count++
1256 0 : if count == 1 {
1257 0 : fmt.Fprintf(&buf, " [")
1258 0 : } else {
1259 0 : fmt.Fprintf(&buf, " ")
1260 0 : }
1261 0 : fmt.Fprintf(&buf, "L%d->L%d", c.inputs[0].level, c.outputLevel)
1262 : }
1263 0 : if count > 0 {
1264 0 : fmt.Fprintf(&buf, "]")
1265 0 : }
1266 0 : fmt.Fprintf(&buf, "\n")
1267 : }
1268 0 : p.opts.Logger.Infof("pickAuto: L%d->L%d\n%s",
1269 0 : pc.startLevel.level, pc.outputLevel.level, buf.String())
1270 : }
1271 :
1272 : // Check for a score-based compaction. candidateLevelInfos are first sorted
1273 : // by whether they should be compacted, so if we find a level which shouldn't
1274 : // be compacted, we can break early.
1275 1 : for i := range scores {
1276 1 : info := &scores[i]
1277 1 : if !info.shouldCompact() {
1278 1 : break
1279 : }
1280 1 : if info.level == numLevels-1 {
1281 1 : continue
1282 : }
1283 :
1284 1 : if info.level == 0 {
1285 1 : pc = pickL0(env, p.opts, p.vers, p.baseLevel)
1286 1 : // Fail-safe to protect against compacting the same sstable
1287 1 : // concurrently.
1288 1 : if pc != nil && !inputRangeAlreadyCompacting(env, pc) {
1289 1 : p.addScoresToPickedCompactionMetrics(pc, scores)
1290 1 : pc.score = info.compensatedScoreRatio
1291 1 : // TODO(bananabrick): Create an EventListener for logCompaction.
1292 1 : if false {
1293 0 : logCompaction(pc)
1294 0 : }
1295 1 : return pc
1296 : }
1297 1 : continue
1298 : }
1299 :
1300 : // info.level > 0
1301 1 : var ok bool
1302 1 : info.file, ok = pickCompactionSeedFile(p.vers, p.virtualBackings, p.opts, info.level, info.outputLevel, env.earliestSnapshotSeqNum)
1303 1 : if !ok {
1304 1 : continue
1305 : }
1306 :
1307 1 : pc := pickAutoLPositive(env, p.opts, p.vers, *info, p.baseLevel)
1308 1 : // Fail-safe to protect against compacting the same sstable concurrently.
1309 1 : if pc != nil && !inputRangeAlreadyCompacting(env, pc) {
1310 1 : p.addScoresToPickedCompactionMetrics(pc, scores)
1311 1 : pc.score = info.compensatedScoreRatio
1312 1 : // TODO(bananabrick): Create an EventListener for logCompaction.
1313 1 : if false {
1314 0 : logCompaction(pc)
1315 0 : }
1316 1 : return pc
1317 : }
1318 : }
1319 :
1320 : // Check for files which contain excessive point tombstones that could slow
1321 : // down reads. Unlike elision-only compactions, these compactions may select
1322 : // a file at any level rather than only the lowest level.
1323 1 : if pc := p.pickTombstoneDensityCompaction(env); pc != nil {
1324 0 : return pc
1325 0 : }
1326 :
1327 : // Check for L6 files with tombstones that may be elided. These files may
1328 : // exist if a snapshot prevented the elision of a tombstone or because of
1329 : // a move compaction. These are low-priority compactions because they
1330 : // don't help us keep up with writes, just reclaim disk space.
1331 1 : if pc := p.pickElisionOnlyCompaction(env); pc != nil {
1332 1 : return pc
1333 1 : }
1334 :
1335 1 : if pc := p.pickReadTriggeredCompaction(env); pc != nil {
1336 1 : return pc
1337 1 : }
1338 :
1339 : // NB: This should only be run if a read compaction wasn't
1340 : // scheduled.
1341 : //
1342 : // We won't be scheduling a read compaction right now, and in
1343 : // read heavy workloads, compactions won't be scheduled frequently
1344 : // because flushes aren't frequent. So we need to signal to the
1345 : // iterator to schedule a compaction when it adds compactions to
1346 : // the read compaction queue.
1347 : //
1348 : // We need the nil check here because without it, we have some
1349 : // tests which don't set that variable fail. Since there's a
1350 : // chance that one of those tests wouldn't want extra compactions
1351 : // to be scheduled, I added this check here, instead of
1352 : // setting rescheduleReadCompaction in those tests.
1353 1 : if env.readCompactionEnv.rescheduleReadCompaction != nil {
1354 1 : *env.readCompactionEnv.rescheduleReadCompaction = true
1355 1 : }
1356 :
1357 : // At the lowest possible compaction-picking priority, look for files marked
1358 : // for compaction. Pebble will mark files for compaction if they have atomic
1359 : // compaction units that span multiple files. While current Pebble code does
1360 : // not construct such sstables, RocksDB and earlier versions of Pebble may
1361 : // have created them. These split user keys form sets of files that must be
1362 : // compacted together for correctness (referred to as "atomic compaction
1363 : // units" within the code). Rewrite them in-place.
1364 : //
1365 : // It's also possible that a file may have been marked for compaction by
1366 : // even earlier versions of Pebble code, since FileMetadata's
1367 : // MarkedForCompaction field is persisted in the manifest. That's okay. We
1368 : // previously would've ignored the designation, whereas now we'll re-compact
1369 : // the file in place.
1370 1 : if p.vers.Stats.MarkedForCompaction > 0 {
1371 1 : if pc := p.pickRewriteCompaction(env); pc != nil {
1372 1 : return pc
1373 1 : }
1374 : }
1375 :
1376 1 : return nil
1377 : }
1378 :
1379 : func (p *compactionPickerByScore) addScoresToPickedCompactionMetrics(
1380 : pc *pickedCompaction, candInfo [numLevels]candidateLevelInfo,
1381 1 : ) {
1382 1 :
1383 1 : // candInfo is sorted by score, not by compaction level.
1384 1 : infoByLevel := [numLevels]candidateLevelInfo{}
1385 1 : for i := range candInfo {
1386 1 : level := candInfo[i].level
1387 1 : infoByLevel[level] = candInfo[i]
1388 1 : }
1389 : // Gather the compaction scores for the levels participating in the compaction.
1390 1 : pc.pickerMetrics.scores = make([]float64, len(pc.inputs))
1391 1 : inputIdx := 0
1392 1 : for i := range infoByLevel {
1393 1 : if pc.inputs[inputIdx].level == infoByLevel[i].level {
1394 1 : pc.pickerMetrics.scores[inputIdx] = infoByLevel[i].compensatedScoreRatio
1395 1 : inputIdx++
1396 1 : }
1397 1 : if inputIdx == len(pc.inputs) {
1398 1 : break
1399 : }
1400 : }
1401 : }
1402 :
1403 : // elisionOnlyAnnotator is a manifest.Annotator that annotates B-Tree
1404 : // nodes with the *fileMetadata of a file meeting the obsolete keys criteria
1405 : // for an elision-only compaction within the subtree. If multiple files meet
1406 : // the criteria, it chooses whichever file has the lowest LargestSeqNum. The
1407 : // lowest LargestSeqNum file will be the first eligible for an elision-only
1408 : // compaction once snapshots less than or equal to its LargestSeqNum are closed.
1409 : var elisionOnlyAnnotator = &manifest.Annotator[fileMetadata]{
1410 : Aggregator: manifest.PickFileAggregator{
1411 1 : Filter: func(f *fileMetadata) (eligible bool, cacheOK bool) {
1412 1 : if f.IsCompacting() {
1413 1 : return false, true
1414 1 : }
1415 1 : if !f.StatsValid() {
1416 1 : return false, false
1417 1 : }
1418 : // Bottommost files are large and not worthwhile to compact just
1419 : // to remove a few tombstones. Consider a file eligible only if
1420 : // either its own range deletions delete at least 10% of its data or
1421 : // its deletion tombstones make at least 10% of its entries.
1422 : //
1423 : // TODO(jackson): This does not account for duplicate user keys
1424 : // which may be collapsed. Ideally, we would have 'obsolete keys'
1425 : // statistics that would include tombstones, the keys that are
1426 : // dropped by tombstones and duplicated user keys. See #847.
1427 : //
1428 : // Note that tables that contain exclusively range keys (i.e. no point keys,
1429 : // `NumEntries` and `RangeDeletionsBytesEstimate` are both zero) are excluded
1430 : // from elision-only compactions.
1431 : // TODO(travers): Consider an alternative heuristic for elision of range-keys.
1432 1 : return f.Stats.RangeDeletionsBytesEstimate*10 >= f.Size || f.Stats.NumDeletions*10 > f.Stats.NumEntries, true
1433 : },
1434 1 : Compare: func(f1 *fileMetadata, f2 *fileMetadata) bool {
1435 1 : return f1.LargestSeqNum < f2.LargestSeqNum
1436 1 : },
1437 : },
1438 : }
1439 :
1440 : // markedForCompactionAnnotator is a manifest.Annotator that annotates B-Tree
1441 : // nodes with the *fileMetadata of a file that is marked for compaction
1442 : // within the subtree. If multiple files meet the criteria, it chooses
1443 : // whichever file has the lowest LargestSeqNum.
1444 : var markedForCompactionAnnotator = &manifest.Annotator[fileMetadata]{
1445 : Aggregator: manifest.PickFileAggregator{
1446 1 : Filter: func(f *fileMetadata) (eligible bool, cacheOK bool) {
1447 1 : return f.MarkedForCompaction, true
1448 1 : },
1449 0 : Compare: func(f1 *fileMetadata, f2 *fileMetadata) bool {
1450 0 : return f1.LargestSeqNum < f2.LargestSeqNum
1451 0 : },
1452 : },
1453 : }
1454 :
1455 : // pickedCompactionFromCandidateFile creates a pickedCompaction from a *fileMetadata
1456 : // with various checks to ensure that the file still exists in the expected level
1457 : // and isn't already being compacted.
1458 : func (p *compactionPickerByScore) pickedCompactionFromCandidateFile(
1459 : candidate *fileMetadata, env compactionEnv, startLevel int, outputLevel int, kind compactionKind,
1460 1 : ) *pickedCompaction {
1461 1 : if candidate == nil || candidate.IsCompacting() {
1462 1 : return nil
1463 1 : }
1464 :
1465 1 : var inputs manifest.LevelSlice
1466 1 : if startLevel == 0 {
1467 1 : // Overlapping L0 files must also be compacted alongside the candidate.
1468 1 : inputs = p.vers.Overlaps(0, candidate.UserKeyBounds())
1469 1 : } else {
1470 1 : inputs = p.vers.Levels[startLevel].Find(p.opts.Comparer.Compare, candidate)
1471 1 : }
1472 1 : if invariants.Enabled {
1473 1 : found := false
1474 1 : inputs.Each(func(f *fileMetadata) {
1475 1 : if f.FileNum == candidate.FileNum {
1476 1 : found = true
1477 1 : }
1478 : })
1479 1 : if !found {
1480 0 : panic(fmt.Sprintf("file %s not found in level %d as expected", candidate.FileNum, startLevel))
1481 : }
1482 : }
1483 :
1484 1 : pc := newPickedCompaction(p.opts, p.vers, startLevel, outputLevel, p.baseLevel)
1485 1 : pc.kind = kind
1486 1 : pc.startLevel.files = inputs
1487 1 : pc.smallest, pc.largest = manifest.KeyRange(pc.cmp, pc.startLevel.files.Iter())
1488 1 :
1489 1 : // Fail-safe to protect against compacting the same sstable concurrently.
1490 1 : if inputRangeAlreadyCompacting(env, pc) {
1491 0 : return nil
1492 0 : }
1493 :
1494 1 : if !pc.setupInputs(p.opts, env.diskAvailBytes, pc.startLevel) {
1495 0 : return nil
1496 0 : }
1497 :
1498 1 : return pc
1499 : }
1500 :
1501 : // pickElisionOnlyCompaction looks for compactions of sstables in the
1502 : // bottommost level containing obsolete records that may now be dropped.
1503 : func (p *compactionPickerByScore) pickElisionOnlyCompaction(
1504 : env compactionEnv,
1505 1 : ) (pc *pickedCompaction) {
1506 1 : if p.opts.private.disableElisionOnlyCompactions {
1507 1 : return nil
1508 1 : }
1509 1 : candidate := elisionOnlyAnnotator.LevelAnnotation(p.vers.Levels[numLevels-1])
1510 1 : if candidate == nil {
1511 1 : return nil
1512 1 : }
1513 1 : if candidate.LargestSeqNum >= env.earliestSnapshotSeqNum {
1514 1 : return nil
1515 1 : }
1516 1 : return p.pickedCompactionFromCandidateFile(candidate, env, numLevels-1, numLevels-1, compactionKindElisionOnly)
1517 : }
1518 :
1519 : // pickRewriteCompaction attempts to construct a compaction that
1520 : // rewrites a file marked for compaction. pickRewriteCompaction will
1521 : // pull in adjacent files in the file's atomic compaction unit if
1522 : // necessary. A rewrite compaction outputs files to the same level as
1523 : // the input level.
1524 1 : func (p *compactionPickerByScore) pickRewriteCompaction(env compactionEnv) (pc *pickedCompaction) {
1525 1 : for l := numLevels - 1; l >= 0; l-- {
1526 1 : candidate := markedForCompactionAnnotator.LevelAnnotation(p.vers.Levels[l])
1527 1 : if candidate == nil {
1528 1 : // Try the next level.
1529 1 : continue
1530 : }
1531 1 : pc := p.pickedCompactionFromCandidateFile(candidate, env, l, l, compactionKindRewrite)
1532 1 : if pc != nil {
1533 1 : return pc
1534 1 : }
1535 : }
1536 0 : return nil
1537 : }
1538 :
1539 : // pickTombstoneDensityCompaction looks for a compaction that eliminates
1540 : // regions of extremely high point tombstone density. For each level, it picks
1541 : // a file where the ratio of tombstone-dense blocks is at least
1542 : // options.Experimental.MinTombstoneDenseRatio, prioritizing compaction of
1543 : // files with higher ratios of tombstone-dense blocks.
1544 : func (p *compactionPickerByScore) pickTombstoneDensityCompaction(
1545 : env compactionEnv,
1546 1 : ) (pc *pickedCompaction) {
1547 1 : if p.opts.Experimental.TombstoneDenseCompactionThreshold <= 0 {
1548 0 : // Tombstone density compactions are disabled.
1549 0 : return nil
1550 0 : }
1551 :
1552 1 : var candidate *fileMetadata
1553 1 : var level int
1554 1 : // If a candidate file has a very high overlapping ratio, point tombstones
1555 1 : // in it are likely sparse in keyspace even if the sstable itself is tombstone
1556 1 : // dense. These tombstones likely wouldn't be slow to iterate over, so we exclude
1557 1 : // these files from tombstone density compactions. The threshold of 40.0 is
1558 1 : // chosen somewhat arbitrarily, after some observations around excessively large
1559 1 : // tombstone density compactions.
1560 1 : const maxOverlappingRatio = 40.0
1561 1 : // NB: we don't consider the lowest level because elision-only compactions
1562 1 : // handle that case.
1563 1 : lastNonEmptyLevel := numLevels - 1
1564 1 : for l := numLevels - 2; l >= 0; l-- {
1565 1 : iter := p.vers.Levels[l].Iter()
1566 1 : for f := iter.First(); f != nil; f = iter.Next() {
1567 1 : if f.IsCompacting() || !f.StatsValid() || f.Size == 0 {
1568 1 : continue
1569 : }
1570 1 : if f.Stats.TombstoneDenseBlocksRatio < p.opts.Experimental.TombstoneDenseCompactionThreshold {
1571 1 : continue
1572 : }
1573 0 : overlaps := p.vers.Overlaps(lastNonEmptyLevel, f.UserKeyBounds())
1574 0 : if float64(overlaps.SizeSum())/float64(f.Size) > maxOverlappingRatio {
1575 0 : continue
1576 : }
1577 0 : if candidate == nil || candidate.Stats.TombstoneDenseBlocksRatio < f.Stats.TombstoneDenseBlocksRatio {
1578 0 : candidate = f
1579 0 : level = l
1580 0 : }
1581 : }
1582 : // We prefer lower level (ie. L5) candidates over higher level (ie. L4) ones.
1583 1 : if candidate != nil {
1584 0 : break
1585 : }
1586 1 : if !p.vers.Levels[l].Empty() {
1587 1 : lastNonEmptyLevel = l
1588 1 : }
1589 : }
1590 :
1591 1 : return p.pickedCompactionFromCandidateFile(candidate, env, level, defaultOutputLevel(level, p.baseLevel), compactionKindTombstoneDensity)
1592 : }
1593 :
1594 : // pickAutoLPositive picks an automatic compaction for the candidate
1595 : // file in a positive-numbered level. This function must not be used for
1596 : // L0.
1597 : func pickAutoLPositive(
1598 : env compactionEnv, opts *Options, vers *version, cInfo candidateLevelInfo, baseLevel int,
1599 1 : ) (pc *pickedCompaction) {
1600 1 : if cInfo.level == 0 {
1601 0 : panic("pebble: pickAutoLPositive called for L0")
1602 : }
1603 :
1604 1 : pc = newPickedCompaction(opts, vers, cInfo.level, defaultOutputLevel(cInfo.level, baseLevel), baseLevel)
1605 1 : if pc.outputLevel.level != cInfo.outputLevel {
1606 0 : panic("pebble: compaction picked unexpected output level")
1607 : }
1608 1 : pc.startLevel.files = cInfo.file.Slice()
1609 1 : // Files in level 0 may overlap each other, so pick up all overlapping ones.
1610 1 : if pc.startLevel.level == 0 {
1611 0 : cmp := opts.Comparer.Compare
1612 0 : smallest, largest := manifest.KeyRange(cmp, pc.startLevel.files.Iter())
1613 0 : pc.startLevel.files = vers.Overlaps(0, base.UserKeyBoundsFromInternal(smallest, largest))
1614 0 : if pc.startLevel.files.Empty() {
1615 0 : panic("pebble: empty compaction")
1616 : }
1617 : }
1618 :
1619 1 : if !pc.setupInputs(opts, env.diskAvailBytes, pc.startLevel) {
1620 0 : return nil
1621 0 : }
1622 1 : return pc.maybeAddLevel(opts, env.diskAvailBytes)
1623 : }
1624 :
1625 : // maybeAddLevel maybe adds a level to the picked compaction.
1626 1 : func (pc *pickedCompaction) maybeAddLevel(opts *Options, diskAvailBytes uint64) *pickedCompaction {
1627 1 : pc.pickerMetrics.singleLevelOverlappingRatio = pc.overlappingRatio()
1628 1 : if pc.outputLevel.level == numLevels-1 {
1629 1 : // Don't add a level if the current output level is in L6
1630 1 : return pc
1631 1 : }
1632 1 : if !opts.Experimental.MultiLevelCompactionHeuristic.allowL0() && pc.startLevel.level == 0 {
1633 1 : return pc
1634 1 : }
1635 1 : if pc.compactionSize() > expandedCompactionByteSizeLimit(
1636 1 : opts, adjustedOutputLevel(pc.outputLevel.level, pc.baseLevel), diskAvailBytes) {
1637 1 : // Don't add a level if the current compaction exceeds the compaction size limit
1638 1 : return pc
1639 1 : }
1640 1 : return opts.Experimental.MultiLevelCompactionHeuristic.pick(pc, opts, diskAvailBytes)
1641 : }
1642 :
1643 : // MultiLevelHeuristic evaluates whether to add files from the next level into the compaction.
1644 : type MultiLevelHeuristic interface {
1645 : // Evaluate returns the preferred compaction.
1646 : pick(pc *pickedCompaction, opts *Options, diskAvailBytes uint64) *pickedCompaction
1647 :
1648 : // Returns if the heuristic allows L0 to be involved in ML compaction
1649 : allowL0() bool
1650 :
1651 : // String implements fmt.Stringer.
1652 : String() string
1653 : }
1654 :
1655 : // NoMultiLevel will never add an additional level to the compaction.
1656 : type NoMultiLevel struct{}
1657 :
1658 : var _ MultiLevelHeuristic = (*NoMultiLevel)(nil)
1659 :
1660 : func (nml NoMultiLevel) pick(
1661 : pc *pickedCompaction, opts *Options, diskAvailBytes uint64,
1662 1 : ) *pickedCompaction {
1663 1 : return pc
1664 1 : }
1665 :
1666 1 : func (nml NoMultiLevel) allowL0() bool { return false }
1667 1 : func (nml NoMultiLevel) String() string { return "none" }
1668 :
1669 1 : func (pc *pickedCompaction) predictedWriteAmp() float64 {
1670 1 : var bytesToCompact uint64
1671 1 : var higherLevelBytes uint64
1672 1 : for i := range pc.inputs {
1673 1 : levelSize := pc.inputs[i].files.SizeSum()
1674 1 : bytesToCompact += levelSize
1675 1 : if i != len(pc.inputs)-1 {
1676 1 : higherLevelBytes += levelSize
1677 1 : }
1678 : }
1679 1 : return float64(bytesToCompact) / float64(higherLevelBytes)
1680 : }
1681 :
1682 1 : func (pc *pickedCompaction) overlappingRatio() float64 {
1683 1 : var higherLevelBytes uint64
1684 1 : var lowestLevelBytes uint64
1685 1 : for i := range pc.inputs {
1686 1 : levelSize := pc.inputs[i].files.SizeSum()
1687 1 : if i == len(pc.inputs)-1 {
1688 1 : lowestLevelBytes += levelSize
1689 1 : continue
1690 : }
1691 1 : higherLevelBytes += levelSize
1692 : }
1693 1 : return float64(lowestLevelBytes) / float64(higherLevelBytes)
1694 : }
1695 :
1696 : // WriteAmpHeuristic defines a multi level compaction heuristic which will add
1697 : // an additional level to the picked compaction if it reduces predicted write
1698 : // amp of the compaction + the addPropensity constant.
1699 : type WriteAmpHeuristic struct {
1700 : // addPropensity is a constant that affects the propensity to conduct multilevel
1701 : // compactions. If positive, a multilevel compaction may get picked even if
1702 : // the single level compaction has lower write amp, and vice versa.
1703 : AddPropensity float64
1704 :
1705 : // AllowL0 if true, allow l0 to be involved in a ML compaction.
1706 : AllowL0 bool
1707 : }
1708 :
1709 : var _ MultiLevelHeuristic = (*WriteAmpHeuristic)(nil)
1710 :
1711 : // TODO(msbutler): microbenchmark the extent to which multilevel compaction
1712 : // picking slows down the compaction picking process. This should be as fast as
1713 : // possible since Compaction-picking holds d.mu, which prevents WAL rotations,
1714 : // in-progress flushes and compactions from completing, etc. Consider ways to
1715 : // deduplicate work, given that setupInputs has already been called.
1716 : func (wa WriteAmpHeuristic) pick(
1717 : pcOrig *pickedCompaction, opts *Options, diskAvailBytes uint64,
1718 1 : ) *pickedCompaction {
1719 1 : pcMulti := pcOrig.clone()
1720 1 : if !pcMulti.setupMultiLevelCandidate(opts, diskAvailBytes) {
1721 1 : return pcOrig
1722 1 : }
1723 : // We consider the addition of a level as an "expansion" of the compaction.
1724 : // If pcMulti is past the expanded compaction byte size limit already,
1725 : // we don't consider it.
1726 1 : if pcMulti.compactionSize() >= expandedCompactionByteSizeLimit(
1727 1 : opts, adjustedOutputLevel(pcMulti.outputLevel.level, pcMulti.baseLevel), diskAvailBytes) {
1728 1 : return pcOrig
1729 1 : }
1730 1 : picked := pcOrig
1731 1 : if pcMulti.predictedWriteAmp() <= pcOrig.predictedWriteAmp()+wa.AddPropensity {
1732 1 : picked = pcMulti
1733 1 : }
1734 : // Regardless of what compaction was picked, log the multilevelOverlapping ratio.
1735 1 : picked.pickerMetrics.multiLevelOverlappingRatio = pcMulti.overlappingRatio()
1736 1 : return picked
1737 : }
1738 :
1739 1 : func (wa WriteAmpHeuristic) allowL0() bool {
1740 1 : return wa.AllowL0
1741 1 : }
1742 :
1743 : // String implements fmt.Stringer.
1744 1 : func (wa WriteAmpHeuristic) String() string {
1745 1 : return fmt.Sprintf("wamp(%.2f, %t)", wa.AddPropensity, wa.AllowL0)
1746 1 : }
1747 :
1748 : // Helper method to pick compactions originating from L0. Uses information about
1749 : // sublevels to generate a compaction.
1750 1 : func pickL0(env compactionEnv, opts *Options, vers *version, baseLevel int) (pc *pickedCompaction) {
1751 1 : // It is important to pass information about Lbase files to L0Sublevels
1752 1 : // so it can pick a compaction that does not conflict with an Lbase => Lbase+1
1753 1 : // compaction. Without this, we observed reduced concurrency of L0=>Lbase
1754 1 : // compactions, and increasing read amplification in L0.
1755 1 : //
1756 1 : // TODO(bilal) Remove the minCompactionDepth parameter once fixing it at 1
1757 1 : // has been shown to not cause a performance regression.
1758 1 : lcf, err := vers.L0Sublevels.PickBaseCompaction(1, vers.Levels[baseLevel].Slice())
1759 1 : if err != nil {
1760 1 : opts.Logger.Errorf("error when picking base compaction: %s", err)
1761 1 : return
1762 1 : }
1763 1 : if lcf != nil {
1764 1 : pc = newPickedCompactionFromL0(lcf, opts, vers, baseLevel, true)
1765 1 : pc.setupInputs(opts, env.diskAvailBytes, pc.startLevel)
1766 1 : if pc.startLevel.files.Empty() {
1767 0 : opts.Logger.Fatalf("empty compaction chosen")
1768 0 : }
1769 1 : return pc.maybeAddLevel(opts, env.diskAvailBytes)
1770 : }
1771 :
1772 : // Couldn't choose a base compaction. Try choosing an intra-L0
1773 : // compaction. Note that we pass in L0CompactionThreshold here as opposed to
1774 : // 1, since choosing a single sublevel intra-L0 compaction is
1775 : // counterproductive.
1776 1 : lcf, err = vers.L0Sublevels.PickIntraL0Compaction(env.earliestUnflushedSeqNum, minIntraL0Count)
1777 1 : if err != nil {
1778 0 : opts.Logger.Errorf("error when picking intra-L0 compaction: %s", err)
1779 0 : return
1780 0 : }
1781 1 : if lcf != nil {
1782 1 : pc = newPickedCompactionFromL0(lcf, opts, vers, 0, false)
1783 1 : if !pc.setupInputs(opts, env.diskAvailBytes, pc.startLevel) {
1784 0 : return nil
1785 0 : }
1786 1 : if pc.startLevel.files.Empty() {
1787 0 : opts.Logger.Fatalf("empty compaction chosen")
1788 0 : }
1789 1 : {
1790 1 : iter := pc.startLevel.files.Iter()
1791 1 : if iter.First() == nil || iter.Next() == nil {
1792 0 : // A single-file intra-L0 compaction is unproductive.
1793 0 : return nil
1794 0 : }
1795 : }
1796 :
1797 1 : pc.smallest, pc.largest = manifest.KeyRange(pc.cmp, pc.startLevel.files.Iter())
1798 : }
1799 1 : return pc
1800 : }
1801 :
1802 : func pickManualCompaction(
1803 : vers *version, opts *Options, env compactionEnv, baseLevel int, manual *manualCompaction,
1804 1 : ) (pc *pickedCompaction, retryLater bool) {
1805 1 : outputLevel := manual.level + 1
1806 1 : if manual.level == 0 {
1807 1 : outputLevel = baseLevel
1808 1 : } else if manual.level < baseLevel {
1809 0 : // The start level for a compaction must be >= Lbase. A manual
1810 0 : // compaction could have been created adhering to that condition, and
1811 0 : // then an automatic compaction came in and compacted all of the
1812 0 : // sstables in Lbase to Lbase+1 which caused Lbase to change. Simply
1813 0 : // ignore this manual compaction as there is nothing to do (manual.level
1814 0 : // points to an empty level).
1815 0 : return nil, false
1816 0 : }
1817 : // This conflictsWithInProgress call is necessary for the manual compaction to
1818 : // be retried when it conflicts with an ongoing automatic compaction. Without
1819 : // it, the compaction is dropped due to pc.setupInputs returning false since
1820 : // the input/output range is already being compacted, and the manual
1821 : // compaction ends with a non-compacted LSM.
1822 1 : if conflictsWithInProgress(manual, outputLevel, env.inProgressCompactions, opts.Comparer.Compare) {
1823 1 : return nil, true
1824 1 : }
1825 1 : pc = newPickedCompaction(opts, vers, manual.level, defaultOutputLevel(manual.level, baseLevel), baseLevel)
1826 1 : manual.outputLevel = pc.outputLevel.level
1827 1 : pc.startLevel.files = vers.Overlaps(manual.level, base.UserKeyBoundsInclusive(manual.start, manual.end))
1828 1 : if pc.startLevel.files.Empty() {
1829 1 : // Nothing to do
1830 1 : return nil, false
1831 1 : }
1832 1 : if !pc.setupInputs(opts, env.diskAvailBytes, pc.startLevel) {
1833 0 : // setupInputs returned false indicating there's a conflicting
1834 0 : // concurrent compaction.
1835 0 : return nil, true
1836 0 : }
1837 1 : if pc = pc.maybeAddLevel(opts, env.diskAvailBytes); pc == nil {
1838 0 : return nil, false
1839 0 : }
1840 1 : if pc.outputLevel.level != outputLevel {
1841 1 : if len(pc.extraLevels) > 0 {
1842 1 : // multilevel compactions relax this invariant
1843 1 : } else {
1844 0 : panic("pebble: compaction picked unexpected output level")
1845 : }
1846 : }
1847 : // Fail-safe to protect against compacting the same sstable concurrently.
1848 1 : if inputRangeAlreadyCompacting(env, pc) {
1849 0 : return nil, true
1850 0 : }
1851 1 : return pc, false
1852 : }
1853 :
1854 : // pickDownloadCompaction picks a download compaction for the downloadSpan,
1855 : // which could be specified as being performed either by a copy compaction of
1856 : // the backing file or a rewrite compaction.
1857 : func pickDownloadCompaction(
1858 : vers *version,
1859 : opts *Options,
1860 : env compactionEnv,
1861 : baseLevel int,
1862 : kind compactionKind,
1863 : level int,
1864 : file *fileMetadata,
1865 1 : ) (pc *pickedCompaction) {
1866 1 : // Check if the file is compacting already.
1867 1 : if file.CompactionState == manifest.CompactionStateCompacting {
1868 0 : return nil
1869 0 : }
1870 1 : if kind != compactionKindCopy && kind != compactionKindRewrite {
1871 0 : panic("invalid download/rewrite compaction kind")
1872 : }
1873 1 : pc = newPickedCompaction(opts, vers, level, level, baseLevel)
1874 1 : pc.kind = kind
1875 1 : pc.startLevel.files = manifest.NewLevelSliceKeySorted(opts.Comparer.Compare, []*fileMetadata{file})
1876 1 : if !pc.setupInputs(opts, env.diskAvailBytes, pc.startLevel) {
1877 0 : // setupInputs returned false indicating there's a conflicting
1878 0 : // concurrent compaction.
1879 0 : return nil
1880 0 : }
1881 1 : if pc.outputLevel.level != level {
1882 0 : panic("pebble: download compaction picked unexpected output level")
1883 : }
1884 : // Fail-safe to protect against compacting the same sstable concurrently.
1885 1 : if inputRangeAlreadyCompacting(env, pc) {
1886 0 : return nil
1887 0 : }
1888 1 : return pc
1889 : }
1890 :
1891 : func (p *compactionPickerByScore) pickReadTriggeredCompaction(
1892 : env compactionEnv,
1893 1 : ) (pc *pickedCompaction) {
1894 1 : // If a flush is in-progress or expected to happen soon, it means more writes are taking place. We would
1895 1 : // soon be scheduling more write focussed compactions. In this case, skip read compactions as they are
1896 1 : // lower priority.
1897 1 : if env.readCompactionEnv.flushing || env.readCompactionEnv.readCompactions == nil {
1898 1 : return nil
1899 1 : }
1900 1 : for env.readCompactionEnv.readCompactions.size > 0 {
1901 1 : rc := env.readCompactionEnv.readCompactions.remove()
1902 1 : if pc = pickReadTriggeredCompactionHelper(p, rc, env); pc != nil {
1903 1 : break
1904 : }
1905 : }
1906 1 : return pc
1907 : }
1908 :
1909 : func pickReadTriggeredCompactionHelper(
1910 : p *compactionPickerByScore, rc *readCompaction, env compactionEnv,
1911 1 : ) (pc *pickedCompaction) {
1912 1 : overlapSlice := p.vers.Overlaps(rc.level, base.UserKeyBoundsInclusive(rc.start, rc.end))
1913 1 : if overlapSlice.Empty() {
1914 1 : // If there is no overlap, then the file with the key range
1915 1 : // must have been compacted away. So, we don't proceed to
1916 1 : // compact the same key range again.
1917 1 : return nil
1918 1 : }
1919 :
1920 1 : iter := overlapSlice.Iter()
1921 1 : var fileMatches bool
1922 1 : for f := iter.First(); f != nil; f = iter.Next() {
1923 1 : if f.FileNum == rc.fileNum {
1924 1 : fileMatches = true
1925 1 : break
1926 : }
1927 : }
1928 1 : if !fileMatches {
1929 1 : return nil
1930 1 : }
1931 :
1932 1 : pc = newPickedCompaction(p.opts, p.vers, rc.level, defaultOutputLevel(rc.level, p.baseLevel), p.baseLevel)
1933 1 :
1934 1 : pc.startLevel.files = overlapSlice
1935 1 : if !pc.setupInputs(p.opts, env.diskAvailBytes, pc.startLevel) {
1936 0 : return nil
1937 0 : }
1938 1 : if inputRangeAlreadyCompacting(env, pc) {
1939 0 : return nil
1940 0 : }
1941 1 : pc.kind = compactionKindRead
1942 1 :
1943 1 : // Prevent read compactions which are too wide.
1944 1 : outputOverlaps := pc.version.Overlaps(pc.outputLevel.level, pc.userKeyBounds())
1945 1 : if outputOverlaps.SizeSum() > pc.maxReadCompactionBytes {
1946 1 : return nil
1947 1 : }
1948 :
1949 : // Prevent compactions which start with a small seed file X, but overlap
1950 : // with over allowedCompactionWidth * X file sizes in the output layer.
1951 1 : const allowedCompactionWidth = 35
1952 1 : if outputOverlaps.SizeSum() > overlapSlice.SizeSum()*allowedCompactionWidth {
1953 0 : return nil
1954 0 : }
1955 :
1956 1 : return pc
1957 : }
1958 :
1959 1 : func (p *compactionPickerByScore) forceBaseLevel1() {
1960 1 : p.baseLevel = 1
1961 1 : }
1962 :
1963 1 : func inputRangeAlreadyCompacting(env compactionEnv, pc *pickedCompaction) bool {
1964 1 : for _, cl := range pc.inputs {
1965 1 : iter := cl.files.Iter()
1966 1 : for f := iter.First(); f != nil; f = iter.Next() {
1967 1 : if f.IsCompacting() {
1968 0 : return true
1969 0 : }
1970 : }
1971 : }
1972 :
1973 : // Look for active compactions outputting to the same region of the key
1974 : // space in the same output level. Two potential compactions may conflict
1975 : // without sharing input files if there are no files in the output level
1976 : // that overlap with the intersection of the compactions' key spaces.
1977 : //
1978 : // Consider an active L0->Lbase compaction compacting two L0 files one
1979 : // [a-f] and the other [t-z] into Lbase.
1980 : //
1981 : // L0
1982 : // ↦ 000100 ↤ ↦ 000101 ↤
1983 : // L1
1984 : // ↦ 000004 ↤
1985 : // a b c d e f g h i j k l m n o p q r s t u v w x y z
1986 : //
1987 : // If a new file 000102 [j-p] is flushed while the existing compaction is
1988 : // still ongoing, new file would not be in any compacting sublevel
1989 : // intervals and would not overlap with any Lbase files that are also
1990 : // compacting. However, this compaction cannot be picked because the
1991 : // compaction's output key space [j-p] would overlap the existing
1992 : // compaction's output key space [a-z].
1993 : //
1994 : // L0
1995 : // ↦ 000100* ↤ ↦ 000102 ↤ ↦ 000101* ↤
1996 : // L1
1997 : // ↦ 000004* ↤
1998 : // a b c d e f g h i j k l m n o p q r s t u v w x y z
1999 : //
2000 : // * - currently compacting
2001 1 : if pc.outputLevel != nil && pc.outputLevel.level != 0 {
2002 1 : for _, c := range env.inProgressCompactions {
2003 1 : if pc.outputLevel.level != c.outputLevel {
2004 1 : continue
2005 : }
2006 1 : if base.InternalCompare(pc.cmp, c.largest, pc.smallest) < 0 ||
2007 1 : base.InternalCompare(pc.cmp, c.smallest, pc.largest) > 0 {
2008 1 : continue
2009 : }
2010 :
2011 : // The picked compaction and the in-progress compaction c are
2012 : // outputting to the same region of the key space of the same
2013 : // level.
2014 1 : return true
2015 : }
2016 : }
2017 1 : return false
2018 : }
2019 :
2020 : // conflictsWithInProgress checks if there are any in-progress compactions with overlapping keyspace.
2021 : func conflictsWithInProgress(
2022 : manual *manualCompaction, outputLevel int, inProgressCompactions []compactionInfo, cmp Compare,
2023 1 : ) bool {
2024 1 : for _, c := range inProgressCompactions {
2025 1 : if (c.outputLevel == manual.level || c.outputLevel == outputLevel) &&
2026 1 : isUserKeysOverlapping(manual.start, manual.end, c.smallest.UserKey, c.largest.UserKey, cmp) {
2027 1 : return true
2028 1 : }
2029 1 : for _, in := range c.inputs {
2030 1 : if in.files.Empty() {
2031 1 : continue
2032 : }
2033 0 : iter := in.files.Iter()
2034 0 : smallest := iter.First().Smallest.UserKey
2035 0 : largest := iter.Last().Largest.UserKey
2036 0 : if (in.level == manual.level || in.level == outputLevel) &&
2037 0 : isUserKeysOverlapping(manual.start, manual.end, smallest, largest, cmp) {
2038 0 : return true
2039 0 : }
2040 : }
2041 : }
2042 1 : return false
2043 : }
2044 :
2045 1 : func isUserKeysOverlapping(x1, x2, y1, y2 []byte, cmp Compare) bool {
2046 1 : return cmp(x1, y2) <= 0 && cmp(y1, x2) <= 0
2047 1 : }
|
