Line data Source code
1 : // Copyright 2013 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 : "runtime/pprof"
13 : "slices"
14 : "sort"
15 : "sync/atomic"
16 : "time"
17 :
18 : "github.com/cockroachdb/crlib/crtime"
19 : "github.com/cockroachdb/errors"
20 : "github.com/cockroachdb/pebble/internal/base"
21 : "github.com/cockroachdb/pebble/internal/compact"
22 : "github.com/cockroachdb/pebble/internal/keyspan"
23 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
24 : "github.com/cockroachdb/pebble/internal/manifest"
25 : "github.com/cockroachdb/pebble/internal/sstableinternal"
26 : "github.com/cockroachdb/pebble/objstorage"
27 : "github.com/cockroachdb/pebble/objstorage/objstorageprovider/objiotracing"
28 : "github.com/cockroachdb/pebble/objstorage/remote"
29 : "github.com/cockroachdb/pebble/sstable"
30 : "github.com/cockroachdb/pebble/vfs"
31 : )
32 :
33 : var errEmptyTable = errors.New("pebble: empty table")
34 :
35 : // ErrCancelledCompaction is returned if a compaction is cancelled by a
36 : // concurrent excise or ingest-split operation.
37 : var ErrCancelledCompaction = errors.New("pebble: compaction cancelled by a concurrent operation, will retry compaction")
38 :
39 : var compactLabels = pprof.Labels("pebble", "compact")
40 : var flushLabels = pprof.Labels("pebble", "flush")
41 : var gcLabels = pprof.Labels("pebble", "gc")
42 :
43 : // expandedCompactionByteSizeLimit is the maximum number of bytes in all
44 : // compacted files. We avoid expanding the lower level file set of a compaction
45 : // if it would make the total compaction cover more than this many bytes.
46 1 : func expandedCompactionByteSizeLimit(opts *Options, level int, availBytes uint64) uint64 {
47 1 : v := uint64(25 * opts.Level(level).TargetFileSize)
48 1 :
49 1 : // Never expand a compaction beyond half the available capacity, divided
50 1 : // by the maximum number of concurrent compactions. Each of the concurrent
51 1 : // compactions may expand up to this limit, so this attempts to limit
52 1 : // compactions to half of available disk space. Note that this will not
53 1 : // prevent compaction picking from pursuing compactions that are larger
54 1 : // than this threshold before expansion.
55 1 : diskMax := (availBytes / 2) / uint64(opts.MaxConcurrentCompactions())
56 1 : if v > diskMax {
57 1 : v = diskMax
58 1 : }
59 1 : return v
60 : }
61 :
62 : // maxGrandparentOverlapBytes is the maximum bytes of overlap with level+1
63 : // before we stop building a single file in a level-1 to level compaction.
64 1 : func maxGrandparentOverlapBytes(opts *Options, level int) uint64 {
65 1 : return uint64(10 * opts.Level(level).TargetFileSize)
66 1 : }
67 :
68 : // maxReadCompactionBytes is used to prevent read compactions which
69 : // are too wide.
70 1 : func maxReadCompactionBytes(opts *Options, level int) uint64 {
71 1 : return uint64(10 * opts.Level(level).TargetFileSize)
72 1 : }
73 :
74 : // noCloseIter wraps around a FragmentIterator, intercepting and eliding
75 : // calls to Close. It is used during compaction to ensure that rangeDelIters
76 : // are not closed prematurely.
77 : type noCloseIter struct {
78 : keyspan.FragmentIterator
79 : }
80 :
81 1 : func (i *noCloseIter) Close() {}
82 :
83 : type compactionLevel struct {
84 : level int
85 : files manifest.LevelSlice
86 : // l0SublevelInfo contains information about L0 sublevels being compacted.
87 : // It's only set for the start level of a compaction starting out of L0 and
88 : // is nil for all other compactions.
89 : l0SublevelInfo []sublevelInfo
90 : }
91 :
92 1 : func (cl compactionLevel) Clone() compactionLevel {
93 1 : newCL := compactionLevel{
94 1 : level: cl.level,
95 1 : files: cl.files,
96 1 : }
97 1 : return newCL
98 1 : }
99 1 : func (cl compactionLevel) String() string {
100 1 : return fmt.Sprintf(`Level %d, Files %s`, cl.level, cl.files)
101 1 : }
102 :
103 : // compactionWritable is a objstorage.Writable wrapper that, on every write,
104 : // updates a metric in `versions` on bytes written by in-progress compactions so
105 : // far. It also increments a per-compaction `written` int.
106 : type compactionWritable struct {
107 : objstorage.Writable
108 :
109 : versions *versionSet
110 : written *int64
111 : }
112 :
113 : // Write is part of the objstorage.Writable interface.
114 1 : func (c *compactionWritable) Write(p []byte) error {
115 1 : if err := c.Writable.Write(p); err != nil {
116 0 : return err
117 0 : }
118 :
119 1 : *c.written += int64(len(p))
120 1 : c.versions.incrementCompactionBytes(int64(len(p)))
121 1 : return nil
122 : }
123 :
124 : type compactionKind int
125 :
126 : const (
127 : compactionKindDefault compactionKind = iota
128 : compactionKindFlush
129 : // compactionKindMove denotes a move compaction where the input file is
130 : // retained and linked in a new level without being obsoleted.
131 : compactionKindMove
132 : // compactionKindCopy denotes a copy compaction where the input file is
133 : // copied byte-by-byte into a new file with a new FileNum in the output level.
134 : compactionKindCopy
135 : // compactionKindDeleteOnly denotes a compaction that only deletes input
136 : // files. It can occur when wide range tombstones completely contain sstables.
137 : compactionKindDeleteOnly
138 : compactionKindElisionOnly
139 : compactionKindRead
140 : compactionKindTombstoneDensity
141 : compactionKindRewrite
142 : compactionKindIngestedFlushable
143 : )
144 :
145 1 : func (k compactionKind) String() string {
146 1 : switch k {
147 1 : case compactionKindDefault:
148 1 : return "default"
149 0 : case compactionKindFlush:
150 0 : return "flush"
151 1 : case compactionKindMove:
152 1 : return "move"
153 1 : case compactionKindDeleteOnly:
154 1 : return "delete-only"
155 1 : case compactionKindElisionOnly:
156 1 : return "elision-only"
157 1 : case compactionKindRead:
158 1 : return "read"
159 0 : case compactionKindTombstoneDensity:
160 0 : return "tombstone-density"
161 1 : case compactionKindRewrite:
162 1 : return "rewrite"
163 0 : case compactionKindIngestedFlushable:
164 0 : return "ingested-flushable"
165 1 : case compactionKindCopy:
166 1 : return "copy"
167 : }
168 0 : return "?"
169 : }
170 :
171 : // compaction is a table compaction from one level to the next, starting from a
172 : // given version.
173 : type compaction struct {
174 : // cancel is a bool that can be used by other goroutines to signal a compaction
175 : // to cancel, such as if a conflicting excise operation raced it to manifest
176 : // application. Only holders of the manifest lock will write to this atomic.
177 : cancel atomic.Bool
178 :
179 : kind compactionKind
180 : // isDownload is true if this compaction was started as part of a Download
181 : // operation. In this case kind is compactionKindCopy or
182 : // compactionKindRewrite.
183 : isDownload bool
184 :
185 : cmp Compare
186 : equal Equal
187 : comparer *base.Comparer
188 : formatKey base.FormatKey
189 : logger Logger
190 : version *version
191 : stats base.InternalIteratorStats
192 : beganAt time.Time
193 : // versionEditApplied is set to true when a compaction has completed and the
194 : // resulting version has been installed (if successful), but the compaction
195 : // goroutine is still cleaning up (eg, deleting obsolete files).
196 : versionEditApplied bool
197 : bufferPool sstable.BufferPool
198 :
199 : // startLevel is the level that is being compacted. Inputs from startLevel
200 : // and outputLevel will be merged to produce a set of outputLevel files.
201 : startLevel *compactionLevel
202 :
203 : // outputLevel is the level that files are being produced in. outputLevel is
204 : // equal to startLevel+1 except when:
205 : // - if startLevel is 0, the output level equals compactionPicker.baseLevel().
206 : // - in multilevel compaction, the output level is the lowest level involved in
207 : // the compaction
208 : // A compaction's outputLevel is nil for delete-only compactions.
209 : outputLevel *compactionLevel
210 :
211 : // extraLevels point to additional levels in between the input and output
212 : // levels that get compacted in multilevel compactions
213 : extraLevels []*compactionLevel
214 :
215 : inputs []compactionLevel
216 :
217 : // maxOutputFileSize is the maximum size of an individual table created
218 : // during compaction.
219 : maxOutputFileSize uint64
220 : // maxOverlapBytes is the maximum number of bytes of overlap allowed for a
221 : // single output table with the tables in the grandparent level.
222 : maxOverlapBytes uint64
223 :
224 : // flushing contains the flushables (aka memtables) that are being flushed.
225 : flushing flushableList
226 : // bytesWritten contains the number of bytes that have been written to outputs.
227 : bytesWritten int64
228 :
229 : // The boundaries of the input data.
230 : smallest InternalKey
231 : largest InternalKey
232 :
233 : // A list of fragment iterators to close when the compaction finishes. Used by
234 : // input iteration to keep rangeDelIters open for the lifetime of the
235 : // compaction, and only close them when the compaction finishes.
236 : closers []*noCloseIter
237 :
238 : // grandparents are the tables in level+2 that overlap with the files being
239 : // compacted. Used to determine output table boundaries. Do not assume that the actual files
240 : // in the grandparent when this compaction finishes will be the same.
241 : grandparents manifest.LevelSlice
242 :
243 : // Boundaries at which flushes to L0 should be split. Determined by
244 : // L0Sublevels. If nil, flushes aren't split.
245 : l0Limits [][]byte
246 :
247 : delElision compact.TombstoneElision
248 : rangeKeyElision compact.TombstoneElision
249 :
250 : // allowedZeroSeqNum is true if seqnums can be zeroed if there are no
251 : // snapshots requiring them to be kept. This determination is made by
252 : // looking for an sstable which overlaps the bounds of the compaction at a
253 : // lower level in the LSM during runCompaction.
254 : allowedZeroSeqNum bool
255 :
256 : // deletionHints are set if this is a compactionKindDeleteOnly. Used to figure
257 : // out whether an input must be deleted in its entirety, or excised into
258 : // virtual sstables.
259 : deletionHints []deleteCompactionHint
260 :
261 : // exciseEnabled is set to true if this is a compactionKindDeleteOnly and
262 : // this compaction is allowed to excise files.
263 : exciseEnabled bool
264 :
265 : metrics map[int]*LevelMetrics
266 :
267 : pickerMetrics compactionPickerMetrics
268 :
269 : slot base.CompactionSlot
270 : }
271 :
272 : // inputLargestSeqNumAbsolute returns the maximum LargestSeqNumAbsolute of any
273 : // input sstables.
274 1 : func (c *compaction) inputLargestSeqNumAbsolute() base.SeqNum {
275 1 : var seqNum base.SeqNum
276 1 : for _, cl := range c.inputs {
277 1 : cl.files.Each(func(m *manifest.FileMetadata) {
278 1 : seqNum = max(seqNum, m.LargestSeqNumAbsolute)
279 1 : })
280 : }
281 1 : return seqNum
282 : }
283 :
284 1 : func (c *compaction) makeInfo(jobID JobID) CompactionInfo {
285 1 : info := CompactionInfo{
286 1 : JobID: int(jobID),
287 1 : Reason: c.kind.String(),
288 1 : Input: make([]LevelInfo, 0, len(c.inputs)),
289 1 : Annotations: []string{},
290 1 : }
291 1 : if c.isDownload {
292 1 : info.Reason = "download," + info.Reason
293 1 : }
294 1 : for _, cl := range c.inputs {
295 1 : inputInfo := LevelInfo{Level: cl.level, Tables: nil}
296 1 : iter := cl.files.Iter()
297 1 : for m := iter.First(); m != nil; m = iter.Next() {
298 1 : inputInfo.Tables = append(inputInfo.Tables, m.TableInfo())
299 1 : }
300 1 : info.Input = append(info.Input, inputInfo)
301 : }
302 1 : if c.outputLevel != nil {
303 1 : info.Output.Level = c.outputLevel.level
304 1 :
305 1 : // If there are no inputs from the output level (eg, a move
306 1 : // compaction), add an empty LevelInfo to info.Input.
307 1 : if len(c.inputs) > 0 && c.inputs[len(c.inputs)-1].level != c.outputLevel.level {
308 0 : info.Input = append(info.Input, LevelInfo{Level: c.outputLevel.level})
309 0 : }
310 1 : } else {
311 1 : // For a delete-only compaction, set the output level to L6. The
312 1 : // output level is not meaningful here, but complicating the
313 1 : // info.Output interface with a pointer doesn't seem worth the
314 1 : // semantic distinction.
315 1 : info.Output.Level = numLevels - 1
316 1 : }
317 :
318 1 : for i, score := range c.pickerMetrics.scores {
319 1 : info.Input[i].Score = score
320 1 : }
321 1 : info.SingleLevelOverlappingRatio = c.pickerMetrics.singleLevelOverlappingRatio
322 1 : info.MultiLevelOverlappingRatio = c.pickerMetrics.multiLevelOverlappingRatio
323 1 : if len(info.Input) > 2 {
324 1 : info.Annotations = append(info.Annotations, "multilevel")
325 1 : }
326 1 : return info
327 : }
328 :
329 1 : func (c *compaction) userKeyBounds() base.UserKeyBounds {
330 1 : return base.UserKeyBoundsFromInternal(c.smallest, c.largest)
331 1 : }
332 :
333 : func newCompaction(
334 : pc *pickedCompaction, opts *Options, beganAt time.Time, provider objstorage.Provider,
335 1 : ) *compaction {
336 1 : c := &compaction{
337 1 : kind: compactionKindDefault,
338 1 : cmp: pc.cmp,
339 1 : equal: opts.Comparer.Equal,
340 1 : comparer: opts.Comparer,
341 1 : formatKey: opts.Comparer.FormatKey,
342 1 : inputs: pc.inputs,
343 1 : smallest: pc.smallest,
344 1 : largest: pc.largest,
345 1 : logger: opts.Logger,
346 1 : version: pc.version,
347 1 : beganAt: beganAt,
348 1 : maxOutputFileSize: pc.maxOutputFileSize,
349 1 : maxOverlapBytes: pc.maxOverlapBytes,
350 1 : pickerMetrics: pc.pickerMetrics,
351 1 : slot: pc.slot,
352 1 : }
353 1 : c.startLevel = &c.inputs[0]
354 1 : if pc.startLevel.l0SublevelInfo != nil {
355 1 : c.startLevel.l0SublevelInfo = pc.startLevel.l0SublevelInfo
356 1 : }
357 1 : c.outputLevel = &c.inputs[1]
358 1 : if c.slot == nil {
359 1 : c.slot = opts.Experimental.CompactionLimiter.TookWithoutPermission(context.TODO())
360 1 : c.slot.CompactionSelected(c.startLevel.level, c.outputLevel.level, c.startLevel.files.SizeSum())
361 1 : }
362 :
363 1 : if len(pc.extraLevels) > 0 {
364 1 : c.extraLevels = pc.extraLevels
365 1 : c.outputLevel = &c.inputs[len(c.inputs)-1]
366 1 : }
367 : // Compute the set of outputLevel+1 files that overlap this compaction (these
368 : // are the grandparent sstables).
369 1 : if c.outputLevel.level+1 < numLevels {
370 1 : c.grandparents = c.version.Overlaps(c.outputLevel.level+1, c.userKeyBounds())
371 1 : }
372 1 : c.delElision, c.rangeKeyElision = compact.SetupTombstoneElision(
373 1 : c.cmp, c.version, c.outputLevel.level, base.UserKeyBoundsFromInternal(c.smallest, c.largest),
374 1 : )
375 1 : c.kind = pc.kind
376 1 :
377 1 : if c.kind == compactionKindDefault && c.outputLevel.files.Empty() && !c.hasExtraLevelData() &&
378 1 : c.startLevel.files.Len() == 1 && c.grandparents.SizeSum() <= c.maxOverlapBytes {
379 1 : // This compaction can be converted into a move or copy from one level
380 1 : // to the next. We avoid such a move if there is lots of overlapping
381 1 : // grandparent data. Otherwise, the move could create a parent file
382 1 : // that will require a very expensive merge later on.
383 1 : iter := c.startLevel.files.Iter()
384 1 : meta := iter.First()
385 1 : isRemote := false
386 1 : // We should always be passed a provider, except in some unit tests.
387 1 : if provider != nil {
388 1 : isRemote = !objstorage.IsLocalTable(provider, meta.FileBacking.DiskFileNum)
389 1 : }
390 : // Avoid a trivial move or copy if all of these are true, as rewriting a
391 : // new file is better:
392 : //
393 : // 1) The source file is a virtual sstable
394 : // 2) The existing file `meta` is on non-remote storage
395 : // 3) The output level prefers shared storage
396 1 : mustCopy := !isRemote && remote.ShouldCreateShared(opts.Experimental.CreateOnShared, c.outputLevel.level)
397 1 : if mustCopy {
398 1 : // If the source is virtual, it's best to just rewrite the file as all
399 1 : // conditions in the above comment are met.
400 1 : if !meta.Virtual {
401 1 : c.kind = compactionKindCopy
402 1 : }
403 1 : } else {
404 1 : c.kind = compactionKindMove
405 1 : }
406 : }
407 1 : return c
408 : }
409 :
410 : func newDeleteOnlyCompaction(
411 : opts *Options,
412 : cur *version,
413 : inputs []compactionLevel,
414 : beganAt time.Time,
415 : hints []deleteCompactionHint,
416 : exciseEnabled bool,
417 1 : ) *compaction {
418 1 : c := &compaction{
419 1 : kind: compactionKindDeleteOnly,
420 1 : cmp: opts.Comparer.Compare,
421 1 : equal: opts.Comparer.Equal,
422 1 : comparer: opts.Comparer,
423 1 : formatKey: opts.Comparer.FormatKey,
424 1 : logger: opts.Logger,
425 1 : version: cur,
426 1 : beganAt: beganAt,
427 1 : inputs: inputs,
428 1 : deletionHints: hints,
429 1 : exciseEnabled: exciseEnabled,
430 1 : }
431 1 :
432 1 : // Set c.smallest, c.largest.
433 1 : files := make([]manifest.LevelIterator, 0, len(inputs))
434 1 : for _, in := range inputs {
435 1 : files = append(files, in.files.Iter())
436 1 : }
437 1 : c.smallest, c.largest = manifest.KeyRange(opts.Comparer.Compare, files...)
438 1 : return c
439 : }
440 :
441 1 : func adjustGrandparentOverlapBytesForFlush(c *compaction, flushingBytes uint64) {
442 1 : // Heuristic to place a lower bound on compaction output file size
443 1 : // caused by Lbase. Prior to this heuristic we have observed an L0 in
444 1 : // production with 310K files of which 290K files were < 10KB in size.
445 1 : // Our hypothesis is that it was caused by L1 having 2600 files and
446 1 : // ~10GB, such that each flush got split into many tiny files due to
447 1 : // overlapping with most of the files in Lbase.
448 1 : //
449 1 : // The computation below is general in that it accounts
450 1 : // for flushing different volumes of data (e.g. we may be flushing
451 1 : // many memtables). For illustration, we consider the typical
452 1 : // example of flushing a 64MB memtable. So 12.8MB output,
453 1 : // based on the compression guess below. If the compressed bytes
454 1 : // guess is an over-estimate we will end up with smaller files,
455 1 : // and if an under-estimate we will end up with larger files.
456 1 : // With a 2MB target file size, 7 files. We are willing to accept
457 1 : // 4x the number of files, if it results in better write amplification
458 1 : // when later compacting to Lbase, i.e., ~450KB files (target file
459 1 : // size / 4).
460 1 : //
461 1 : // Note that this is a pessimistic heuristic in that
462 1 : // fileCountUpperBoundDueToGrandparents could be far from the actual
463 1 : // number of files produced due to the grandparent limits. For
464 1 : // example, in the extreme, consider a flush that overlaps with 1000
465 1 : // files in Lbase f0...f999, and the initially calculated value of
466 1 : // maxOverlapBytes will cause splits at f10, f20,..., f990, which
467 1 : // means an upper bound file count of 100 files. Say the input bytes
468 1 : // in the flush are such that acceptableFileCount=10. We will fatten
469 1 : // up maxOverlapBytes by 10x to ensure that the upper bound file count
470 1 : // drops to 10. However, it is possible that in practice, even without
471 1 : // this change, we would have produced no more than 10 files, and that
472 1 : // this change makes the files unnecessarily wide. Say the input bytes
473 1 : // are distributed such that 10% are in f0...f9, 10% in f10...f19, ...
474 1 : // 10% in f80...f89 and 10% in f990...f999. The original value of
475 1 : // maxOverlapBytes would have actually produced only 10 sstables. But
476 1 : // by increasing maxOverlapBytes by 10x, we may produce 1 sstable that
477 1 : // spans f0...f89, i.e., a much wider sstable than necessary.
478 1 : //
479 1 : // We could produce a tighter estimate of
480 1 : // fileCountUpperBoundDueToGrandparents if we had knowledge of the key
481 1 : // distribution of the flush. The 4x multiplier mentioned earlier is
482 1 : // a way to try to compensate for this pessimism.
483 1 : //
484 1 : // TODO(sumeer): we don't have compression info for the data being
485 1 : // flushed, but it is likely that existing files that overlap with
486 1 : // this flush in Lbase are representative wrt compression ratio. We
487 1 : // could store the uncompressed size in FileMetadata and estimate
488 1 : // the compression ratio.
489 1 : const approxCompressionRatio = 0.2
490 1 : approxOutputBytes := approxCompressionRatio * float64(flushingBytes)
491 1 : approxNumFilesBasedOnTargetSize :=
492 1 : int(math.Ceil(approxOutputBytes / float64(c.maxOutputFileSize)))
493 1 : acceptableFileCount := float64(4 * approxNumFilesBasedOnTargetSize)
494 1 : // The byte calculation is linear in numGrandparentFiles, but we will
495 1 : // incur this linear cost in compact.Runner.TableSplitLimit() too, so we are
496 1 : // also willing to pay it now. We could approximate this cheaply by using the
497 1 : // mean file size of Lbase.
498 1 : grandparentFileBytes := c.grandparents.SizeSum()
499 1 : fileCountUpperBoundDueToGrandparents :=
500 1 : float64(grandparentFileBytes) / float64(c.maxOverlapBytes)
501 1 : if fileCountUpperBoundDueToGrandparents > acceptableFileCount {
502 1 : c.maxOverlapBytes = uint64(
503 1 : float64(c.maxOverlapBytes) *
504 1 : (fileCountUpperBoundDueToGrandparents / acceptableFileCount))
505 1 : }
506 : }
507 :
508 : func newFlush(
509 : opts *Options, cur *version, baseLevel int, flushing flushableList, beganAt time.Time,
510 1 : ) (*compaction, error) {
511 1 : c := &compaction{
512 1 : kind: compactionKindFlush,
513 1 : cmp: opts.Comparer.Compare,
514 1 : equal: opts.Comparer.Equal,
515 1 : comparer: opts.Comparer,
516 1 : formatKey: opts.Comparer.FormatKey,
517 1 : logger: opts.Logger,
518 1 : version: cur,
519 1 : beganAt: beganAt,
520 1 : inputs: []compactionLevel{{level: -1}, {level: 0}},
521 1 : maxOutputFileSize: math.MaxUint64,
522 1 : maxOverlapBytes: math.MaxUint64,
523 1 : flushing: flushing,
524 1 : }
525 1 : c.startLevel = &c.inputs[0]
526 1 : c.outputLevel = &c.inputs[1]
527 1 :
528 1 : // Flush slots are always taken without permission.
529 1 : //
530 1 : // NB: CompactionLimiter defaults to a no-op limiter unless one is implemented
531 1 : // and passed-in as an option during Open.
532 1 : slot := opts.Experimental.CompactionLimiter.TookWithoutPermission(context.TODO())
533 1 : var flushingSize uint64
534 1 : for i := range flushing {
535 1 : flushingSize += flushing[i].totalBytes()
536 1 : }
537 1 : slot.CompactionSelected(-1, 0, flushingSize)
538 1 : c.slot = slot
539 1 :
540 1 : if len(flushing) > 0 {
541 1 : if _, ok := flushing[0].flushable.(*ingestedFlushable); ok {
542 1 : if len(flushing) != 1 {
543 0 : panic("pebble: ingestedFlushable must be flushed one at a time.")
544 : }
545 1 : c.kind = compactionKindIngestedFlushable
546 1 : return c, nil
547 : }
548 : }
549 :
550 : // Make sure there's no ingestedFlushable after the first flushable in the
551 : // list.
552 1 : for _, f := range flushing {
553 1 : if _, ok := f.flushable.(*ingestedFlushable); ok {
554 0 : panic("pebble: flushing shouldn't contain ingestedFlushable flushable")
555 : }
556 : }
557 :
558 1 : if cur.L0Sublevels != nil {
559 1 : c.l0Limits = cur.L0Sublevels.FlushSplitKeys()
560 1 : }
561 :
562 1 : smallestSet, largestSet := false, false
563 1 : updatePointBounds := func(iter internalIterator) {
564 1 : if kv := iter.First(); kv != nil {
565 1 : if !smallestSet ||
566 1 : base.InternalCompare(c.cmp, c.smallest, kv.K) > 0 {
567 1 : smallestSet = true
568 1 : c.smallest = kv.K.Clone()
569 1 : }
570 : }
571 1 : if kv := iter.Last(); kv != nil {
572 1 : if !largestSet ||
573 1 : base.InternalCompare(c.cmp, c.largest, kv.K) < 0 {
574 1 : largestSet = true
575 1 : c.largest = kv.K.Clone()
576 1 : }
577 : }
578 : }
579 :
580 1 : updateRangeBounds := func(iter keyspan.FragmentIterator) error {
581 1 : // File bounds require s != nil && !s.Empty(). We only need to check for
582 1 : // s != nil here, as the memtable's FragmentIterator would never surface
583 1 : // empty spans.
584 1 : if s, err := iter.First(); err != nil {
585 0 : return err
586 1 : } else if s != nil {
587 1 : if key := s.SmallestKey(); !smallestSet ||
588 1 : base.InternalCompare(c.cmp, c.smallest, key) > 0 {
589 1 : smallestSet = true
590 1 : c.smallest = key.Clone()
591 1 : }
592 : }
593 1 : if s, err := iter.Last(); err != nil {
594 0 : return err
595 1 : } else if s != nil {
596 1 : if key := s.LargestKey(); !largestSet ||
597 1 : base.InternalCompare(c.cmp, c.largest, key) < 0 {
598 1 : largestSet = true
599 1 : c.largest = key.Clone()
600 1 : }
601 : }
602 1 : return nil
603 : }
604 :
605 1 : var flushingBytes uint64
606 1 : for i := range flushing {
607 1 : f := flushing[i]
608 1 : updatePointBounds(f.newIter(nil))
609 1 : if rangeDelIter := f.newRangeDelIter(nil); rangeDelIter != nil {
610 1 : if err := updateRangeBounds(rangeDelIter); err != nil {
611 0 : return nil, err
612 0 : }
613 : }
614 1 : if rangeKeyIter := f.newRangeKeyIter(nil); rangeKeyIter != nil {
615 1 : if err := updateRangeBounds(rangeKeyIter); err != nil {
616 0 : return nil, err
617 0 : }
618 : }
619 1 : flushingBytes += f.inuseBytes()
620 : }
621 :
622 1 : if opts.FlushSplitBytes > 0 {
623 1 : c.maxOutputFileSize = uint64(opts.Level(0).TargetFileSize)
624 1 : c.maxOverlapBytes = maxGrandparentOverlapBytes(opts, 0)
625 1 : c.grandparents = c.version.Overlaps(baseLevel, c.userKeyBounds())
626 1 : adjustGrandparentOverlapBytesForFlush(c, flushingBytes)
627 1 : }
628 :
629 : // We don't elide tombstones for flushes.
630 1 : c.delElision, c.rangeKeyElision = compact.NoTombstoneElision(), compact.NoTombstoneElision()
631 1 : return c, nil
632 : }
633 :
634 1 : func (c *compaction) hasExtraLevelData() bool {
635 1 : if len(c.extraLevels) == 0 {
636 1 : // not a multi level compaction
637 1 : return false
638 1 : } else if c.extraLevels[0].files.Empty() {
639 1 : // a multi level compaction without data in the intermediate input level;
640 1 : // e.g. for a multi level compaction with levels 4,5, and 6, this could
641 1 : // occur if there is no files to compact in 5, or in 5 and 6 (i.e. a move).
642 1 : return false
643 1 : }
644 1 : return true
645 : }
646 :
647 : // errorOnUserKeyOverlap returns an error if the last two written sstables in
648 : // this compaction have revisions of the same user key present in both sstables,
649 : // when it shouldn't (eg. when splitting flushes).
650 1 : func (c *compaction) errorOnUserKeyOverlap(ve *versionEdit) error {
651 1 : if n := len(ve.NewFiles); n > 1 {
652 1 : meta := ve.NewFiles[n-1].Meta
653 1 : prevMeta := ve.NewFiles[n-2].Meta
654 1 : if !prevMeta.Largest.IsExclusiveSentinel() &&
655 1 : c.cmp(prevMeta.Largest.UserKey, meta.Smallest.UserKey) >= 0 {
656 1 : return errors.Errorf("pebble: compaction split user key across two sstables: %s in %s and %s",
657 1 : prevMeta.Largest.Pretty(c.formatKey),
658 1 : prevMeta.FileNum,
659 1 : meta.FileNum)
660 1 : }
661 : }
662 1 : return nil
663 : }
664 :
665 : // allowZeroSeqNum returns true if seqnum's can be zeroed if there are no
666 : // snapshots requiring them to be kept. It performs this determination by
667 : // looking at the TombstoneElision values which are set up based on sstables
668 : // which overlap the bounds of the compaction at a lower level in the LSM.
669 1 : func (c *compaction) allowZeroSeqNum() bool {
670 1 : // TODO(peter): we disable zeroing of seqnums during flushing to match
671 1 : // RocksDB behavior and to avoid generating overlapping sstables during
672 1 : // DB.replayWAL. When replaying WAL files at startup, we flush after each
673 1 : // WAL is replayed building up a single version edit that is
674 1 : // applied. Because we don't apply the version edit after each flush, this
675 1 : // code doesn't know that L0 contains files and zeroing of seqnums should
676 1 : // be disabled. That is fixable, but it seems safer to just match the
677 1 : // RocksDB behavior for now.
678 1 : return len(c.flushing) == 0 && c.delElision.ElidesEverything() && c.rangeKeyElision.ElidesEverything()
679 1 : }
680 :
681 : // newInputIters returns an iterator over all the input tables in a compaction.
682 : func (c *compaction) newInputIters(
683 : newIters tableNewIters, newRangeKeyIter keyspanimpl.TableNewSpanIter,
684 : ) (
685 : pointIter internalIterator,
686 : rangeDelIter, rangeKeyIter keyspan.FragmentIterator,
687 : retErr error,
688 1 : ) {
689 1 : // Validate the ordering of compaction input files for defense in depth.
690 1 : if len(c.flushing) == 0 {
691 1 : if c.startLevel.level >= 0 {
692 1 : err := manifest.CheckOrdering(c.cmp, c.formatKey,
693 1 : manifest.Level(c.startLevel.level), c.startLevel.files.Iter())
694 1 : if err != nil {
695 1 : return nil, nil, nil, err
696 1 : }
697 : }
698 1 : err := manifest.CheckOrdering(c.cmp, c.formatKey,
699 1 : manifest.Level(c.outputLevel.level), c.outputLevel.files.Iter())
700 1 : if err != nil {
701 1 : return nil, nil, nil, err
702 1 : }
703 1 : if c.startLevel.level == 0 {
704 1 : if c.startLevel.l0SublevelInfo == nil {
705 0 : panic("l0SublevelInfo not created for compaction out of L0")
706 : }
707 1 : for _, info := range c.startLevel.l0SublevelInfo {
708 1 : err := manifest.CheckOrdering(c.cmp, c.formatKey,
709 1 : info.sublevel, info.Iter())
710 1 : if err != nil {
711 1 : return nil, nil, nil, err
712 1 : }
713 : }
714 : }
715 1 : if len(c.extraLevels) > 0 {
716 1 : if len(c.extraLevels) > 1 {
717 0 : panic("n>2 multi level compaction not implemented yet")
718 : }
719 1 : interLevel := c.extraLevels[0]
720 1 : err := manifest.CheckOrdering(c.cmp, c.formatKey,
721 1 : manifest.Level(interLevel.level), interLevel.files.Iter())
722 1 : if err != nil {
723 0 : return nil, nil, nil, err
724 0 : }
725 : }
726 : }
727 :
728 : // There are three classes of keys that a compaction needs to process: point
729 : // keys, range deletion tombstones and range keys. Collect all iterators for
730 : // all these classes of keys from all the levels. We'll aggregate them
731 : // together farther below.
732 : //
733 : // numInputLevels is an approximation of the number of iterator levels. Due
734 : // to idiosyncrasies in iterator construction, we may (rarely) exceed this
735 : // initial capacity.
736 1 : numInputLevels := max(len(c.flushing), len(c.inputs))
737 1 : iters := make([]internalIterator, 0, numInputLevels)
738 1 : rangeDelIters := make([]keyspan.FragmentIterator, 0, numInputLevels)
739 1 : rangeKeyIters := make([]keyspan.FragmentIterator, 0, numInputLevels)
740 1 :
741 1 : // If construction of the iterator inputs fails, ensure that we close all
742 1 : // the consitutent iterators.
743 1 : defer func() {
744 1 : if retErr != nil {
745 0 : for _, iter := range iters {
746 0 : if iter != nil {
747 0 : iter.Close()
748 0 : }
749 : }
750 0 : for _, rangeDelIter := range rangeDelIters {
751 0 : rangeDelIter.Close()
752 0 : }
753 : }
754 : }()
755 1 : iterOpts := IterOptions{
756 1 : Category: categoryCompaction,
757 1 : logger: c.logger,
758 1 : }
759 1 :
760 1 : // Populate iters, rangeDelIters and rangeKeyIters with the appropriate
761 1 : // constituent iterators. This depends on whether this is a flush or a
762 1 : // compaction.
763 1 : if len(c.flushing) != 0 {
764 1 : // If flushing, we need to build the input iterators over the memtables
765 1 : // stored in c.flushing.
766 1 : for i := range c.flushing {
767 1 : f := c.flushing[i]
768 1 : iters = append(iters, f.newFlushIter(nil))
769 1 : rangeDelIter := f.newRangeDelIter(nil)
770 1 : if rangeDelIter != nil {
771 1 : rangeDelIters = append(rangeDelIters, rangeDelIter)
772 1 : }
773 1 : if rangeKeyIter := f.newRangeKeyIter(nil); rangeKeyIter != nil {
774 1 : rangeKeyIters = append(rangeKeyIters, rangeKeyIter)
775 1 : }
776 : }
777 1 : } else {
778 1 : addItersForLevel := func(level *compactionLevel, l manifest.Layer) error {
779 1 : // Add a *levelIter for point iterators. Because we don't call
780 1 : // initRangeDel, the levelIter will close and forget the range
781 1 : // deletion iterator when it steps on to a new file. Surfacing range
782 1 : // deletions to compactions are handled below.
783 1 : iters = append(iters, newLevelIter(context.Background(),
784 1 : iterOpts, c.comparer, newIters, level.files.Iter(), l, internalIterOpts{
785 1 : compaction: true,
786 1 : bufferPool: &c.bufferPool,
787 1 : stats: &c.stats,
788 1 : }))
789 1 : // TODO(jackson): Use keyspanimpl.LevelIter to avoid loading all the range
790 1 : // deletions into memory upfront. (See #2015, which reverted this.) There
791 1 : // will be no user keys that are split between sstables within a level in
792 1 : // Cockroach 23.1, which unblocks this optimization.
793 1 :
794 1 : // Add the range deletion iterator for each file as an independent level
795 1 : // in mergingIter, as opposed to making a levelIter out of those. This
796 1 : // is safer as levelIter expects all keys coming from underlying
797 1 : // iterators to be in order. Due to compaction / tombstone writing
798 1 : // logic in finishOutput(), it is possible for range tombstones to not
799 1 : // be strictly ordered across all files in one level.
800 1 : //
801 1 : // Consider this example from the metamorphic tests (also repeated in
802 1 : // finishOutput()), consisting of three L3 files with their bounds
803 1 : // specified in square brackets next to the file name:
804 1 : //
805 1 : // ./000240.sst [tmgc#391,MERGE-tmgc#391,MERGE]
806 1 : // tmgc#391,MERGE [786e627a]
807 1 : // tmgc-udkatvs#331,RANGEDEL
808 1 : //
809 1 : // ./000241.sst [tmgc#384,MERGE-tmgc#384,MERGE]
810 1 : // tmgc#384,MERGE [666c7070]
811 1 : // tmgc-tvsalezade#383,RANGEDEL
812 1 : // tmgc-tvsalezade#331,RANGEDEL
813 1 : //
814 1 : // ./000242.sst [tmgc#383,RANGEDEL-tvsalezade#72057594037927935,RANGEDEL]
815 1 : // tmgc-tvsalezade#383,RANGEDEL
816 1 : // tmgc#375,SET [72646c78766965616c72776865676e79]
817 1 : // tmgc-tvsalezade#356,RANGEDEL
818 1 : //
819 1 : // Here, the range tombstone in 000240.sst falls "after" one in
820 1 : // 000241.sst, despite 000240.sst being ordered "before" 000241.sst for
821 1 : // levelIter's purposes. While each file is still consistent before its
822 1 : // bounds, it's safer to have all rangedel iterators be visible to
823 1 : // mergingIter.
824 1 : iter := level.files.Iter()
825 1 : for f := iter.First(); f != nil; f = iter.Next() {
826 1 : rangeDelIter, err := c.newRangeDelIter(newIters, iter.Take(), iterOpts, l)
827 1 : if err != nil {
828 0 : // The error will already be annotated with the BackingFileNum, so
829 0 : // we annotate it with the FileNum.
830 0 : return errors.Wrapf(err, "pebble: could not open table %s", errors.Safe(f.FileNum))
831 0 : }
832 1 : if rangeDelIter == nil {
833 1 : continue
834 : }
835 1 : rangeDelIters = append(rangeDelIters, rangeDelIter)
836 1 : c.closers = append(c.closers, rangeDelIter)
837 : }
838 :
839 : // Check if this level has any range keys.
840 1 : hasRangeKeys := false
841 1 : for f := iter.First(); f != nil; f = iter.Next() {
842 1 : if f.HasRangeKeys {
843 1 : hasRangeKeys = true
844 1 : break
845 : }
846 : }
847 1 : if hasRangeKeys {
848 1 : newRangeKeyIterWrapper := func(ctx context.Context, file *manifest.FileMetadata, iterOptions keyspan.SpanIterOptions) (keyspan.FragmentIterator, error) {
849 1 : rangeKeyIter, err := newRangeKeyIter(ctx, file, iterOptions)
850 1 : if err != nil {
851 0 : return nil, err
852 1 : } else if rangeKeyIter == nil {
853 0 : return emptyKeyspanIter, nil
854 0 : }
855 : // Ensure that the range key iter is not closed until the compaction is
856 : // finished. This is necessary because range key processing
857 : // requires the range keys to be held in memory for up to the
858 : // lifetime of the compaction.
859 1 : noCloseIter := &noCloseIter{rangeKeyIter}
860 1 : c.closers = append(c.closers, noCloseIter)
861 1 :
862 1 : // We do not need to truncate range keys to sstable boundaries, or
863 1 : // only read within the file's atomic compaction units, unlike with
864 1 : // range tombstones. This is because range keys were added after we
865 1 : // stopped splitting user keys across sstables, so all the range keys
866 1 : // in this sstable must wholly lie within the file's bounds.
867 1 : return noCloseIter, err
868 : }
869 1 : li := keyspanimpl.NewLevelIter(
870 1 : context.Background(), keyspan.SpanIterOptions{}, c.cmp,
871 1 : newRangeKeyIterWrapper, level.files.Iter(), l, manifest.KeyTypeRange,
872 1 : )
873 1 : rangeKeyIters = append(rangeKeyIters, li)
874 : }
875 1 : return nil
876 : }
877 :
878 1 : for i := range c.inputs {
879 1 : // If the level is annotated with l0SublevelInfo, expand it into one
880 1 : // level per sublevel.
881 1 : // TODO(jackson): Perform this expansion even earlier when we pick the
882 1 : // compaction?
883 1 : if len(c.inputs[i].l0SublevelInfo) > 0 {
884 1 : for _, info := range c.startLevel.l0SublevelInfo {
885 1 : sublevelCompactionLevel := &compactionLevel{0, info.LevelSlice, nil}
886 1 : if err := addItersForLevel(sublevelCompactionLevel, info.sublevel); err != nil {
887 0 : return nil, nil, nil, err
888 0 : }
889 : }
890 1 : continue
891 : }
892 1 : if err := addItersForLevel(&c.inputs[i], manifest.Level(c.inputs[i].level)); err != nil {
893 0 : return nil, nil, nil, err
894 0 : }
895 : }
896 : }
897 :
898 : // If there's only one constituent point iterator, we can avoid the overhead
899 : // of a *mergingIter. This is possible, for example, when performing a flush
900 : // of a single memtable. Otherwise, combine all the iterators into a merging
901 : // iter.
902 1 : pointIter = iters[0]
903 1 : if len(iters) > 1 {
904 1 : pointIter = newMergingIter(c.logger, &c.stats, c.cmp, nil, iters...)
905 1 : }
906 :
907 : // In normal operation, levelIter iterates over the point operations in a
908 : // level, and initializes a rangeDelIter pointer for the range deletions in
909 : // each table. During compaction, we want to iterate over the merged view of
910 : // point operations and range deletions. In order to do this we create one
911 : // levelIter per level to iterate over the point operations, and collect up
912 : // all the range deletion files.
913 : //
914 : // The range deletion levels are combined with a keyspanimpl.MergingIter. The
915 : // resulting merged rangedel iterator is then included using an
916 : // InterleavingIter.
917 : // TODO(jackson): Consider using a defragmenting iterator to stitch together
918 : // logical range deletions that were fragmented due to previous file
919 : // boundaries.
920 1 : if len(rangeDelIters) > 0 {
921 1 : mi := &keyspanimpl.MergingIter{}
922 1 : mi.Init(c.comparer, keyspan.NoopTransform, new(keyspanimpl.MergingBuffers), rangeDelIters...)
923 1 : rangeDelIter = mi
924 1 : }
925 :
926 : // If there are range key iterators, we need to combine them using
927 : // keyspanimpl.MergingIter, and then interleave them among the points.
928 1 : if len(rangeKeyIters) > 0 {
929 1 : mi := &keyspanimpl.MergingIter{}
930 1 : mi.Init(c.comparer, keyspan.NoopTransform, new(keyspanimpl.MergingBuffers), rangeKeyIters...)
931 1 : // TODO(radu): why do we have a defragmenter here but not above?
932 1 : di := &keyspan.DefragmentingIter{}
933 1 : di.Init(c.comparer, mi, keyspan.DefragmentInternal, keyspan.StaticDefragmentReducer, new(keyspan.DefragmentingBuffers))
934 1 : rangeKeyIter = di
935 1 : }
936 1 : return pointIter, rangeDelIter, rangeKeyIter, nil
937 : }
938 :
939 : func (c *compaction) newRangeDelIter(
940 : newIters tableNewIters, f manifest.LevelFile, opts IterOptions, l manifest.Layer,
941 1 : ) (*noCloseIter, error) {
942 1 : opts.layer = l
943 1 : iterSet, err := newIters(context.Background(), f.FileMetadata, &opts,
944 1 : internalIterOpts{
945 1 : compaction: true,
946 1 : bufferPool: &c.bufferPool,
947 1 : }, iterRangeDeletions)
948 1 : if err != nil {
949 0 : return nil, err
950 1 : } else if iterSet.rangeDeletion == nil {
951 1 : // The file doesn't contain any range deletions.
952 1 : return nil, nil
953 1 : }
954 : // Ensure that rangeDelIter is not closed until the compaction is
955 : // finished. This is necessary because range tombstone processing
956 : // requires the range tombstones to be held in memory for up to the
957 : // lifetime of the compaction.
958 1 : return &noCloseIter{iterSet.rangeDeletion}, nil
959 : }
960 :
961 1 : func (c *compaction) String() string {
962 1 : if len(c.flushing) != 0 {
963 0 : return "flush\n"
964 0 : }
965 :
966 1 : var buf bytes.Buffer
967 1 : for level := c.startLevel.level; level <= c.outputLevel.level; level++ {
968 1 : i := level - c.startLevel.level
969 1 : fmt.Fprintf(&buf, "%d:", level)
970 1 : iter := c.inputs[i].files.Iter()
971 1 : for f := iter.First(); f != nil; f = iter.Next() {
972 1 : fmt.Fprintf(&buf, " %s:%s-%s", f.FileNum, f.Smallest, f.Largest)
973 1 : }
974 1 : fmt.Fprintf(&buf, "\n")
975 : }
976 1 : return buf.String()
977 : }
978 :
979 : type manualCompaction struct {
980 : // Count of the retries either due to too many concurrent compactions, or a
981 : // concurrent compaction to overlapping levels.
982 : retries int
983 : level int
984 : outputLevel int
985 : done chan error
986 : start []byte
987 : end []byte
988 : split bool
989 : }
990 :
991 : type readCompaction struct {
992 : level int
993 : // [start, end] key ranges are used for de-duping.
994 : start []byte
995 : end []byte
996 :
997 : // The file associated with the compaction.
998 : // If the file no longer belongs in the same
999 : // level, then we skip the compaction.
1000 : fileNum base.FileNum
1001 : }
1002 :
1003 1 : func (d *DB) addInProgressCompaction(c *compaction) {
1004 1 : d.mu.compact.inProgress[c] = struct{}{}
1005 1 : var isBase, isIntraL0 bool
1006 1 : for _, cl := range c.inputs {
1007 1 : iter := cl.files.Iter()
1008 1 : for f := iter.First(); f != nil; f = iter.Next() {
1009 1 : if f.IsCompacting() {
1010 0 : d.opts.Logger.Fatalf("L%d->L%d: %s already being compacted", c.startLevel.level, c.outputLevel.level, f.FileNum)
1011 0 : }
1012 1 : f.SetCompactionState(manifest.CompactionStateCompacting)
1013 1 : if c.startLevel != nil && c.outputLevel != nil && c.startLevel.level == 0 {
1014 1 : if c.outputLevel.level == 0 {
1015 1 : f.IsIntraL0Compacting = true
1016 1 : isIntraL0 = true
1017 1 : } else {
1018 1 : isBase = true
1019 1 : }
1020 : }
1021 : }
1022 : }
1023 :
1024 1 : if (isIntraL0 || isBase) && c.version.L0Sublevels != nil {
1025 1 : l0Inputs := []manifest.LevelSlice{c.startLevel.files}
1026 1 : if isIntraL0 {
1027 1 : l0Inputs = append(l0Inputs, c.outputLevel.files)
1028 1 : }
1029 1 : if err := c.version.L0Sublevels.UpdateStateForStartedCompaction(l0Inputs, isBase); err != nil {
1030 0 : d.opts.Logger.Fatalf("could not update state for compaction: %s", err)
1031 0 : }
1032 : }
1033 : }
1034 :
1035 : // Removes compaction markers from files in a compaction. The rollback parameter
1036 : // indicates whether the compaction state should be rolled back to its original
1037 : // state in the case of an unsuccessful compaction.
1038 : //
1039 : // DB.mu must be held when calling this method, however this method can drop and
1040 : // re-acquire that mutex. All writes to the manifest for this compaction should
1041 : // have completed by this point.
1042 1 : func (d *DB) clearCompactingState(c *compaction, rollback bool) {
1043 1 : c.versionEditApplied = true
1044 1 : for _, cl := range c.inputs {
1045 1 : iter := cl.files.Iter()
1046 1 : for f := iter.First(); f != nil; f = iter.Next() {
1047 1 : if !f.IsCompacting() {
1048 0 : d.opts.Logger.Fatalf("L%d->L%d: %s not being compacted", c.startLevel.level, c.outputLevel.level, f.FileNum)
1049 0 : }
1050 1 : if !rollback {
1051 1 : // On success all compactions other than move and delete-only compactions
1052 1 : // transition the file into the Compacted state. Move-compacted files
1053 1 : // become eligible for compaction again and transition back to NotCompacting.
1054 1 : // Delete-only compactions could, on rare occasion, leave files untouched
1055 1 : // (eg. if files have a loose bound), so we revert them all to NotCompacting
1056 1 : // just in case they need to be compacted again.
1057 1 : if c.kind != compactionKindMove && c.kind != compactionKindDeleteOnly {
1058 1 : f.SetCompactionState(manifest.CompactionStateCompacted)
1059 1 : } else {
1060 1 : f.SetCompactionState(manifest.CompactionStateNotCompacting)
1061 1 : }
1062 1 : } else {
1063 1 : // Else, on rollback, all input files unconditionally transition back to
1064 1 : // NotCompacting.
1065 1 : f.SetCompactionState(manifest.CompactionStateNotCompacting)
1066 1 : }
1067 1 : f.IsIntraL0Compacting = false
1068 : }
1069 : }
1070 1 : l0InProgress := inProgressL0Compactions(d.getInProgressCompactionInfoLocked(c))
1071 1 : func() {
1072 1 : // InitCompactingFileInfo requires that no other manifest writes be
1073 1 : // happening in parallel with it, i.e. we're not in the midst of installing
1074 1 : // another version. Otherwise, it's possible that we've created another
1075 1 : // L0Sublevels instance, but not added it to the versions list, causing
1076 1 : // all the indices in FileMetadata to be inaccurate. To ensure this,
1077 1 : // grab the manifest lock.
1078 1 : d.mu.versions.logLock()
1079 1 : defer d.mu.versions.logUnlock()
1080 1 : d.mu.versions.currentVersion().L0Sublevels.InitCompactingFileInfo(l0InProgress)
1081 1 : }()
1082 : }
1083 :
1084 1 : func (d *DB) calculateDiskAvailableBytes() uint64 {
1085 1 : if space, err := d.opts.FS.GetDiskUsage(d.dirname); err == nil {
1086 1 : d.diskAvailBytes.Store(space.AvailBytes)
1087 1 : return space.AvailBytes
1088 1 : } else if !errors.Is(err, vfs.ErrUnsupported) {
1089 1 : d.opts.EventListener.BackgroundError(err)
1090 1 : }
1091 1 : return d.diskAvailBytes.Load()
1092 : }
1093 :
1094 : // maybeScheduleFlush schedules a flush if necessary.
1095 : //
1096 : // d.mu must be held when calling this.
1097 1 : func (d *DB) maybeScheduleFlush() {
1098 1 : if d.mu.compact.flushing || d.closed.Load() != nil || d.opts.ReadOnly {
1099 1 : return
1100 1 : }
1101 1 : if len(d.mu.mem.queue) <= 1 {
1102 1 : return
1103 1 : }
1104 :
1105 1 : if !d.passedFlushThreshold() {
1106 1 : return
1107 1 : }
1108 :
1109 1 : d.mu.compact.flushing = true
1110 1 : go d.flush()
1111 : }
1112 :
1113 1 : func (d *DB) passedFlushThreshold() bool {
1114 1 : var n int
1115 1 : var size uint64
1116 1 : for ; n < len(d.mu.mem.queue)-1; n++ {
1117 1 : if !d.mu.mem.queue[n].readyForFlush() {
1118 1 : break
1119 : }
1120 1 : if d.mu.mem.queue[n].flushForced {
1121 1 : // A flush was forced. Pretend the memtable size is the configured
1122 1 : // size. See minFlushSize below.
1123 1 : size += d.opts.MemTableSize
1124 1 : } else {
1125 1 : size += d.mu.mem.queue[n].totalBytes()
1126 1 : }
1127 : }
1128 1 : if n == 0 {
1129 1 : // None of the immutable memtables are ready for flushing.
1130 1 : return false
1131 1 : }
1132 :
1133 : // Only flush once the sum of the queued memtable sizes exceeds half the
1134 : // configured memtable size. This prevents flushing of memtables at startup
1135 : // while we're undergoing the ramp period on the memtable size. See
1136 : // DB.newMemTable().
1137 1 : minFlushSize := d.opts.MemTableSize / 2
1138 1 : return size >= minFlushSize
1139 : }
1140 :
1141 1 : func (d *DB) maybeScheduleDelayedFlush(tbl *memTable, dur time.Duration) {
1142 1 : var mem *flushableEntry
1143 1 : for _, m := range d.mu.mem.queue {
1144 1 : if m.flushable == tbl {
1145 1 : mem = m
1146 1 : break
1147 : }
1148 : }
1149 1 : if mem == nil || mem.flushForced {
1150 1 : return
1151 1 : }
1152 1 : deadline := d.timeNow().Add(dur)
1153 1 : if !mem.delayedFlushForcedAt.IsZero() && deadline.After(mem.delayedFlushForcedAt) {
1154 1 : // Already scheduled to flush sooner than within `dur`.
1155 1 : return
1156 1 : }
1157 1 : mem.delayedFlushForcedAt = deadline
1158 1 : go func() {
1159 1 : timer := time.NewTimer(dur)
1160 1 : defer timer.Stop()
1161 1 :
1162 1 : select {
1163 1 : case <-d.closedCh:
1164 1 : return
1165 1 : case <-mem.flushed:
1166 1 : return
1167 1 : case <-timer.C:
1168 1 : d.commit.mu.Lock()
1169 1 : defer d.commit.mu.Unlock()
1170 1 : d.mu.Lock()
1171 1 : defer d.mu.Unlock()
1172 1 :
1173 1 : // NB: The timer may fire concurrently with a call to Close. If a
1174 1 : // Close call beat us to acquiring d.mu, d.closed holds ErrClosed,
1175 1 : // and it's too late to flush anything. Otherwise, the Close call
1176 1 : // will block on locking d.mu until we've finished scheduling the
1177 1 : // flush and set `d.mu.compact.flushing` to true. Close will wait
1178 1 : // for the current flush to complete.
1179 1 : if d.closed.Load() != nil {
1180 1 : return
1181 1 : }
1182 :
1183 1 : if d.mu.mem.mutable == tbl {
1184 1 : d.makeRoomForWrite(nil)
1185 1 : } else {
1186 1 : mem.flushForced = true
1187 1 : }
1188 1 : d.maybeScheduleFlush()
1189 : }
1190 : }()
1191 : }
1192 :
1193 1 : func (d *DB) flush() {
1194 1 : pprof.Do(context.Background(), flushLabels, func(context.Context) {
1195 1 : flushingWorkStart := crtime.NowMono()
1196 1 : d.mu.Lock()
1197 1 : defer d.mu.Unlock()
1198 1 : idleDuration := flushingWorkStart.Sub(d.mu.compact.noOngoingFlushStartTime)
1199 1 : var bytesFlushed uint64
1200 1 : var err error
1201 1 : if bytesFlushed, err = d.flush1(); err != nil {
1202 1 : // TODO(peter): count consecutive flush errors and backoff.
1203 1 : d.opts.EventListener.BackgroundError(err)
1204 1 : }
1205 1 : d.mu.compact.flushing = false
1206 1 : d.mu.compact.noOngoingFlushStartTime = crtime.NowMono()
1207 1 : workDuration := d.mu.compact.noOngoingFlushStartTime.Sub(flushingWorkStart)
1208 1 : d.mu.compact.flushWriteThroughput.Bytes += int64(bytesFlushed)
1209 1 : d.mu.compact.flushWriteThroughput.WorkDuration += workDuration
1210 1 : d.mu.compact.flushWriteThroughput.IdleDuration += idleDuration
1211 1 : // More flush work may have arrived while we were flushing, so schedule
1212 1 : // another flush if needed.
1213 1 : d.maybeScheduleFlush()
1214 1 : // The flush may have produced too many files in a level, so schedule a
1215 1 : // compaction if needed.
1216 1 : d.maybeScheduleCompaction()
1217 1 : d.mu.compact.cond.Broadcast()
1218 : })
1219 : }
1220 :
1221 : // runIngestFlush is used to generate a flush version edit for sstables which
1222 : // were ingested as flushables. Both DB.mu and the manifest lock must be held
1223 : // while runIngestFlush is called.
1224 1 : func (d *DB) runIngestFlush(c *compaction) (*manifest.VersionEdit, error) {
1225 1 : if len(c.flushing) != 1 {
1226 0 : panic("pebble: ingestedFlushable must be flushed one at a time.")
1227 : }
1228 :
1229 : // Construct the VersionEdit, levelMetrics etc.
1230 1 : c.metrics = make(map[int]*LevelMetrics, numLevels)
1231 1 : // Finding the target level for ingestion must use the latest version
1232 1 : // after the logLock has been acquired.
1233 1 : c.version = d.mu.versions.currentVersion()
1234 1 :
1235 1 : baseLevel := d.mu.versions.picker.getBaseLevel()
1236 1 : ve := &versionEdit{}
1237 1 : var ingestSplitFiles []ingestSplitFile
1238 1 : ingestFlushable := c.flushing[0].flushable.(*ingestedFlushable)
1239 1 :
1240 1 : updateLevelMetricsOnExcise := func(m *fileMetadata, level int, added []newFileEntry) {
1241 1 : levelMetrics := c.metrics[level]
1242 1 : if levelMetrics == nil {
1243 1 : levelMetrics = &LevelMetrics{}
1244 1 : c.metrics[level] = levelMetrics
1245 1 : }
1246 1 : levelMetrics.NumFiles--
1247 1 : levelMetrics.Size -= int64(m.Size)
1248 1 : for i := range added {
1249 1 : levelMetrics.NumFiles++
1250 1 : levelMetrics.Size += int64(added[i].Meta.Size)
1251 1 : }
1252 : }
1253 :
1254 1 : suggestSplit := d.opts.Experimental.IngestSplit != nil && d.opts.Experimental.IngestSplit() &&
1255 1 : d.FormatMajorVersion() >= FormatVirtualSSTables
1256 1 :
1257 1 : if suggestSplit || ingestFlushable.exciseSpan.Valid() {
1258 1 : // We could add deleted files to ve.
1259 1 : ve.DeletedFiles = make(map[manifest.DeletedFileEntry]*manifest.FileMetadata)
1260 1 : }
1261 :
1262 1 : ctx := context.Background()
1263 1 : overlapChecker := &overlapChecker{
1264 1 : comparer: d.opts.Comparer,
1265 1 : newIters: d.newIters,
1266 1 : opts: IterOptions{
1267 1 : logger: d.opts.Logger,
1268 1 : Category: categoryIngest,
1269 1 : },
1270 1 : v: c.version,
1271 1 : }
1272 1 : replacedFiles := make(map[base.FileNum][]newFileEntry)
1273 1 : for _, file := range ingestFlushable.files {
1274 1 : var fileToSplit *fileMetadata
1275 1 : var level int
1276 1 :
1277 1 : // This file fits perfectly within the excise span, so we can slot it at L6.
1278 1 : if ingestFlushable.exciseSpan.Valid() &&
1279 1 : ingestFlushable.exciseSpan.Contains(d.cmp, file.FileMetadata.Smallest) &&
1280 1 : ingestFlushable.exciseSpan.Contains(d.cmp, file.FileMetadata.Largest) {
1281 1 : level = 6
1282 1 : } else {
1283 1 : // TODO(radu): this can perform I/O; we should not do this while holding DB.mu.
1284 1 : lsmOverlap, err := overlapChecker.DetermineLSMOverlap(ctx, file.UserKeyBounds())
1285 1 : if err != nil {
1286 0 : return nil, err
1287 0 : }
1288 1 : level, fileToSplit, err = ingestTargetLevel(
1289 1 : ctx, d.cmp, lsmOverlap, baseLevel, d.mu.compact.inProgress, file.FileMetadata, suggestSplit,
1290 1 : )
1291 1 : if err != nil {
1292 0 : return nil, err
1293 0 : }
1294 : }
1295 :
1296 : // Add the current flushableIngest file to the version.
1297 1 : ve.NewFiles = append(ve.NewFiles, newFileEntry{Level: level, Meta: file.FileMetadata})
1298 1 : if fileToSplit != nil {
1299 0 : ingestSplitFiles = append(ingestSplitFiles, ingestSplitFile{
1300 0 : ingestFile: file.FileMetadata,
1301 0 : splitFile: fileToSplit,
1302 0 : level: level,
1303 0 : })
1304 0 : }
1305 1 : levelMetrics := c.metrics[level]
1306 1 : if levelMetrics == nil {
1307 1 : levelMetrics = &LevelMetrics{}
1308 1 : c.metrics[level] = levelMetrics
1309 1 : }
1310 1 : levelMetrics.BytesIngested += file.Size
1311 1 : levelMetrics.TablesIngested++
1312 : }
1313 1 : if ingestFlushable.exciseSpan.Valid() {
1314 1 : // Iterate through all levels and find files that intersect with exciseSpan.
1315 1 : for l := range c.version.Levels {
1316 1 : overlaps := c.version.Overlaps(l, base.UserKeyBoundsEndExclusive(ingestFlushable.exciseSpan.Start, ingestFlushable.exciseSpan.End))
1317 1 : iter := overlaps.Iter()
1318 1 :
1319 1 : for m := iter.First(); m != nil; m = iter.Next() {
1320 1 : newFiles, err := d.excise(context.TODO(), ingestFlushable.exciseSpan.UserKeyBounds(), m, ve, l)
1321 1 : if err != nil {
1322 0 : return nil, err
1323 0 : }
1324 :
1325 1 : if _, ok := ve.DeletedFiles[deletedFileEntry{
1326 1 : Level: l,
1327 1 : FileNum: m.FileNum,
1328 1 : }]; !ok {
1329 1 : // We did not excise this file.
1330 1 : continue
1331 : }
1332 1 : replacedFiles[m.FileNum] = newFiles
1333 1 : updateLevelMetricsOnExcise(m, l, newFiles)
1334 : }
1335 : }
1336 : }
1337 :
1338 1 : if len(ingestSplitFiles) > 0 {
1339 0 : if err := d.ingestSplit(context.TODO(), ve, updateLevelMetricsOnExcise, ingestSplitFiles, replacedFiles); err != nil {
1340 0 : return nil, err
1341 0 : }
1342 : }
1343 :
1344 1 : return ve, nil
1345 : }
1346 :
1347 : // flush runs a compaction that copies the immutable memtables from memory to
1348 : // disk.
1349 : //
1350 : // d.mu must be held when calling this, but the mutex may be dropped and
1351 : // re-acquired during the course of this method.
1352 1 : func (d *DB) flush1() (bytesFlushed uint64, err error) {
1353 1 : // NB: The flushable queue can contain flushables of type ingestedFlushable.
1354 1 : // The sstables in ingestedFlushable.files must be placed into the appropriate
1355 1 : // level in the lsm. Let's say the flushable queue contains a prefix of
1356 1 : // regular immutable memtables, then an ingestedFlushable, and then the
1357 1 : // mutable memtable. When the flush of the ingestedFlushable is performed,
1358 1 : // it needs an updated view of the lsm. That is, the prefix of immutable
1359 1 : // memtables must have already been flushed. Similarly, if there are two
1360 1 : // contiguous ingestedFlushables in the queue, then the first flushable must
1361 1 : // be flushed, so that the second flushable can see an updated view of the
1362 1 : // lsm.
1363 1 : //
1364 1 : // Given the above, we restrict flushes to either some prefix of regular
1365 1 : // memtables, or a single flushable of type ingestedFlushable. The DB.flush
1366 1 : // function will call DB.maybeScheduleFlush again, so a new flush to finish
1367 1 : // the remaining flush work should be scheduled right away.
1368 1 : //
1369 1 : // NB: Large batches placed in the flushable queue share the WAL with the
1370 1 : // previous memtable in the queue. We must ensure the property that both the
1371 1 : // large batch and the memtable with which it shares a WAL are flushed
1372 1 : // together. The property ensures that the minimum unflushed log number
1373 1 : // isn't incremented incorrectly. Since a flushableBatch.readyToFlush always
1374 1 : // returns true, and since the large batch will always be placed right after
1375 1 : // the memtable with which it shares a WAL, the property is naturally
1376 1 : // ensured. The large batch will always be placed after the memtable with
1377 1 : // which it shares a WAL because we ensure it in DB.commitWrite by holding
1378 1 : // the commitPipeline.mu and then holding DB.mu. As an extra defensive
1379 1 : // measure, if we try to flush the memtable without also flushing the
1380 1 : // flushable batch in the same flush, since the memtable and flushableBatch
1381 1 : // have the same logNum, the logNum invariant check below will trigger.
1382 1 : var n, inputs int
1383 1 : var inputBytes uint64
1384 1 : var ingest bool
1385 1 : for ; n < len(d.mu.mem.queue)-1; n++ {
1386 1 : if f, ok := d.mu.mem.queue[n].flushable.(*ingestedFlushable); ok {
1387 1 : if n == 0 {
1388 1 : // The first flushable is of type ingestedFlushable. Since these
1389 1 : // must be flushed individually, we perform a flush for just
1390 1 : // this.
1391 1 : if !f.readyForFlush() {
1392 0 : // This check is almost unnecessary, but we guard against it
1393 0 : // just in case this invariant changes in the future.
1394 0 : panic("pebble: ingestedFlushable should always be ready to flush.")
1395 : }
1396 : // By setting n = 1, we ensure that the first flushable(n == 0)
1397 : // is scheduled for a flush. The number of tables added is equal to the
1398 : // number of files in the ingest operation.
1399 1 : n = 1
1400 1 : inputs = len(f.files)
1401 1 : ingest = true
1402 1 : break
1403 1 : } else {
1404 1 : // There was some prefix of flushables which weren't of type
1405 1 : // ingestedFlushable. So, perform a flush for those.
1406 1 : break
1407 : }
1408 : }
1409 1 : if !d.mu.mem.queue[n].readyForFlush() {
1410 1 : break
1411 : }
1412 1 : inputBytes += d.mu.mem.queue[n].inuseBytes()
1413 : }
1414 1 : if n == 0 {
1415 0 : // None of the immutable memtables are ready for flushing.
1416 0 : return 0, nil
1417 0 : }
1418 1 : if !ingest {
1419 1 : // Flushes of memtables add the prefix of n memtables from the flushable
1420 1 : // queue.
1421 1 : inputs = n
1422 1 : }
1423 :
1424 : // Require that every memtable being flushed has a log number less than the
1425 : // new minimum unflushed log number.
1426 1 : minUnflushedLogNum := d.mu.mem.queue[n].logNum
1427 1 : if !d.opts.DisableWAL {
1428 1 : for i := 0; i < n; i++ {
1429 1 : if logNum := d.mu.mem.queue[i].logNum; logNum >= minUnflushedLogNum {
1430 0 : panic(errors.AssertionFailedf("logNum invariant violated: flushing %d items; %d:type=%T,logNum=%d; %d:type=%T,logNum=%d",
1431 0 : n,
1432 0 : i, d.mu.mem.queue[i].flushable, logNum,
1433 0 : n, d.mu.mem.queue[n].flushable, minUnflushedLogNum))
1434 : }
1435 : }
1436 : }
1437 :
1438 1 : c, err := newFlush(d.opts, d.mu.versions.currentVersion(),
1439 1 : d.mu.versions.picker.getBaseLevel(), d.mu.mem.queue[:n], d.timeNow())
1440 1 : if err != nil {
1441 0 : return 0, err
1442 0 : }
1443 1 : d.addInProgressCompaction(c)
1444 1 :
1445 1 : jobID := d.newJobIDLocked()
1446 1 : d.opts.EventListener.FlushBegin(FlushInfo{
1447 1 : JobID: int(jobID),
1448 1 : Input: inputs,
1449 1 : InputBytes: inputBytes,
1450 1 : Ingest: ingest,
1451 1 : })
1452 1 : startTime := d.timeNow()
1453 1 :
1454 1 : var ve *manifest.VersionEdit
1455 1 : var stats compact.Stats
1456 1 : // To determine the target level of the files in the ingestedFlushable, we
1457 1 : // need to acquire the logLock, and not release it for that duration. Since,
1458 1 : // we need to acquire the logLock below to perform the logAndApply step
1459 1 : // anyway, we create the VersionEdit for ingestedFlushable outside of
1460 1 : // runCompaction. For all other flush cases, we construct the VersionEdit
1461 1 : // inside runCompaction.
1462 1 : if c.kind != compactionKindIngestedFlushable {
1463 1 : ve, stats, err = d.runCompaction(jobID, c)
1464 1 : }
1465 :
1466 : // Acquire logLock. This will be released either on an error, by way of
1467 : // logUnlock, or through a call to logAndApply if there is no error.
1468 1 : d.mu.versions.logLock()
1469 1 :
1470 1 : if c.kind == compactionKindIngestedFlushable {
1471 1 : ve, err = d.runIngestFlush(c)
1472 1 : }
1473 :
1474 1 : info := FlushInfo{
1475 1 : JobID: int(jobID),
1476 1 : Input: inputs,
1477 1 : InputBytes: inputBytes,
1478 1 : Duration: d.timeNow().Sub(startTime),
1479 1 : Done: true,
1480 1 : Ingest: ingest,
1481 1 : Err: err,
1482 1 : }
1483 1 : if err == nil {
1484 1 : validateVersionEdit(ve, d.opts.Experimental.KeyValidationFunc, d.opts.Comparer.FormatKey, d.opts.Logger)
1485 1 : for i := range ve.NewFiles {
1486 1 : e := &ve.NewFiles[i]
1487 1 : info.Output = append(info.Output, e.Meta.TableInfo())
1488 1 : // Ingested tables are not necessarily flushed to L0. Record the level of
1489 1 : // each ingested file explicitly.
1490 1 : if ingest {
1491 1 : info.IngestLevels = append(info.IngestLevels, e.Level)
1492 1 : }
1493 : }
1494 1 : if len(ve.NewFiles) == 0 {
1495 1 : info.Err = errEmptyTable
1496 1 : }
1497 :
1498 : // The flush succeeded or it produced an empty sstable. In either case we
1499 : // want to bump the minimum unflushed log number to the log number of the
1500 : // oldest unflushed memtable.
1501 1 : ve.MinUnflushedLogNum = minUnflushedLogNum
1502 1 : if c.kind != compactionKindIngestedFlushable {
1503 1 : metrics := c.metrics[0]
1504 1 : if d.opts.DisableWAL {
1505 1 : // If the WAL is disabled, every flushable has a zero [logSize],
1506 1 : // resulting in zero bytes in. Instead, use the number of bytes we
1507 1 : // flushed as the BytesIn. This ensures we get a reasonable w-amp
1508 1 : // calculation even when the WAL is disabled.
1509 1 : metrics.BytesIn = metrics.BytesFlushed
1510 1 : } else {
1511 1 : for i := 0; i < n; i++ {
1512 1 : metrics.BytesIn += d.mu.mem.queue[i].logSize
1513 1 : }
1514 : }
1515 1 : } else {
1516 1 : // c.kind == compactionKindIngestedFlushable && we could have deleted files due
1517 1 : // to ingest-time splits or excises.
1518 1 : ingestFlushable := c.flushing[0].flushable.(*ingestedFlushable)
1519 1 : for c2 := range d.mu.compact.inProgress {
1520 1 : // Check if this compaction overlaps with the excise span. Note that just
1521 1 : // checking if the inputs individually overlap with the excise span
1522 1 : // isn't sufficient; for instance, a compaction could have [a,b] and [e,f]
1523 1 : // as inputs and write it all out as [a,b,e,f] in one sstable. If we're
1524 1 : // doing a [c,d) excise at the same time as this compaction, we will have
1525 1 : // to error out the whole compaction as we can't guarantee it hasn't/won't
1526 1 : // write a file overlapping with the excise span.
1527 1 : if ingestFlushable.exciseSpan.OverlapsInternalKeyRange(d.cmp, c2.smallest, c2.largest) {
1528 1 : c2.cancel.Store(true)
1529 1 : continue
1530 : }
1531 : }
1532 :
1533 1 : if len(ve.DeletedFiles) > 0 {
1534 1 : // Iterate through all other compactions, and check if their inputs have
1535 1 : // been replaced due to an ingest-time split or excise. In that case,
1536 1 : // cancel the compaction.
1537 1 : for c2 := range d.mu.compact.inProgress {
1538 1 : for i := range c2.inputs {
1539 1 : iter := c2.inputs[i].files.Iter()
1540 1 : for f := iter.First(); f != nil; f = iter.Next() {
1541 1 : if _, ok := ve.DeletedFiles[deletedFileEntry{FileNum: f.FileNum, Level: c2.inputs[i].level}]; ok {
1542 1 : c2.cancel.Store(true)
1543 1 : break
1544 : }
1545 : }
1546 : }
1547 : }
1548 : }
1549 : }
1550 1 : err = d.mu.versions.logAndApply(jobID, ve, c.metrics, false, /* forceRotation */
1551 1 : func() []compactionInfo { return d.getInProgressCompactionInfoLocked(c) })
1552 1 : if err != nil {
1553 1 : info.Err = err
1554 1 : }
1555 1 : } else {
1556 1 : // We won't be performing the logAndApply step because of the error,
1557 1 : // so logUnlock.
1558 1 : d.mu.versions.logUnlock()
1559 1 : }
1560 :
1561 : // If err != nil, then the flush will be retried, and we will recalculate
1562 : // these metrics.
1563 1 : if err == nil {
1564 1 : d.mu.snapshots.cumulativePinnedCount += stats.CumulativePinnedKeys
1565 1 : d.mu.snapshots.cumulativePinnedSize += stats.CumulativePinnedSize
1566 1 : d.mu.versions.metrics.Keys.MissizedTombstonesCount += stats.CountMissizedDels
1567 1 : }
1568 :
1569 1 : d.clearCompactingState(c, err != nil)
1570 1 : delete(d.mu.compact.inProgress, c)
1571 1 : d.mu.versions.incrementCompactions(c.kind, c.extraLevels, c.pickerMetrics)
1572 1 :
1573 1 : var flushed flushableList
1574 1 : if err == nil {
1575 1 : flushed = d.mu.mem.queue[:n]
1576 1 : d.mu.mem.queue = d.mu.mem.queue[n:]
1577 1 : d.updateReadStateLocked(d.opts.DebugCheck)
1578 1 : d.updateTableStatsLocked(ve.NewFiles)
1579 1 : if ingest {
1580 1 : d.mu.versions.metrics.Flush.AsIngestCount++
1581 1 : for _, l := range c.metrics {
1582 1 : d.mu.versions.metrics.Flush.AsIngestBytes += l.BytesIngested
1583 1 : d.mu.versions.metrics.Flush.AsIngestTableCount += l.TablesIngested
1584 1 : }
1585 : }
1586 1 : d.maybeTransitionSnapshotsToFileOnlyLocked()
1587 :
1588 : }
1589 : // Signal FlushEnd after installing the new readState. This helps for unit
1590 : // tests that use the callback to trigger a read using an iterator with
1591 : // IterOptions.OnlyReadGuaranteedDurable.
1592 1 : info.TotalDuration = d.timeNow().Sub(startTime)
1593 1 : d.opts.EventListener.FlushEnd(info)
1594 1 :
1595 1 : // The order of these operations matters here for ease of testing.
1596 1 : // Removing the reader reference first allows tests to be guaranteed that
1597 1 : // the memtable reservation has been released by the time a synchronous
1598 1 : // flush returns. readerUnrefLocked may also produce obsolete files so the
1599 1 : // call to deleteObsoleteFiles must happen after it.
1600 1 : for i := range flushed {
1601 1 : flushed[i].readerUnrefLocked(true)
1602 1 : }
1603 :
1604 1 : d.deleteObsoleteFiles(jobID)
1605 1 :
1606 1 : // Mark all the memtables we flushed as flushed.
1607 1 : for i := range flushed {
1608 1 : close(flushed[i].flushed)
1609 1 : }
1610 :
1611 1 : return inputBytes, err
1612 : }
1613 :
1614 : // maybeTransitionSnapshotsToFileOnlyLocked transitions any "eventually
1615 : // file-only" snapshots to be file-only if all their visible state has been
1616 : // flushed to sstables.
1617 : //
1618 : // REQUIRES: d.mu.
1619 1 : func (d *DB) maybeTransitionSnapshotsToFileOnlyLocked() {
1620 1 : earliestUnflushedSeqNum := d.getEarliestUnflushedSeqNumLocked()
1621 1 : currentVersion := d.mu.versions.currentVersion()
1622 1 : for s := d.mu.snapshots.root.next; s != &d.mu.snapshots.root; {
1623 1 : if s.efos == nil {
1624 1 : s = s.next
1625 1 : continue
1626 : }
1627 1 : overlapsFlushable := false
1628 1 : if base.Visible(earliestUnflushedSeqNum, s.efos.seqNum, base.SeqNumMax) {
1629 1 : // There are some unflushed keys that are still visible to the EFOS.
1630 1 : // Check if any memtables older than the EFOS contain keys within a
1631 1 : // protected range of the EFOS. If no, we can transition.
1632 1 : protectedRanges := make([]bounded, len(s.efos.protectedRanges))
1633 1 : for i := range s.efos.protectedRanges {
1634 1 : protectedRanges[i] = s.efos.protectedRanges[i]
1635 1 : }
1636 1 : for i := range d.mu.mem.queue {
1637 1 : if !base.Visible(d.mu.mem.queue[i].logSeqNum, s.efos.seqNum, base.SeqNumMax) {
1638 0 : // All keys in this memtable are newer than the EFOS. Skip this
1639 0 : // memtable.
1640 0 : continue
1641 : }
1642 : // NB: computePossibleOverlaps could have false positives, such as if
1643 : // the flushable is a flushable ingest and not a memtable. In that
1644 : // case we don't open the sstables to check; we just pessimistically
1645 : // assume an overlap.
1646 1 : d.mu.mem.queue[i].computePossibleOverlaps(func(b bounded) shouldContinue {
1647 1 : overlapsFlushable = true
1648 1 : return stopIteration
1649 1 : }, protectedRanges...)
1650 1 : if overlapsFlushable {
1651 1 : break
1652 : }
1653 : }
1654 : }
1655 1 : if overlapsFlushable {
1656 1 : s = s.next
1657 1 : continue
1658 : }
1659 1 : currentVersion.Ref()
1660 1 :
1661 1 : // NB: s.efos.transitionToFileOnlySnapshot could close s, in which
1662 1 : // case s.next would be nil. Save it before calling it.
1663 1 : next := s.next
1664 1 : _ = s.efos.transitionToFileOnlySnapshot(currentVersion)
1665 1 : s = next
1666 : }
1667 : }
1668 :
1669 : // maybeScheduleCompactionAsync should be used when
1670 : // we want to possibly schedule a compaction, but don't
1671 : // want to eat the cost of running maybeScheduleCompaction.
1672 : // This method should be launched in a separate goroutine.
1673 : // d.mu must not be held when this is called.
1674 0 : func (d *DB) maybeScheduleCompactionAsync() {
1675 0 : defer d.compactionSchedulers.Done()
1676 0 :
1677 0 : d.mu.Lock()
1678 0 : d.maybeScheduleCompaction()
1679 0 : d.mu.Unlock()
1680 0 : }
1681 :
1682 : // maybeScheduleCompaction schedules a compaction if necessary.
1683 : //
1684 : // d.mu must be held when calling this.
1685 1 : func (d *DB) maybeScheduleCompaction() {
1686 1 : d.maybeScheduleCompactionPicker(pickAuto)
1687 1 : }
1688 :
1689 1 : func pickAuto(picker compactionPicker, env compactionEnv) *pickedCompaction {
1690 1 : return picker.pickAuto(env)
1691 1 : }
1692 :
1693 1 : func pickElisionOnly(picker compactionPicker, env compactionEnv) *pickedCompaction {
1694 1 : return picker.pickElisionOnlyCompaction(env)
1695 1 : }
1696 :
1697 : // tryScheduleDownloadCompaction tries to start a download compaction.
1698 : //
1699 : // Returns true if we started a download compaction (or completed it
1700 : // immediately because it is a no-op or we hit an error).
1701 : //
1702 : // Requires d.mu to be held. Updates d.mu.compact.downloads.
1703 1 : func (d *DB) tryScheduleDownloadCompaction(env compactionEnv, maxConcurrentDownloads int) bool {
1704 1 : vers := d.mu.versions.currentVersion()
1705 1 : for i := 0; i < len(d.mu.compact.downloads); {
1706 1 : download := d.mu.compact.downloads[i]
1707 1 : switch d.tryLaunchDownloadCompaction(download, vers, env, maxConcurrentDownloads) {
1708 1 : case launchedCompaction:
1709 1 : return true
1710 0 : case didNotLaunchCompaction:
1711 0 : // See if we can launch a compaction for another download task.
1712 0 : i++
1713 1 : case downloadTaskCompleted:
1714 1 : // Task is completed and must be removed.
1715 1 : d.mu.compact.downloads = slices.Delete(d.mu.compact.downloads, i, i+1)
1716 : }
1717 : }
1718 1 : return false
1719 : }
1720 :
1721 : // maybeScheduleCompactionPicker schedules a compaction if necessary,
1722 : // calling `pickFunc` to pick automatic compactions.
1723 : //
1724 : // Requires d.mu to be held.
1725 : func (d *DB) maybeScheduleCompactionPicker(
1726 : pickFunc func(compactionPicker, compactionEnv) *pickedCompaction,
1727 1 : ) {
1728 1 : if d.closed.Load() != nil || d.opts.ReadOnly {
1729 1 : return
1730 1 : }
1731 1 : maxCompactions := d.opts.MaxConcurrentCompactions()
1732 1 : maxDownloads := d.opts.MaxConcurrentDownloads()
1733 1 :
1734 1 : if d.mu.compact.compactingCount >= maxCompactions &&
1735 1 : (len(d.mu.compact.downloads) == 0 || d.mu.compact.downloadingCount >= maxDownloads) {
1736 1 : if len(d.mu.compact.manual) > 0 {
1737 1 : // Inability to run head blocks later manual compactions.
1738 1 : d.mu.compact.manual[0].retries++
1739 1 : }
1740 1 : return
1741 : }
1742 :
1743 : // Compaction picking needs a coherent view of a Version. In particular, we
1744 : // need to exclude concurrent ingestions from making a decision on which level
1745 : // to ingest into that conflicts with our compaction
1746 : // decision. versionSet.logLock provides the necessary mutual exclusion.
1747 1 : d.mu.versions.logLock()
1748 1 : defer d.mu.versions.logUnlock()
1749 1 :
1750 1 : // Check for the closed flag again, in case the DB was closed while we were
1751 1 : // waiting for logLock().
1752 1 : if d.closed.Load() != nil {
1753 1 : return
1754 1 : }
1755 :
1756 1 : env := compactionEnv{
1757 1 : diskAvailBytes: d.diskAvailBytes.Load(),
1758 1 : earliestSnapshotSeqNum: d.mu.snapshots.earliest(),
1759 1 : earliestUnflushedSeqNum: d.getEarliestUnflushedSeqNumLocked(),
1760 1 : }
1761 1 :
1762 1 : if d.mu.compact.compactingCount < maxCompactions {
1763 1 : // Check for delete-only compactions first, because they're expected to be
1764 1 : // cheap and reduce future compaction work.
1765 1 : if !d.opts.private.disableDeleteOnlyCompactions &&
1766 1 : !d.opts.DisableAutomaticCompactions &&
1767 1 : len(d.mu.compact.deletionHints) > 0 {
1768 1 : d.tryScheduleDeleteOnlyCompaction()
1769 1 : }
1770 :
1771 1 : for len(d.mu.compact.manual) > 0 && d.mu.compact.compactingCount < maxCompactions {
1772 1 : if manual := d.mu.compact.manual[0]; !d.tryScheduleManualCompaction(env, manual) {
1773 1 : // Inability to run head blocks later manual compactions.
1774 1 : manual.retries++
1775 1 : break
1776 : }
1777 1 : d.mu.compact.manual = d.mu.compact.manual[1:]
1778 : }
1779 :
1780 1 : for !d.opts.DisableAutomaticCompactions && d.mu.compact.compactingCount < maxCompactions &&
1781 1 : d.tryScheduleAutoCompaction(env, pickFunc) {
1782 1 : }
1783 : }
1784 :
1785 1 : for len(d.mu.compact.downloads) > 0 && d.mu.compact.downloadingCount < maxDownloads &&
1786 1 : d.tryScheduleDownloadCompaction(env, maxDownloads) {
1787 1 : }
1788 : }
1789 :
1790 : // tryScheduleDeleteOnlyCompaction tries to kick off a delete-only compaction
1791 : // for all files that can be deleted as suggested by deletionHints.
1792 : //
1793 : // Requires d.mu to be held. Updates d.mu.compact.deletionHints.
1794 1 : func (d *DB) tryScheduleDeleteOnlyCompaction() {
1795 1 : v := d.mu.versions.currentVersion()
1796 1 : snapshots := d.mu.snapshots.toSlice()
1797 1 : // We need to save the value of exciseEnabled in the compaction itself, as
1798 1 : // it can change dynamically between now and when the compaction runs.
1799 1 : exciseEnabled := d.FormatMajorVersion() >= FormatVirtualSSTables &&
1800 1 : d.opts.Experimental.EnableDeleteOnlyCompactionExcises != nil && d.opts.Experimental.EnableDeleteOnlyCompactionExcises()
1801 1 : inputs, resolvedHints, unresolvedHints := checkDeleteCompactionHints(d.cmp, v, d.mu.compact.deletionHints, snapshots, exciseEnabled)
1802 1 : d.mu.compact.deletionHints = unresolvedHints
1803 1 :
1804 1 : if len(inputs) > 0 {
1805 1 : c := newDeleteOnlyCompaction(d.opts, v, inputs, d.timeNow(), resolvedHints, exciseEnabled)
1806 1 : d.mu.compact.compactingCount++
1807 1 : d.addInProgressCompaction(c)
1808 1 : go d.compact(c, nil)
1809 1 : }
1810 : }
1811 :
1812 : // tryScheduleManualCompaction tries to kick off the given manual compaction.
1813 : //
1814 : // Returns false if we are not able to run this compaction at this time.
1815 : //
1816 : // Requires d.mu to be held.
1817 1 : func (d *DB) tryScheduleManualCompaction(env compactionEnv, manual *manualCompaction) bool {
1818 1 : v := d.mu.versions.currentVersion()
1819 1 : env.inProgressCompactions = d.getInProgressCompactionInfoLocked(nil)
1820 1 : pc, retryLater := pickManualCompaction(v, d.opts, env, d.mu.versions.picker.getBaseLevel(), manual)
1821 1 : if pc == nil {
1822 1 : if !retryLater {
1823 1 : // Manual compaction is a no-op. Signal completion and exit.
1824 1 : manual.done <- nil
1825 1 : return true
1826 1 : }
1827 : // We are not able to run this manual compaction at this time.
1828 1 : return false
1829 : }
1830 :
1831 1 : c := newCompaction(pc, d.opts, d.timeNow(), d.ObjProvider())
1832 1 : d.mu.compact.compactingCount++
1833 1 : d.addInProgressCompaction(c)
1834 1 : go d.compact(c, manual.done)
1835 1 : return true
1836 : }
1837 :
1838 : // tryScheduleAutoCompaction tries to kick off an automatic compaction.
1839 : //
1840 : // Returns false if no automatic compactions are necessary or able to run at
1841 : // this time.
1842 : //
1843 : // Requires d.mu to be held.
1844 : func (d *DB) tryScheduleAutoCompaction(
1845 : env compactionEnv, pickFunc func(compactionPicker, compactionEnv) *pickedCompaction,
1846 1 : ) bool {
1847 1 : env.inProgressCompactions = d.getInProgressCompactionInfoLocked(nil)
1848 1 : env.readCompactionEnv = readCompactionEnv{
1849 1 : readCompactions: &d.mu.compact.readCompactions,
1850 1 : flushing: d.mu.compact.flushing || d.passedFlushThreshold(),
1851 1 : rescheduleReadCompaction: &d.mu.compact.rescheduleReadCompaction,
1852 1 : }
1853 1 : pc := pickFunc(d.mu.versions.picker, env)
1854 1 : if pc == nil {
1855 1 : return false
1856 1 : }
1857 1 : c := newCompaction(pc, d.opts, d.timeNow(), d.ObjProvider())
1858 1 : d.mu.compact.compactingCount++
1859 1 : d.addInProgressCompaction(c)
1860 1 : go d.compact(c, nil)
1861 1 : return true
1862 : }
1863 :
1864 : // deleteCompactionHintType indicates whether the deleteCompactionHint was
1865 : // generated from a span containing a range del (point key only), a range key
1866 : // delete (range key only), or both a point and range key.
1867 : type deleteCompactionHintType uint8
1868 :
1869 : const (
1870 : // NOTE: While these are primarily used as enumeration types, they are also
1871 : // used for some bitwise operations. Care should be taken when updating.
1872 : deleteCompactionHintTypeUnknown deleteCompactionHintType = iota
1873 : deleteCompactionHintTypePointKeyOnly
1874 : deleteCompactionHintTypeRangeKeyOnly
1875 : deleteCompactionHintTypePointAndRangeKey
1876 : )
1877 :
1878 : // String implements fmt.Stringer.
1879 1 : func (h deleteCompactionHintType) String() string {
1880 1 : switch h {
1881 0 : case deleteCompactionHintTypeUnknown:
1882 0 : return "unknown"
1883 1 : case deleteCompactionHintTypePointKeyOnly:
1884 1 : return "point-key-only"
1885 1 : case deleteCompactionHintTypeRangeKeyOnly:
1886 1 : return "range-key-only"
1887 1 : case deleteCompactionHintTypePointAndRangeKey:
1888 1 : return "point-and-range-key"
1889 0 : default:
1890 0 : panic(fmt.Sprintf("unknown hint type: %d", h))
1891 : }
1892 : }
1893 :
1894 : // compactionHintFromKeys returns a deleteCompactionHintType given a slice of
1895 : // keyspan.Keys.
1896 1 : func compactionHintFromKeys(keys []keyspan.Key) deleteCompactionHintType {
1897 1 : var hintType deleteCompactionHintType
1898 1 : for _, k := range keys {
1899 1 : switch k.Kind() {
1900 1 : case base.InternalKeyKindRangeDelete:
1901 1 : hintType |= deleteCompactionHintTypePointKeyOnly
1902 1 : case base.InternalKeyKindRangeKeyDelete:
1903 1 : hintType |= deleteCompactionHintTypeRangeKeyOnly
1904 0 : default:
1905 0 : panic(fmt.Sprintf("unsupported key kind: %s", k.Kind()))
1906 : }
1907 : }
1908 1 : return hintType
1909 : }
1910 :
1911 : // A deleteCompactionHint records a user key and sequence number span that has been
1912 : // deleted by a range tombstone. A hint is recorded if at least one sstable
1913 : // falls completely within both the user key and sequence number spans.
1914 : // Once the tombstones and the observed completely-contained sstables fall
1915 : // into the same snapshot stripe, a delete-only compaction may delete any
1916 : // sstables within the range.
1917 : type deleteCompactionHint struct {
1918 : // The type of key span that generated this hint (point key, range key, or
1919 : // both).
1920 : hintType deleteCompactionHintType
1921 : // start and end are user keys specifying a key range [start, end) of
1922 : // deleted keys.
1923 : start []byte
1924 : end []byte
1925 : // The level of the file containing the range tombstone(s) when the hint
1926 : // was created. Only lower levels need to be searched for files that may
1927 : // be deleted.
1928 : tombstoneLevel int
1929 : // The file containing the range tombstone(s) that created the hint.
1930 : tombstoneFile *fileMetadata
1931 : // The smallest and largest sequence numbers of the abutting tombstones
1932 : // merged to form this hint. All of a tables' keys must be less than the
1933 : // tombstone smallest sequence number to be deleted. All of a tables'
1934 : // sequence numbers must fall into the same snapshot stripe as the
1935 : // tombstone largest sequence number to be deleted.
1936 : tombstoneLargestSeqNum base.SeqNum
1937 : tombstoneSmallestSeqNum base.SeqNum
1938 : // The smallest sequence number of a sstable that was found to be covered
1939 : // by this hint. The hint cannot be resolved until this sequence number is
1940 : // in the same snapshot stripe as the largest tombstone sequence number.
1941 : // This is set when a hint is created, so the LSM may look different and
1942 : // notably no longer contain the sstable that contained the key at this
1943 : // sequence number.
1944 : fileSmallestSeqNum base.SeqNum
1945 : }
1946 :
1947 : type deletionHintOverlap int8
1948 :
1949 : const (
1950 : // hintDoesNotApply indicates that the hint does not apply to the file.
1951 : hintDoesNotApply deletionHintOverlap = iota
1952 : // hintExcisesFile indicates that the hint excises a portion of the file,
1953 : // and the format major version of the DB supports excises.
1954 : hintExcisesFile
1955 : // hintDeletesFile indicates that the hint deletes the entirety of the file.
1956 : hintDeletesFile
1957 : )
1958 :
1959 1 : func (h deleteCompactionHint) String() string {
1960 1 : return fmt.Sprintf(
1961 1 : "L%d.%s %s-%s seqnums(tombstone=%d-%d, file-smallest=%d, type=%s)",
1962 1 : h.tombstoneLevel, h.tombstoneFile.FileNum, h.start, h.end,
1963 1 : h.tombstoneSmallestSeqNum, h.tombstoneLargestSeqNum, h.fileSmallestSeqNum,
1964 1 : h.hintType,
1965 1 : )
1966 1 : }
1967 :
1968 : func (h *deleteCompactionHint) canDeleteOrExcise(
1969 : cmp Compare, m *fileMetadata, snapshots compact.Snapshots, exciseEnabled bool,
1970 1 : ) deletionHintOverlap {
1971 1 : // The file can only be deleted if all of its keys are older than the
1972 1 : // earliest tombstone aggregated into the hint. Note that we use
1973 1 : // m.LargestSeqNumAbsolute, not m.LargestSeqNum. Consider a compaction that
1974 1 : // zeroes sequence numbers. A compaction may zero the sequence number of a
1975 1 : // key with a sequence number > h.tombstoneSmallestSeqNum and set it to
1976 1 : // zero. If we looked at m.LargestSeqNum, the resulting output file would
1977 1 : // appear to not contain any keys more recent than the oldest tombstone. To
1978 1 : // avoid this error, the largest pre-zeroing sequence number is maintained
1979 1 : // in LargestSeqNumAbsolute and used here to make the determination whether
1980 1 : // the file's keys are older than all of the hint's tombstones.
1981 1 : if m.LargestSeqNumAbsolute >= h.tombstoneSmallestSeqNum || m.SmallestSeqNum < h.fileSmallestSeqNum {
1982 1 : return hintDoesNotApply
1983 1 : }
1984 :
1985 : // The file's oldest key must be in the same snapshot stripe as the
1986 : // newest tombstone. NB: We already checked the hint's sequence numbers,
1987 : // but this file's oldest sequence number might be lower than the hint's
1988 : // smallest sequence number despite the file falling within the key range
1989 : // if this file was constructed after the hint by a compaction.
1990 1 : if snapshots.Index(h.tombstoneLargestSeqNum) != snapshots.Index(m.SmallestSeqNum) {
1991 0 : return hintDoesNotApply
1992 0 : }
1993 :
1994 1 : switch h.hintType {
1995 1 : case deleteCompactionHintTypePointKeyOnly:
1996 1 : // A hint generated by a range del span cannot delete tables that contain
1997 1 : // range keys.
1998 1 : if m.HasRangeKeys {
1999 1 : return hintDoesNotApply
2000 1 : }
2001 1 : case deleteCompactionHintTypeRangeKeyOnly:
2002 1 : // A hint generated by a range key del span cannot delete tables that
2003 1 : // contain point keys.
2004 1 : if m.HasPointKeys {
2005 1 : return hintDoesNotApply
2006 1 : }
2007 1 : case deleteCompactionHintTypePointAndRangeKey:
2008 : // A hint from a span that contains both range dels *and* range keys can
2009 : // only be deleted if both bounds fall within the hint. The next check takes
2010 : // care of this.
2011 0 : default:
2012 0 : panic(fmt.Sprintf("pebble: unknown delete compaction hint type: %d", h.hintType))
2013 : }
2014 1 : if cmp(h.start, m.Smallest.UserKey) <= 0 &&
2015 1 : base.UserKeyExclusive(h.end).CompareUpperBounds(cmp, m.UserKeyBounds().End) >= 0 {
2016 1 : return hintDeletesFile
2017 1 : }
2018 1 : if !exciseEnabled {
2019 1 : // The file's keys must be completely contained within the hint range; excises
2020 1 : // aren't allowed.
2021 1 : return hintDoesNotApply
2022 1 : }
2023 : // Check for any overlap. In cases of partial overlap, we can excise the part of the file
2024 : // that overlaps with the deletion hint.
2025 1 : if cmp(h.end, m.Smallest.UserKey) > 0 &&
2026 1 : (m.UserKeyBounds().End.CompareUpperBounds(cmp, base.UserKeyInclusive(h.start)) >= 0) {
2027 1 : return hintExcisesFile
2028 1 : }
2029 0 : return hintDoesNotApply
2030 : }
2031 :
2032 : // checkDeleteCompactionHints checks the passed-in deleteCompactionHints for those that
2033 : // can be resolved and those that cannot. A hint is considered resolved when its largest
2034 : // tombstone sequence number and the smallest sequence number of covered files fall in
2035 : // the same snapshot stripe. No more than maxHintsPerDeleteOnlyCompaction will be resolved
2036 : // per method call. Resolved and unresolved hints are returned in separate return values.
2037 : // The files that the resolved hints apply to, are returned as compactionLevels.
2038 : func checkDeleteCompactionHints(
2039 : cmp Compare,
2040 : v *version,
2041 : hints []deleteCompactionHint,
2042 : snapshots compact.Snapshots,
2043 : exciseEnabled bool,
2044 1 : ) (levels []compactionLevel, resolved, unresolved []deleteCompactionHint) {
2045 1 : var files map[*fileMetadata]bool
2046 1 : var byLevel [numLevels][]*fileMetadata
2047 1 :
2048 1 : // Delete-only compactions can be quadratic (O(mn)) in terms of runtime
2049 1 : // where m = number of files in the delete-only compaction and n = number
2050 1 : // of resolved hints. To prevent these from growing unbounded, we cap
2051 1 : // the number of hints we resolve for one delete-only compaction. This
2052 1 : // cap only applies if exciseEnabled == true.
2053 1 : const maxHintsPerDeleteOnlyCompaction = 10
2054 1 :
2055 1 : unresolvedHints := hints[:0]
2056 1 : // Lazily populate resolvedHints, similar to files above.
2057 1 : resolvedHints := make([]deleteCompactionHint, 0)
2058 1 : for _, h := range hints {
2059 1 : // Check each compaction hint to see if it's resolvable. Resolvable
2060 1 : // hints are removed and trigger a delete-only compaction if any files
2061 1 : // in the current LSM still meet their criteria. Unresolvable hints
2062 1 : // are saved and don't trigger a delete-only compaction.
2063 1 : //
2064 1 : // When a compaction hint is created, the sequence numbers of the
2065 1 : // range tombstones and the covered file with the oldest key are
2066 1 : // recorded. The largest tombstone sequence number and the smallest
2067 1 : // file sequence number must be in the same snapshot stripe for the
2068 1 : // hint to be resolved. The below graphic models a compaction hint
2069 1 : // covering the keyspace [b, r). The hint completely contains two
2070 1 : // files, 000002 and 000003. The file 000003 contains the lowest
2071 1 : // covered sequence number at #90. The tombstone b.RANGEDEL.230:h has
2072 1 : // the highest tombstone sequence number incorporated into the hint.
2073 1 : // The hint may be resolved only once the snapshots at #100, #180 and
2074 1 : // #210 are all closed. File 000001 is not included within the hint
2075 1 : // because it extends beyond the range tombstones in user key space.
2076 1 : //
2077 1 : // 250
2078 1 : //
2079 1 : // |-b...230:h-|
2080 1 : // _____________________________________________________ snapshot #210
2081 1 : // 200 |--h.RANGEDEL.200:r--|
2082 1 : //
2083 1 : // _____________________________________________________ snapshot #180
2084 1 : //
2085 1 : // 150 +--------+
2086 1 : // +---------+ | 000003 |
2087 1 : // | 000002 | | |
2088 1 : // +_________+ | |
2089 1 : // 100_____________________|________|___________________ snapshot #100
2090 1 : // +--------+
2091 1 : // _____________________________________________________ snapshot #70
2092 1 : // +---------------+
2093 1 : // 50 | 000001 |
2094 1 : // | |
2095 1 : // +---------------+
2096 1 : // ______________________________________________________________
2097 1 : // 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
2098 1 :
2099 1 : if snapshots.Index(h.tombstoneLargestSeqNum) != snapshots.Index(h.fileSmallestSeqNum) ||
2100 1 : (len(resolvedHints) >= maxHintsPerDeleteOnlyCompaction && exciseEnabled) {
2101 1 : // Cannot resolve yet.
2102 1 : unresolvedHints = append(unresolvedHints, h)
2103 1 : continue
2104 : }
2105 :
2106 : // The hint h will be resolved and dropped, if it either affects no files at all
2107 : // or if the number of files it creates (eg. through excision) is less than or
2108 : // equal to the number of files it deletes. First, determine how many files are
2109 : // affected by this hint.
2110 1 : filesDeletedByCurrentHint := 0
2111 1 : var filesDeletedByLevel [7][]*fileMetadata
2112 1 : for l := h.tombstoneLevel + 1; l < numLevels; l++ {
2113 1 : overlaps := v.Overlaps(l, base.UserKeyBoundsEndExclusive(h.start, h.end))
2114 1 : iter := overlaps.Iter()
2115 1 :
2116 1 : for m := iter.First(); m != nil; m = iter.Next() {
2117 1 : doesHintApply := h.canDeleteOrExcise(cmp, m, snapshots, exciseEnabled)
2118 1 : if m.IsCompacting() || doesHintApply == hintDoesNotApply || files[m] {
2119 1 : continue
2120 : }
2121 1 : switch doesHintApply {
2122 1 : case hintDeletesFile:
2123 1 : filesDeletedByCurrentHint++
2124 1 : case hintExcisesFile:
2125 1 : // Account for the original file being deleted.
2126 1 : filesDeletedByCurrentHint++
2127 1 : // An excise could produce up to 2 new files. If the hint
2128 1 : // leaves a fragment of the file on the left, decrement
2129 1 : // the counter once. If the hint leaves a fragment of the
2130 1 : // file on the right, decrement the counter once.
2131 1 : if cmp(h.start, m.Smallest.UserKey) > 0 {
2132 1 : filesDeletedByCurrentHint--
2133 1 : }
2134 1 : if m.UserKeyBounds().End.IsUpperBoundFor(cmp, h.end) {
2135 1 : filesDeletedByCurrentHint--
2136 1 : }
2137 : }
2138 1 : filesDeletedByLevel[l] = append(filesDeletedByLevel[l], m)
2139 : }
2140 : }
2141 1 : if filesDeletedByCurrentHint < 0 {
2142 1 : // This hint does not delete a sufficient number of files to warrant
2143 1 : // a delete-only compaction at this stage. Drop it (ie. don't add it
2144 1 : // to either resolved or unresolved hints) so it doesn't stick around
2145 1 : // forever.
2146 1 : continue
2147 : }
2148 : // This hint will be resolved and dropped.
2149 1 : for l := h.tombstoneLevel + 1; l < numLevels; l++ {
2150 1 : byLevel[l] = append(byLevel[l], filesDeletedByLevel[l]...)
2151 1 : for _, m := range filesDeletedByLevel[l] {
2152 1 : if files == nil {
2153 1 : // Construct files lazily, assuming most calls will not
2154 1 : // produce delete-only compactions.
2155 1 : files = make(map[*fileMetadata]bool)
2156 1 : }
2157 1 : files[m] = true
2158 : }
2159 : }
2160 1 : resolvedHints = append(resolvedHints, h)
2161 : }
2162 :
2163 1 : var compactLevels []compactionLevel
2164 1 : for l, files := range byLevel {
2165 1 : if len(files) == 0 {
2166 1 : continue
2167 : }
2168 1 : compactLevels = append(compactLevels, compactionLevel{
2169 1 : level: l,
2170 1 : files: manifest.NewLevelSliceKeySorted(cmp, files),
2171 1 : })
2172 : }
2173 1 : return compactLevels, resolvedHints, unresolvedHints
2174 : }
2175 :
2176 : // compact runs one compaction and maybe schedules another call to compact.
2177 1 : func (d *DB) compact(c *compaction, errChannel chan error) {
2178 1 : pprof.Do(context.Background(), compactLabels, func(context.Context) {
2179 1 : d.mu.Lock()
2180 1 : defer d.mu.Unlock()
2181 1 : if err := d.compact1(c, errChannel); err != nil {
2182 1 : // TODO(peter): count consecutive compaction errors and backoff.
2183 1 : d.opts.EventListener.BackgroundError(err)
2184 1 : }
2185 1 : if c.isDownload {
2186 1 : d.mu.compact.downloadingCount--
2187 1 : } else {
2188 1 : d.mu.compact.compactingCount--
2189 1 : }
2190 1 : delete(d.mu.compact.inProgress, c)
2191 1 : // Add this compaction's duration to the cumulative duration. NB: This
2192 1 : // must be atomic with the above removal of c from
2193 1 : // d.mu.compact.InProgress to ensure Metrics.Compact.Duration does not
2194 1 : // miss or double count a completing compaction's duration.
2195 1 : d.mu.compact.duration += d.timeNow().Sub(c.beganAt)
2196 1 :
2197 1 : // The previous compaction may have produced too many files in a
2198 1 : // level, so reschedule another compaction if needed.
2199 1 : d.maybeScheduleCompaction()
2200 1 : d.mu.compact.cond.Broadcast()
2201 : })
2202 : }
2203 :
2204 : // cleanupVersionEdit cleans up any on-disk artifacts that were created
2205 : // for the application of a versionEdit that is no longer going to be applied.
2206 : //
2207 : // d.mu must be held when calling this method.
2208 1 : func (d *DB) cleanupVersionEdit(ve *versionEdit) {
2209 1 : obsoleteFiles := make([]*fileBacking, 0, len(ve.NewFiles))
2210 1 : deletedFiles := make(map[base.FileNum]struct{})
2211 1 : for key := range ve.DeletedFiles {
2212 1 : deletedFiles[key.FileNum] = struct{}{}
2213 1 : }
2214 1 : for i := range ve.NewFiles {
2215 1 : if ve.NewFiles[i].Meta.Virtual {
2216 0 : // We handle backing files separately.
2217 0 : continue
2218 : }
2219 1 : if _, ok := deletedFiles[ve.NewFiles[i].Meta.FileNum]; ok {
2220 0 : // This file is being moved in this ve to a different level.
2221 0 : // Don't mark it as obsolete.
2222 0 : continue
2223 : }
2224 1 : obsoleteFiles = append(obsoleteFiles, ve.NewFiles[i].Meta.PhysicalMeta().FileBacking)
2225 : }
2226 1 : for i := range ve.CreatedBackingTables {
2227 0 : if ve.CreatedBackingTables[i].IsUnused() {
2228 0 : obsoleteFiles = append(obsoleteFiles, ve.CreatedBackingTables[i])
2229 0 : }
2230 : }
2231 1 : for i := range obsoleteFiles {
2232 1 : // Add this file to zombie tables as well, as the versionSet
2233 1 : // asserts on whether every obsolete file was at one point
2234 1 : // marked zombie.
2235 1 : d.mu.versions.zombieTables[obsoleteFiles[i].DiskFileNum] = tableInfo{
2236 1 : fileInfo: fileInfo{
2237 1 : FileNum: obsoleteFiles[i].DiskFileNum,
2238 1 : FileSize: obsoleteFiles[i].Size,
2239 1 : },
2240 1 : // TODO(bilal): This is harmless if it's wrong, as it only causes
2241 1 : // incorrect accounting for the size of it in metrics. Currently
2242 1 : // all compactions only write to local files anyway except with
2243 1 : // disaggregated storage; if this becomes the norm, we should do
2244 1 : // an objprovider lookup here.
2245 1 : isLocal: true,
2246 1 : }
2247 1 : }
2248 1 : d.mu.versions.addObsoleteLocked(obsoleteFiles)
2249 : }
2250 :
2251 : // compact1 runs one compaction.
2252 : //
2253 : // d.mu must be held when calling this, but the mutex may be dropped and
2254 : // re-acquired during the course of this method.
2255 1 : func (d *DB) compact1(c *compaction, errChannel chan error) (err error) {
2256 1 : if errChannel != nil {
2257 1 : defer func() {
2258 1 : errChannel <- err
2259 1 : }()
2260 : }
2261 :
2262 1 : jobID := d.newJobIDLocked()
2263 1 : info := c.makeInfo(jobID)
2264 1 : d.opts.EventListener.CompactionBegin(info)
2265 1 : startTime := d.timeNow()
2266 1 :
2267 1 : ve, stats, err := d.runCompaction(jobID, c)
2268 1 :
2269 1 : info.Duration = d.timeNow().Sub(startTime)
2270 1 : if err == nil {
2271 1 : validateVersionEdit(ve, d.opts.Experimental.KeyValidationFunc, d.opts.Comparer.FormatKey, d.opts.Logger)
2272 1 : err = func() error {
2273 1 : var err error
2274 1 : d.mu.versions.logLock()
2275 1 : // Check if this compaction had a conflicting operation (eg. a d.excise())
2276 1 : // that necessitates it restarting from scratch. Note that since we hold
2277 1 : // the manifest lock, we don't expect this bool to change its value
2278 1 : // as only the holder of the manifest lock will ever write to it.
2279 1 : if c.cancel.Load() {
2280 1 : d.mu.versions.metrics.Compact.CancelledCount++
2281 1 : d.mu.versions.metrics.Compact.CancelledBytes += c.bytesWritten
2282 1 :
2283 1 : err = firstError(err, ErrCancelledCompaction)
2284 1 : // This is the first time we've seen a cancellation during the
2285 1 : // life of this compaction (or the original condition on err == nil
2286 1 : // would not have been true). We should delete any tables already
2287 1 : // created, as d.runCompaction did not do that.
2288 1 : d.cleanupVersionEdit(ve)
2289 1 : // logAndApply calls logUnlock. If we didn't call it, we need to call
2290 1 : // logUnlock ourselves.
2291 1 : d.mu.versions.logUnlock()
2292 1 : return err
2293 1 : }
2294 1 : return d.mu.versions.logAndApply(jobID, ve, c.metrics, false /* forceRotation */, func() []compactionInfo {
2295 1 : return d.getInProgressCompactionInfoLocked(c)
2296 1 : })
2297 : }()
2298 : }
2299 :
2300 1 : info.Done = true
2301 1 : info.Err = err
2302 1 : if err == nil {
2303 1 : for i := range ve.NewFiles {
2304 1 : e := &ve.NewFiles[i]
2305 1 : info.Output.Tables = append(info.Output.Tables, e.Meta.TableInfo())
2306 1 : }
2307 1 : d.mu.snapshots.cumulativePinnedCount += stats.CumulativePinnedKeys
2308 1 : d.mu.snapshots.cumulativePinnedSize += stats.CumulativePinnedSize
2309 1 : d.mu.versions.metrics.Keys.MissizedTombstonesCount += stats.CountMissizedDels
2310 : }
2311 :
2312 : // NB: clearing compacting state must occur before updating the read state;
2313 : // L0Sublevels initialization depends on it.
2314 1 : d.clearCompactingState(c, err != nil)
2315 1 : if err != nil && errors.Is(err, ErrCancelledCompaction) {
2316 1 : d.mu.versions.metrics.Compact.CancelledCount++
2317 1 : d.mu.versions.metrics.Compact.CancelledBytes += c.bytesWritten
2318 1 : }
2319 1 : d.mu.versions.incrementCompactions(c.kind, c.extraLevels, c.pickerMetrics)
2320 1 : d.mu.versions.incrementCompactionBytes(-c.bytesWritten)
2321 1 :
2322 1 : info.TotalDuration = d.timeNow().Sub(c.beganAt)
2323 1 : d.opts.EventListener.CompactionEnd(info)
2324 1 :
2325 1 : // Update the read state before deleting obsolete files because the
2326 1 : // read-state update will cause the previous version to be unref'd and if
2327 1 : // there are no references obsolete tables will be added to the obsolete
2328 1 : // table list.
2329 1 : if err == nil {
2330 1 : d.updateReadStateLocked(d.opts.DebugCheck)
2331 1 : d.updateTableStatsLocked(ve.NewFiles)
2332 1 : }
2333 1 : d.deleteObsoleteFiles(jobID)
2334 1 :
2335 1 : return err
2336 : }
2337 :
2338 : // runCopyCompaction runs a copy compaction where a new FileNum is created that
2339 : // is a byte-for-byte copy of the input file or span thereof in some cases. This
2340 : // is used in lieu of a move compaction when a file is being moved across the
2341 : // local/remote storage boundary. It could also be used in lieu of a rewrite
2342 : // compaction as part of a Download() call, which allows copying only a span of
2343 : // the external file, provided the file does not contain range keys or value
2344 : // blocks (see sstable.CopySpan).
2345 : //
2346 : // d.mu must be held when calling this method. The mutex will be released when
2347 : // doing IO.
2348 : func (d *DB) runCopyCompaction(
2349 : jobID JobID, c *compaction,
2350 1 : ) (ve *versionEdit, stats compact.Stats, _ error) {
2351 1 : iter := c.startLevel.files.Iter()
2352 1 : inputMeta := iter.First()
2353 1 : if iter.Next() != nil {
2354 0 : return nil, compact.Stats{}, base.AssertionFailedf("got more than one file for a move compaction")
2355 0 : }
2356 1 : if c.cancel.Load() {
2357 0 : return nil, compact.Stats{}, ErrCancelledCompaction
2358 0 : }
2359 1 : ve = &versionEdit{
2360 1 : DeletedFiles: map[deletedFileEntry]*fileMetadata{
2361 1 : {Level: c.startLevel.level, FileNum: inputMeta.FileNum}: inputMeta,
2362 1 : },
2363 1 : }
2364 1 :
2365 1 : objMeta, err := d.objProvider.Lookup(fileTypeTable, inputMeta.FileBacking.DiskFileNum)
2366 1 : if err != nil {
2367 0 : return nil, compact.Stats{}, err
2368 0 : }
2369 1 : if !objMeta.IsExternal() {
2370 1 : if objMeta.IsRemote() || !remote.ShouldCreateShared(d.opts.Experimental.CreateOnShared, c.outputLevel.level) {
2371 0 : panic("pebble: scheduled a copy compaction that is not actually moving files to shared storage")
2372 : }
2373 : // Note that based on logic in the compaction picker, we're guaranteed
2374 : // inputMeta.Virtual is false.
2375 1 : if inputMeta.Virtual {
2376 0 : panic(errors.AssertionFailedf("cannot do a copy compaction of a virtual sstable across local/remote storage"))
2377 : }
2378 : }
2379 :
2380 : // We are in the relatively more complex case where we need to copy this
2381 : // file to remote storage. Drop the db mutex while we do the copy
2382 : //
2383 : // To ease up cleanup of the local file and tracking of refs, we create
2384 : // a new FileNum. This has the potential of making the block cache less
2385 : // effective, however.
2386 1 : newMeta := &fileMetadata{
2387 1 : Size: inputMeta.Size,
2388 1 : CreationTime: inputMeta.CreationTime,
2389 1 : SmallestSeqNum: inputMeta.SmallestSeqNum,
2390 1 : LargestSeqNum: inputMeta.LargestSeqNum,
2391 1 : LargestSeqNumAbsolute: inputMeta.LargestSeqNumAbsolute,
2392 1 : Stats: inputMeta.Stats,
2393 1 : Virtual: inputMeta.Virtual,
2394 1 : SyntheticPrefixAndSuffix: inputMeta.SyntheticPrefixAndSuffix,
2395 1 : }
2396 1 : if inputMeta.HasPointKeys {
2397 1 : newMeta.ExtendPointKeyBounds(c.cmp, inputMeta.SmallestPointKey, inputMeta.LargestPointKey)
2398 1 : }
2399 1 : if inputMeta.HasRangeKeys {
2400 1 : newMeta.ExtendRangeKeyBounds(c.cmp, inputMeta.SmallestRangeKey, inputMeta.LargestRangeKey)
2401 1 : }
2402 1 : newMeta.FileNum = d.mu.versions.getNextFileNum()
2403 1 : if objMeta.IsExternal() {
2404 1 : // external -> local/shared copy. File must be virtual.
2405 1 : // We will update this size later after we produce the new backing file.
2406 1 : newMeta.InitProviderBacking(base.DiskFileNum(newMeta.FileNum), inputMeta.FileBacking.Size)
2407 1 : } else {
2408 1 : // local -> shared copy. New file is guaranteed to not be virtual.
2409 1 : newMeta.InitPhysicalBacking()
2410 1 : }
2411 :
2412 : // Before dropping the db mutex, grab a ref to the current version. This
2413 : // prevents any concurrent excises from deleting files that this compaction
2414 : // needs to read/maintain a reference to.
2415 1 : vers := d.mu.versions.currentVersion()
2416 1 : vers.Ref()
2417 1 : defer vers.UnrefLocked()
2418 1 :
2419 1 : // NB: The order here is reversed, lock after unlock. This is similar to
2420 1 : // runCompaction.
2421 1 : d.mu.Unlock()
2422 1 : defer d.mu.Lock()
2423 1 :
2424 1 : deleteOnExit := false
2425 1 : defer func() {
2426 1 : if deleteOnExit {
2427 1 : _ = d.objProvider.Remove(fileTypeTable, newMeta.FileBacking.DiskFileNum)
2428 1 : }
2429 : }()
2430 :
2431 : // If the src obj is external, we're doing an external to local/shared copy.
2432 1 : if objMeta.IsExternal() {
2433 1 : ctx := context.TODO()
2434 1 : src, err := d.objProvider.OpenForReading(
2435 1 : ctx, fileTypeTable, inputMeta.FileBacking.DiskFileNum, objstorage.OpenOptions{},
2436 1 : )
2437 1 : if err != nil {
2438 0 : return nil, compact.Stats{}, err
2439 0 : }
2440 1 : defer func() {
2441 1 : if src != nil {
2442 0 : src.Close()
2443 0 : }
2444 : }()
2445 :
2446 1 : w, _, err := d.objProvider.Create(
2447 1 : ctx, fileTypeTable, newMeta.FileBacking.DiskFileNum,
2448 1 : objstorage.CreateOptions{
2449 1 : PreferSharedStorage: remote.ShouldCreateShared(d.opts.Experimental.CreateOnShared, c.outputLevel.level),
2450 1 : },
2451 1 : )
2452 1 : if err != nil {
2453 0 : return nil, compact.Stats{}, err
2454 0 : }
2455 1 : deleteOnExit = true
2456 1 :
2457 1 : start, end := newMeta.Smallest, newMeta.Largest
2458 1 : if newMeta.SyntheticPrefixAndSuffix.HasPrefix() {
2459 1 : syntheticPrefix := newMeta.SyntheticPrefixAndSuffix.Prefix()
2460 1 : start.UserKey = syntheticPrefix.Invert(start.UserKey)
2461 1 : end.UserKey = syntheticPrefix.Invert(end.UserKey)
2462 1 : }
2463 1 : if newMeta.SyntheticPrefixAndSuffix.HasSuffix() {
2464 0 : // Extend the bounds as necessary so that the keys don't include suffixes.
2465 0 : start.UserKey = start.UserKey[:c.comparer.Split(start.UserKey)]
2466 0 : if n := c.comparer.Split(end.UserKey); n < len(end.UserKey) {
2467 0 : end = base.MakeRangeDeleteSentinelKey(c.comparer.ImmediateSuccessor(nil, end.UserKey[:n]))
2468 0 : }
2469 : }
2470 :
2471 : // NB: external files are always virtual.
2472 1 : var wrote uint64
2473 1 : err = d.fileCache.withVirtualReader(inputMeta.VirtualMeta(), func(r sstable.VirtualReader) error {
2474 1 : var err error
2475 1 : wrote, err = sstable.CopySpan(ctx,
2476 1 : src, r.UnsafeReader(), d.opts.MakeReaderOptions(),
2477 1 : w, d.opts.MakeWriterOptions(c.outputLevel.level, d.TableFormat()),
2478 1 : start, end,
2479 1 : )
2480 1 : return err
2481 1 : })
2482 :
2483 1 : src = nil // We passed src to CopySpan; it's responsible for closing it.
2484 1 : if err != nil {
2485 1 : if errors.Is(err, sstable.ErrEmptySpan) {
2486 1 : // The virtual table was empty. Just remove the backing file.
2487 1 : // Note that deleteOnExit is true so we will delete the created object.
2488 1 : c.metrics = map[int]*LevelMetrics{
2489 1 : c.outputLevel.level: {
2490 1 : BytesIn: inputMeta.Size,
2491 1 : },
2492 1 : }
2493 1 : return ve, compact.Stats{}, nil
2494 1 : }
2495 0 : return nil, compact.Stats{}, err
2496 : }
2497 1 : newMeta.FileBacking.Size = wrote
2498 1 : newMeta.Size = wrote
2499 1 : } else {
2500 1 : _, err := d.objProvider.LinkOrCopyFromLocal(context.TODO(), d.opts.FS,
2501 1 : d.objProvider.Path(objMeta), fileTypeTable, newMeta.FileBacking.DiskFileNum,
2502 1 : objstorage.CreateOptions{PreferSharedStorage: true})
2503 1 : if err != nil {
2504 0 : return nil, compact.Stats{}, err
2505 0 : }
2506 1 : deleteOnExit = true
2507 : }
2508 1 : ve.NewFiles = []newFileEntry{{
2509 1 : Level: c.outputLevel.level,
2510 1 : Meta: newMeta,
2511 1 : }}
2512 1 : if newMeta.Virtual {
2513 1 : ve.CreatedBackingTables = []*fileBacking{newMeta.FileBacking}
2514 1 : }
2515 1 : c.metrics = map[int]*LevelMetrics{
2516 1 : c.outputLevel.level: {
2517 1 : BytesIn: inputMeta.Size,
2518 1 : BytesCompacted: newMeta.Size,
2519 1 : TablesCompacted: 1,
2520 1 : },
2521 1 : }
2522 1 :
2523 1 : if err := d.objProvider.Sync(); err != nil {
2524 0 : return nil, compact.Stats{}, err
2525 0 : }
2526 1 : deleteOnExit = false
2527 1 : return ve, compact.Stats{}, nil
2528 : }
2529 :
2530 : // applyHintOnFile applies a deleteCompactionHint to a file, and updates the
2531 : // versionEdit accordingly. It returns a list of new files that were created
2532 : // if the hint was applied partially to a file (eg. through an excise as opposed
2533 : // to an outright deletion). levelMetrics is kept up-to-date with the number
2534 : // of tables deleted or excised.
2535 : func (d *DB) applyHintOnFile(
2536 : h deleteCompactionHint,
2537 : f *fileMetadata,
2538 : level int,
2539 : levelMetrics *LevelMetrics,
2540 : ve *versionEdit,
2541 : hintOverlap deletionHintOverlap,
2542 1 : ) (newFiles []manifest.NewFileEntry, err error) {
2543 1 : if hintOverlap == hintDoesNotApply {
2544 0 : return nil, nil
2545 0 : }
2546 :
2547 : // The hint overlaps with at least part of the file.
2548 1 : if hintOverlap == hintDeletesFile {
2549 1 : // The hint deletes the entirety of this file.
2550 1 : ve.DeletedFiles[deletedFileEntry{
2551 1 : Level: level,
2552 1 : FileNum: f.FileNum,
2553 1 : }] = f
2554 1 : levelMetrics.TablesDeleted++
2555 1 : return nil, nil
2556 1 : }
2557 : // The hint overlaps with only a part of the file, not the entirety of it. We need
2558 : // to use d.excise. (hintOverlap == hintExcisesFile)
2559 1 : if d.FormatMajorVersion() < FormatVirtualSSTables {
2560 0 : panic("pebble: delete-only compaction hint excising a file is not supported in this version")
2561 : }
2562 :
2563 1 : levelMetrics.TablesExcised++
2564 1 : newFiles, err = d.excise(context.TODO(), base.UserKeyBoundsEndExclusive(h.start, h.end), f, ve, level)
2565 1 : if err != nil {
2566 0 : return nil, errors.Wrap(err, "error when running excise for delete-only compaction")
2567 0 : }
2568 1 : if _, ok := ve.DeletedFiles[deletedFileEntry{
2569 1 : Level: level,
2570 1 : FileNum: f.FileNum,
2571 1 : }]; !ok {
2572 0 : panic("pebble: delete-only compaction hint overlapping a file did not excise that file")
2573 : }
2574 1 : return newFiles, nil
2575 : }
2576 :
2577 : func (d *DB) runDeleteOnlyCompactionForLevel(
2578 : cl compactionLevel,
2579 : levelMetrics *LevelMetrics,
2580 : ve *versionEdit,
2581 : snapshots compact.Snapshots,
2582 : fragments []deleteCompactionHintFragment,
2583 : exciseEnabled bool,
2584 1 : ) error {
2585 1 : curFragment := 0
2586 1 : iter := cl.files.Iter()
2587 1 : if cl.level == 0 {
2588 0 : panic("cannot run delete-only compaction for L0")
2589 : }
2590 :
2591 : // Outer loop loops on files. Middle loop loops on fragments. Inner loop
2592 : // loops on raw fragments of hints. Number of fragments are bounded by
2593 : // the number of hints this compaction was created with, which is capped
2594 : // in the compaction picker to avoid very CPU-hot loops here.
2595 1 : for f := iter.First(); f != nil; f = iter.Next() {
2596 1 : // curFile usually matches f, except if f got excised in which case
2597 1 : // it maps to a virtual file that replaces f, or nil if f got removed
2598 1 : // in its entirety.
2599 1 : curFile := f
2600 1 : for curFragment < len(fragments) && d.cmp(fragments[curFragment].start, f.Smallest.UserKey) <= 0 {
2601 1 : curFragment++
2602 1 : }
2603 1 : if curFragment > 0 {
2604 1 : curFragment--
2605 1 : }
2606 :
2607 1 : for ; curFragment < len(fragments); curFragment++ {
2608 1 : if f.UserKeyBounds().End.CompareUpperBounds(d.cmp, base.UserKeyInclusive(fragments[curFragment].start)) < 0 {
2609 1 : break
2610 : }
2611 : // Process all overlapping hints with this file. Note that applying
2612 : // a hint twice is idempotent; curFile should have already been excised
2613 : // the first time, resulting in no change the second time.
2614 1 : for _, h := range fragments[curFragment].hints {
2615 1 : if h.tombstoneLevel >= cl.level {
2616 1 : // We cannot excise out the deletion tombstone itself, or anything
2617 1 : // above it.
2618 1 : continue
2619 : }
2620 1 : hintOverlap := h.canDeleteOrExcise(d.cmp, curFile, snapshots, exciseEnabled)
2621 1 : if hintOverlap == hintDoesNotApply {
2622 1 : continue
2623 : }
2624 1 : newFiles, err := d.applyHintOnFile(h, curFile, cl.level, levelMetrics, ve, hintOverlap)
2625 1 : if err != nil {
2626 0 : return err
2627 0 : }
2628 1 : if _, ok := ve.DeletedFiles[manifest.DeletedFileEntry{Level: cl.level, FileNum: curFile.FileNum}]; ok {
2629 1 : curFile = nil
2630 1 : }
2631 1 : if len(newFiles) > 0 {
2632 1 : curFile = newFiles[len(newFiles)-1].Meta
2633 1 : } else if curFile == nil {
2634 1 : // Nothing remains of the file.
2635 1 : break
2636 : }
2637 : }
2638 1 : if curFile == nil {
2639 1 : // Nothing remains of the file.
2640 1 : break
2641 : }
2642 : }
2643 1 : if _, ok := ve.DeletedFiles[deletedFileEntry{
2644 1 : Level: cl.level,
2645 1 : FileNum: f.FileNum,
2646 1 : }]; !ok {
2647 0 : panic("pebble: delete-only compaction scheduled with hints that did not delete or excise a file")
2648 : }
2649 : }
2650 1 : return nil
2651 : }
2652 :
2653 : // deleteCompactionHintFragment represents a fragment of the key space and
2654 : // contains a set of deleteCompactionHints that apply to that fragment; a
2655 : // fragment starts at the start field and ends where the next fragment starts.
2656 : type deleteCompactionHintFragment struct {
2657 : start []byte
2658 : hints []deleteCompactionHint
2659 : }
2660 :
2661 : // Delete compaction hints can overlap with each other, and multiple fragments
2662 : // can apply to a single file. This function takes a list of hints and fragments
2663 : // them, to make it easier to apply them to non-overlapping files occupying a level;
2664 : // that way, files and hint fragments can be iterated on in lockstep, while efficiently
2665 : // being able to apply all hints overlapping with a given file.
2666 : func fragmentDeleteCompactionHints(
2667 : cmp Compare, hints []deleteCompactionHint,
2668 1 : ) []deleteCompactionHintFragment {
2669 1 : fragments := make([]deleteCompactionHintFragment, 0, len(hints)*2)
2670 1 : for i := range hints {
2671 1 : fragments = append(fragments, deleteCompactionHintFragment{start: hints[i].start},
2672 1 : deleteCompactionHintFragment{start: hints[i].end})
2673 1 : }
2674 1 : slices.SortFunc(fragments, func(i, j deleteCompactionHintFragment) int {
2675 1 : return cmp(i.start, j.start)
2676 1 : })
2677 1 : fragments = slices.CompactFunc(fragments, func(i, j deleteCompactionHintFragment) bool {
2678 1 : return bytes.Equal(i.start, j.start)
2679 1 : })
2680 1 : for _, h := range hints {
2681 1 : startIdx := sort.Search(len(fragments), func(i int) bool {
2682 1 : return cmp(fragments[i].start, h.start) >= 0
2683 1 : })
2684 1 : endIdx := sort.Search(len(fragments), func(i int) bool {
2685 1 : return cmp(fragments[i].start, h.end) >= 0
2686 1 : })
2687 1 : for i := startIdx; i < endIdx; i++ {
2688 1 : fragments[i].hints = append(fragments[i].hints, h)
2689 1 : }
2690 : }
2691 1 : return fragments
2692 : }
2693 :
2694 : // Runs a delete-only compaction.
2695 : //
2696 : // d.mu must *not* be held when calling this.
2697 : func (d *DB) runDeleteOnlyCompaction(
2698 : jobID JobID, c *compaction, snapshots compact.Snapshots,
2699 1 : ) (ve *versionEdit, stats compact.Stats, retErr error) {
2700 1 : c.metrics = make(map[int]*LevelMetrics, len(c.inputs))
2701 1 : fragments := fragmentDeleteCompactionHints(d.cmp, c.deletionHints)
2702 1 : ve = &versionEdit{
2703 1 : DeletedFiles: map[deletedFileEntry]*fileMetadata{},
2704 1 : }
2705 1 : for _, cl := range c.inputs {
2706 1 : levelMetrics := &LevelMetrics{}
2707 1 : if err := d.runDeleteOnlyCompactionForLevel(cl, levelMetrics, ve, snapshots, fragments, c.exciseEnabled); err != nil {
2708 0 : return nil, stats, err
2709 0 : }
2710 1 : c.metrics[cl.level] = levelMetrics
2711 : }
2712 : // Remove any files that were added and deleted in the same versionEdit.
2713 1 : ve.NewFiles = slices.DeleteFunc(ve.NewFiles, func(e manifest.NewFileEntry) bool {
2714 1 : deletedFileEntry := manifest.DeletedFileEntry{Level: e.Level, FileNum: e.Meta.FileNum}
2715 1 : if _, deleted := ve.DeletedFiles[deletedFileEntry]; deleted {
2716 1 : delete(ve.DeletedFiles, deletedFileEntry)
2717 1 : return true
2718 1 : }
2719 1 : return false
2720 : })
2721 : // Remove any entries from CreatedBackingTables that are not used in any
2722 : // NewFiles.
2723 1 : usedBackingFiles := make(map[base.DiskFileNum]struct{})
2724 1 : for _, e := range ve.NewFiles {
2725 1 : if e.Meta.Virtual {
2726 1 : usedBackingFiles[e.Meta.FileBacking.DiskFileNum] = struct{}{}
2727 1 : }
2728 : }
2729 1 : ve.CreatedBackingTables = slices.DeleteFunc(ve.CreatedBackingTables, func(b *fileBacking) bool {
2730 1 : _, used := usedBackingFiles[b.DiskFileNum]
2731 1 : return !used
2732 1 : })
2733 : // Refresh the disk available statistic whenever a compaction/flush
2734 : // completes, before re-acquiring the mutex.
2735 1 : d.calculateDiskAvailableBytes()
2736 1 : return ve, stats, nil
2737 : }
2738 :
2739 : func (d *DB) runMoveCompaction(
2740 : jobID JobID, c *compaction,
2741 1 : ) (ve *versionEdit, stats compact.Stats, _ error) {
2742 1 : iter := c.startLevel.files.Iter()
2743 1 : meta := iter.First()
2744 1 : if iter.Next() != nil {
2745 0 : return nil, stats, base.AssertionFailedf("got more than one file for a move compaction")
2746 0 : }
2747 1 : if c.cancel.Load() {
2748 0 : return ve, stats, ErrCancelledCompaction
2749 0 : }
2750 1 : c.metrics = map[int]*LevelMetrics{
2751 1 : c.outputLevel.level: {
2752 1 : BytesMoved: meta.Size,
2753 1 : TablesMoved: 1,
2754 1 : },
2755 1 : }
2756 1 : ve = &versionEdit{
2757 1 : DeletedFiles: map[deletedFileEntry]*fileMetadata{
2758 1 : {Level: c.startLevel.level, FileNum: meta.FileNum}: meta,
2759 1 : },
2760 1 : NewFiles: []newFileEntry{
2761 1 : {Level: c.outputLevel.level, Meta: meta},
2762 1 : },
2763 1 : }
2764 1 :
2765 1 : return ve, stats, nil
2766 : }
2767 :
2768 : // runCompaction runs a compaction that produces new on-disk tables from
2769 : // memtables or old on-disk tables.
2770 : //
2771 : // runCompaction cannot be used for compactionKindIngestedFlushable.
2772 : //
2773 : // d.mu must be held when calling this, but the mutex may be dropped and
2774 : // re-acquired during the course of this method.
2775 : func (d *DB) runCompaction(
2776 : jobID JobID, c *compaction,
2777 1 : ) (ve *versionEdit, stats compact.Stats, retErr error) {
2778 1 : if c.cancel.Load() {
2779 1 : return ve, stats, ErrCancelledCompaction
2780 1 : }
2781 1 : switch c.kind {
2782 1 : case compactionKindDeleteOnly:
2783 1 : // Before dropping the db mutex, grab a ref to the current version. This
2784 1 : // prevents any concurrent excises from deleting files that this compaction
2785 1 : // needs to read/maintain a reference to.
2786 1 : //
2787 1 : // Note that delete-only compactions can call excise(), which needs to be able
2788 1 : // to read these files.
2789 1 : vers := d.mu.versions.currentVersion()
2790 1 : vers.Ref()
2791 1 : defer vers.UnrefLocked()
2792 1 : // Release the d.mu lock while doing I/O.
2793 1 : // Note the unusual order: Unlock and then Lock.
2794 1 : snapshots := d.mu.snapshots.toSlice()
2795 1 : d.mu.Unlock()
2796 1 : defer d.mu.Lock()
2797 1 : return d.runDeleteOnlyCompaction(jobID, c, snapshots)
2798 1 : case compactionKindMove:
2799 1 : return d.runMoveCompaction(jobID, c)
2800 1 : case compactionKindCopy:
2801 1 : return d.runCopyCompaction(jobID, c)
2802 0 : case compactionKindIngestedFlushable:
2803 0 : panic("pebble: runCompaction cannot handle compactionKindIngestedFlushable.")
2804 : }
2805 :
2806 1 : snapshots := d.mu.snapshots.toSlice()
2807 1 :
2808 1 : if c.flushing == nil {
2809 1 : // Before dropping the db mutex, grab a ref to the current version. This
2810 1 : // prevents any concurrent excises from deleting files that this compaction
2811 1 : // needs to read/maintain a reference to.
2812 1 : //
2813 1 : // Note that unlike user iterators, compactionIter does not maintain a ref
2814 1 : // of the version or read state.
2815 1 : vers := d.mu.versions.currentVersion()
2816 1 : vers.Ref()
2817 1 : defer vers.UnrefLocked()
2818 1 : }
2819 :
2820 : // The table is typically written at the maximum allowable format implied by
2821 : // the current format major version of the DB, but Options may define
2822 : // additional constraints.
2823 1 : tableFormat := d.TableFormat()
2824 1 :
2825 1 : // Release the d.mu lock while doing I/O.
2826 1 : // Note the unusual order: Unlock and then Lock.
2827 1 : d.mu.Unlock()
2828 1 : defer d.mu.Lock()
2829 1 :
2830 1 : result := d.compactAndWrite(jobID, c, snapshots, tableFormat)
2831 1 : if result.Err == nil {
2832 1 : ve, result.Err = c.makeVersionEdit(result)
2833 1 : }
2834 1 : if result.Err != nil {
2835 1 : // Delete any created tables.
2836 1 : obsoleteFiles := make([]*fileBacking, 0, len(result.Tables))
2837 1 : d.mu.Lock()
2838 1 : for i := range result.Tables {
2839 1 : backing := &fileBacking{
2840 1 : DiskFileNum: result.Tables[i].ObjMeta.DiskFileNum,
2841 1 : Size: result.Tables[i].WriterMeta.Size,
2842 1 : }
2843 1 : obsoleteFiles = append(obsoleteFiles, backing)
2844 1 : // Add this file to zombie tables as well, as the versionSet
2845 1 : // asserts on whether every obsolete file was at one point
2846 1 : // marked zombie.
2847 1 : d.mu.versions.zombieTables[backing.DiskFileNum] = tableInfo{
2848 1 : fileInfo: fileInfo{
2849 1 : FileNum: backing.DiskFileNum,
2850 1 : FileSize: backing.Size,
2851 1 : },
2852 1 : isLocal: true,
2853 1 : }
2854 1 : }
2855 1 : d.mu.versions.addObsoleteLocked(obsoleteFiles)
2856 1 : d.mu.Unlock()
2857 : }
2858 : // Refresh the disk available statistic whenever a compaction/flush
2859 : // completes, before re-acquiring the mutex.
2860 1 : d.calculateDiskAvailableBytes()
2861 1 : return ve, result.Stats, result.Err
2862 : }
2863 :
2864 : // compactAndWrite runs the data part of a compaction, where we set up a
2865 : // compaction iterator and use it to write output tables.
2866 : func (d *DB) compactAndWrite(
2867 : jobID JobID, c *compaction, snapshots compact.Snapshots, tableFormat sstable.TableFormat,
2868 1 : ) (result compact.Result) {
2869 1 : // Compactions use a pool of buffers to read blocks, avoiding polluting the
2870 1 : // block cache with blocks that will not be read again. We initialize the
2871 1 : // buffer pool with a size 12. This initial size does not need to be
2872 1 : // accurate, because the pool will grow to accommodate the maximum number of
2873 1 : // blocks allocated at a given time over the course of the compaction. But
2874 1 : // choosing a size larger than that working set avoids any additional
2875 1 : // allocations to grow the size of the pool over the course of iteration.
2876 1 : //
2877 1 : // Justification for initial size 12: In a two-level compaction, at any
2878 1 : // given moment we'll have 2 index blocks in-use and 2 data blocks in-use.
2879 1 : // Additionally, when decoding a compressed block, we'll temporarily
2880 1 : // allocate 1 additional block to hold the compressed buffer. In the worst
2881 1 : // case that all input sstables have two-level index blocks (+2), value
2882 1 : // blocks (+2), range deletion blocks (+n) and range key blocks (+n), we'll
2883 1 : // additionally require 2n+4 blocks where n is the number of input sstables.
2884 1 : // Range deletion and range key blocks are relatively rare, and the cost of
2885 1 : // an additional allocation or two over the course of the compaction is
2886 1 : // considered to be okay. A larger initial size would cause the pool to hold
2887 1 : // on to more memory, even when it's not in-use because the pool will
2888 1 : // recycle buffers up to the current capacity of the pool. The memory use of
2889 1 : // a 12-buffer pool is expected to be within reason, even if all the buffers
2890 1 : // grow to the typical size of an index block (256 KiB) which would
2891 1 : // translate to 3 MiB per compaction.
2892 1 : c.bufferPool.Init(12)
2893 1 : defer c.bufferPool.Release()
2894 1 :
2895 1 : pointIter, rangeDelIter, rangeKeyIter, err := c.newInputIters(d.newIters, d.tableNewRangeKeyIter)
2896 1 : defer func() {
2897 1 : for _, closer := range c.closers {
2898 1 : closer.FragmentIterator.Close()
2899 1 : }
2900 : }()
2901 1 : if err != nil {
2902 0 : return compact.Result{Err: err}
2903 0 : }
2904 1 : c.allowedZeroSeqNum = c.allowZeroSeqNum()
2905 1 : cfg := compact.IterConfig{
2906 1 : Comparer: c.comparer,
2907 1 : Merge: d.merge,
2908 1 : TombstoneElision: c.delElision,
2909 1 : RangeKeyElision: c.rangeKeyElision,
2910 1 : Snapshots: snapshots,
2911 1 : AllowZeroSeqNum: c.allowedZeroSeqNum,
2912 1 : IneffectualSingleDeleteCallback: d.opts.Experimental.IneffectualSingleDeleteCallback,
2913 1 : SingleDeleteInvariantViolationCallback: d.opts.Experimental.SingleDeleteInvariantViolationCallback,
2914 1 : }
2915 1 : iter := compact.NewIter(cfg, pointIter, rangeDelIter, rangeKeyIter)
2916 1 :
2917 1 : runnerCfg := compact.RunnerConfig{
2918 1 : CompactionBounds: base.UserKeyBoundsFromInternal(c.smallest, c.largest),
2919 1 : L0SplitKeys: c.l0Limits,
2920 1 : Grandparents: c.grandparents,
2921 1 : MaxGrandparentOverlapBytes: c.maxOverlapBytes,
2922 1 : TargetOutputFileSize: c.maxOutputFileSize,
2923 1 : Slot: c.slot,
2924 1 : IteratorStats: &c.stats,
2925 1 : }
2926 1 : runner := compact.NewRunner(runnerCfg, iter)
2927 1 : for runner.MoreDataToWrite() {
2928 1 : if c.cancel.Load() {
2929 1 : return runner.Finish().WithError(ErrCancelledCompaction)
2930 1 : }
2931 : // Create a new table.
2932 1 : writerOpts := d.opts.MakeWriterOptions(c.outputLevel.level, tableFormat)
2933 1 : objMeta, tw, cpuWorkHandle, err := d.newCompactionOutput(jobID, c, writerOpts)
2934 1 : if err != nil {
2935 1 : return runner.Finish().WithError(err)
2936 1 : }
2937 1 : runner.WriteTable(objMeta, tw)
2938 1 : d.opts.Experimental.CPUWorkPermissionGranter.CPUWorkDone(cpuWorkHandle)
2939 : }
2940 1 : result = runner.Finish()
2941 1 : if result.Err == nil {
2942 1 : result.Err = d.objProvider.Sync()
2943 1 : }
2944 1 : return result
2945 : }
2946 :
2947 : // makeVersionEdit creates the version edit for a compaction, based on the
2948 : // tables in compact.Result.
2949 1 : func (c *compaction) makeVersionEdit(result compact.Result) (*versionEdit, error) {
2950 1 : ve := &versionEdit{
2951 1 : DeletedFiles: map[deletedFileEntry]*fileMetadata{},
2952 1 : }
2953 1 : for _, cl := range c.inputs {
2954 1 : iter := cl.files.Iter()
2955 1 : for f := iter.First(); f != nil; f = iter.Next() {
2956 1 : ve.DeletedFiles[deletedFileEntry{
2957 1 : Level: cl.level,
2958 1 : FileNum: f.FileNum,
2959 1 : }] = f
2960 1 : }
2961 : }
2962 :
2963 1 : startLevelBytes := c.startLevel.files.SizeSum()
2964 1 : outputMetrics := &LevelMetrics{
2965 1 : BytesIn: startLevelBytes,
2966 1 : BytesRead: c.outputLevel.files.SizeSum(),
2967 1 : }
2968 1 : if len(c.extraLevels) > 0 {
2969 1 : outputMetrics.BytesIn += c.extraLevels[0].files.SizeSum()
2970 1 : }
2971 1 : outputMetrics.BytesRead += outputMetrics.BytesIn
2972 1 :
2973 1 : c.metrics = map[int]*LevelMetrics{
2974 1 : c.outputLevel.level: outputMetrics,
2975 1 : }
2976 1 : if len(c.flushing) == 0 && c.metrics[c.startLevel.level] == nil {
2977 1 : c.metrics[c.startLevel.level] = &LevelMetrics{}
2978 1 : }
2979 1 : if len(c.extraLevels) > 0 {
2980 1 : c.metrics[c.extraLevels[0].level] = &LevelMetrics{}
2981 1 : outputMetrics.MultiLevel.BytesInTop = startLevelBytes
2982 1 : outputMetrics.MultiLevel.BytesIn = outputMetrics.BytesIn
2983 1 : outputMetrics.MultiLevel.BytesRead = outputMetrics.BytesRead
2984 1 : }
2985 :
2986 1 : inputLargestSeqNumAbsolute := c.inputLargestSeqNumAbsolute()
2987 1 : ve.NewFiles = make([]newFileEntry, len(result.Tables))
2988 1 : for i := range result.Tables {
2989 1 : t := &result.Tables[i]
2990 1 :
2991 1 : fileMeta := &fileMetadata{
2992 1 : FileNum: base.PhysicalTableFileNum(t.ObjMeta.DiskFileNum),
2993 1 : CreationTime: t.CreationTime.Unix(),
2994 1 : Size: t.WriterMeta.Size,
2995 1 : SmallestSeqNum: t.WriterMeta.SmallestSeqNum,
2996 1 : LargestSeqNum: t.WriterMeta.LargestSeqNum,
2997 1 : }
2998 1 : if c.flushing == nil {
2999 1 : // Set the file's LargestSeqNumAbsolute to be the maximum value of any
3000 1 : // of the compaction's input sstables.
3001 1 : // TODO(jackson): This could be narrowed to be the maximum of input
3002 1 : // sstables that overlap the output sstable's key range.
3003 1 : fileMeta.LargestSeqNumAbsolute = inputLargestSeqNumAbsolute
3004 1 : } else {
3005 1 : fileMeta.LargestSeqNumAbsolute = t.WriterMeta.LargestSeqNum
3006 1 : }
3007 1 : fileMeta.InitPhysicalBacking()
3008 1 :
3009 1 : // If the file didn't contain any range deletions, we can fill its
3010 1 : // table stats now, avoiding unnecessarily loading the table later.
3011 1 : maybeSetStatsFromProperties(
3012 1 : fileMeta.PhysicalMeta(), &t.WriterMeta.Properties,
3013 1 : )
3014 1 :
3015 1 : if t.WriterMeta.HasPointKeys {
3016 1 : fileMeta.ExtendPointKeyBounds(c.cmp, t.WriterMeta.SmallestPoint, t.WriterMeta.LargestPoint)
3017 1 : }
3018 1 : if t.WriterMeta.HasRangeDelKeys {
3019 1 : fileMeta.ExtendPointKeyBounds(c.cmp, t.WriterMeta.SmallestRangeDel, t.WriterMeta.LargestRangeDel)
3020 1 : }
3021 1 : if t.WriterMeta.HasRangeKeys {
3022 1 : fileMeta.ExtendRangeKeyBounds(c.cmp, t.WriterMeta.SmallestRangeKey, t.WriterMeta.LargestRangeKey)
3023 1 : }
3024 :
3025 1 : ve.NewFiles[i] = newFileEntry{
3026 1 : Level: c.outputLevel.level,
3027 1 : Meta: fileMeta,
3028 1 : }
3029 1 :
3030 1 : // Update metrics.
3031 1 : if c.flushing == nil {
3032 1 : outputMetrics.TablesCompacted++
3033 1 : outputMetrics.BytesCompacted += fileMeta.Size
3034 1 : } else {
3035 1 : outputMetrics.TablesFlushed++
3036 1 : outputMetrics.BytesFlushed += fileMeta.Size
3037 1 : }
3038 1 : outputMetrics.Size += int64(fileMeta.Size)
3039 1 : outputMetrics.NumFiles++
3040 1 : outputMetrics.Additional.BytesWrittenDataBlocks += t.WriterMeta.Properties.DataSize
3041 1 : outputMetrics.Additional.BytesWrittenValueBlocks += t.WriterMeta.Properties.ValueBlocksSize
3042 : }
3043 :
3044 : // Sanity check that the tables are ordered and don't overlap.
3045 1 : for i := 1; i < len(ve.NewFiles); i++ {
3046 1 : if ve.NewFiles[i-1].Meta.UserKeyBounds().End.IsUpperBoundFor(c.cmp, ve.NewFiles[i].Meta.Smallest.UserKey) {
3047 0 : return nil, base.AssertionFailedf("pebble: compaction output tables overlap: %s and %s",
3048 0 : ve.NewFiles[i-1].Meta.DebugString(c.formatKey, true),
3049 0 : ve.NewFiles[i].Meta.DebugString(c.formatKey, true),
3050 0 : )
3051 0 : }
3052 : }
3053 :
3054 1 : return ve, nil
3055 : }
3056 :
3057 : // newCompactionOutput creates an object for a new table produced by a
3058 : // compaction or flush.
3059 : func (d *DB) newCompactionOutput(
3060 : jobID JobID, c *compaction, writerOpts sstable.WriterOptions,
3061 1 : ) (objstorage.ObjectMetadata, sstable.RawWriter, CPUWorkHandle, error) {
3062 1 : diskFileNum := d.mu.versions.getNextDiskFileNum()
3063 1 :
3064 1 : var writeCategory vfs.DiskWriteCategory
3065 1 : if d.opts.EnableSQLRowSpillMetrics {
3066 0 : // In the scenario that the Pebble engine is used for SQL row spills the
3067 0 : // data written to the memtable will correspond to spills to disk and
3068 0 : // should be categorized as such.
3069 0 : writeCategory = "sql-row-spill"
3070 1 : } else if c.kind == compactionKindFlush {
3071 1 : writeCategory = "pebble-memtable-flush"
3072 1 : } else {
3073 1 : writeCategory = "pebble-compaction"
3074 1 : }
3075 :
3076 1 : var reason string
3077 1 : if c.kind == compactionKindFlush {
3078 1 : reason = "flushing"
3079 1 : } else {
3080 1 : reason = "compacting"
3081 1 : }
3082 :
3083 1 : ctx := context.TODO()
3084 1 : if objiotracing.Enabled {
3085 0 : ctx = objiotracing.WithLevel(ctx, c.outputLevel.level)
3086 0 : if c.kind == compactionKindFlush {
3087 0 : ctx = objiotracing.WithReason(ctx, objiotracing.ForFlush)
3088 0 : } else {
3089 0 : ctx = objiotracing.WithReason(ctx, objiotracing.ForCompaction)
3090 0 : }
3091 : }
3092 :
3093 : // Prefer shared storage if present.
3094 1 : createOpts := objstorage.CreateOptions{
3095 1 : PreferSharedStorage: remote.ShouldCreateShared(d.opts.Experimental.CreateOnShared, c.outputLevel.level),
3096 1 : WriteCategory: writeCategory,
3097 1 : }
3098 1 : writable, objMeta, err := d.objProvider.Create(ctx, fileTypeTable, diskFileNum, createOpts)
3099 1 : if err != nil {
3100 1 : return objstorage.ObjectMetadata{}, nil, nil, err
3101 1 : }
3102 :
3103 1 : if c.kind != compactionKindFlush {
3104 1 : writable = &compactionWritable{
3105 1 : Writable: writable,
3106 1 : versions: d.mu.versions,
3107 1 : written: &c.bytesWritten,
3108 1 : }
3109 1 : }
3110 1 : d.opts.EventListener.TableCreated(TableCreateInfo{
3111 1 : JobID: int(jobID),
3112 1 : Reason: reason,
3113 1 : Path: d.objProvider.Path(objMeta),
3114 1 : FileNum: diskFileNum,
3115 1 : })
3116 1 :
3117 1 : writerOpts.SetInternal(sstableinternal.WriterOptions{
3118 1 : CacheOpts: sstableinternal.CacheOptions{
3119 1 : Cache: d.opts.Cache,
3120 1 : CacheID: d.cacheID,
3121 1 : FileNum: diskFileNum,
3122 1 : },
3123 1 : })
3124 1 :
3125 1 : const MaxFileWriteAdditionalCPUTime = time.Millisecond * 100
3126 1 : cpuWorkHandle := d.opts.Experimental.CPUWorkPermissionGranter.GetPermission(
3127 1 : MaxFileWriteAdditionalCPUTime,
3128 1 : )
3129 1 : writerOpts.Parallelism =
3130 1 : d.opts.Experimental.MaxWriterConcurrency > 0 &&
3131 1 : (cpuWorkHandle.Permitted() || d.opts.Experimental.ForceWriterParallelism)
3132 1 :
3133 1 : // TODO(jackson): Make the compaction body generic over the RawWriter type,
3134 1 : // so that we don't need to pay the cost of dynamic dispatch?
3135 1 : tw := sstable.NewRawWriter(writable, writerOpts)
3136 1 : return objMeta, tw, cpuWorkHandle, nil
3137 : }
3138 :
3139 : // validateVersionEdit validates that start and end keys across new and deleted
3140 : // files in a versionEdit pass the given validation function.
3141 : func validateVersionEdit(
3142 : ve *versionEdit, validateFn func([]byte) error, format base.FormatKey, logger Logger,
3143 1 : ) {
3144 1 : validateKey := func(f *manifest.FileMetadata, key []byte) {
3145 1 : if err := validateFn(key); err != nil {
3146 1 : logger.Fatalf("pebble: version edit validation failed (key=%s file=%s): %v", format(key), f, err)
3147 1 : }
3148 : }
3149 :
3150 : // Validate both new and deleted files.
3151 1 : for _, f := range ve.NewFiles {
3152 1 : validateKey(f.Meta, f.Meta.Smallest.UserKey)
3153 1 : validateKey(f.Meta, f.Meta.Largest.UserKey)
3154 1 : }
3155 1 : for _, m := range ve.DeletedFiles {
3156 1 : validateKey(m, m.Smallest.UserKey)
3157 1 : validateKey(m, m.Largest.UserKey)
3158 1 : }
3159 : }
|
