Line data Source code
1 : // Copyright 2020 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 : "context"
9 : "fmt"
10 : "math"
11 :
12 : "github.com/cockroachdb/errors"
13 : "github.com/cockroachdb/pebble/internal/base"
14 : "github.com/cockroachdb/pebble/internal/invariants"
15 : "github.com/cockroachdb/pebble/internal/keyspan"
16 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
17 : "github.com/cockroachdb/pebble/internal/manifest"
18 : "github.com/cockroachdb/pebble/sstable"
19 : "github.com/cockroachdb/pebble/sstable/block"
20 : )
21 :
22 : // In-memory statistics about tables help inform compaction picking, but may
23 : // be expensive to calculate or load from disk. Every time a database is
24 : // opened, these statistics must be reloaded or recalculated. To minimize
25 : // impact on user activity and compactions, we load these statistics
26 : // asynchronously in the background and store loaded statistics in each
27 : // table's *TableMetadata.
28 : //
29 : // This file implements the asynchronous loading of statistics by maintaining
30 : // a list of files that require statistics, alongside their LSM levels.
31 : // Whenever new files are added to the LSM, the files are appended to
32 : // d.mu.tableStats.pending. If a stats collection job is not currently
33 : // running, one is started in a separate goroutine.
34 : //
35 : // The stats collection job grabs and clears the pending list, computes table
36 : // statistics relative to the current readState and updates the tables' file
37 : // metadata. New pending files may accumulate during a stats collection job,
38 : // so a completing job triggers a new job if necessary. Only one job runs at a
39 : // time.
40 : //
41 : // When an existing database is opened, all files lack in-memory statistics.
42 : // These files' stats are loaded incrementally whenever the pending list is
43 : // empty by scanning a current readState for files missing statistics. Once a
44 : // job completes a scan without finding any remaining files without
45 : // statistics, it flips a `loadedInitial` flag. From then on, the stats
46 : // collection job only needs to load statistics for new files appended to the
47 : // pending list.
48 :
49 1 : func (d *DB) maybeCollectTableStatsLocked() {
50 1 : if d.shouldCollectTableStatsLocked() {
51 1 : go d.collectTableStats()
52 1 : }
53 : }
54 :
55 : // updateTableStatsLocked is called when new files are introduced, after the
56 : // read state has been updated. It may trigger a new stat collection.
57 : // DB.mu must be locked when calling.
58 1 : func (d *DB) updateTableStatsLocked(newTables []manifest.NewTableEntry) {
59 1 : var needStats bool
60 1 : for _, nf := range newTables {
61 1 : if !nf.Meta.StatsValid() {
62 1 : needStats = true
63 1 : break
64 : }
65 : }
66 1 : if !needStats {
67 1 : return
68 1 : }
69 :
70 1 : d.mu.tableStats.pending = append(d.mu.tableStats.pending, newTables...)
71 1 : d.maybeCollectTableStatsLocked()
72 : }
73 :
74 1 : func (d *DB) shouldCollectTableStatsLocked() bool {
75 1 : return !d.mu.tableStats.loading &&
76 1 : d.closed.Load() == nil &&
77 1 : !d.opts.DisableTableStats &&
78 1 : (len(d.mu.tableStats.pending) > 0 || !d.mu.tableStats.loadedInitial)
79 1 : }
80 :
81 : // collectTableStats runs a table stats collection job, returning true if the
82 : // invocation did the collection work, false otherwise (e.g. if another job was
83 : // already running).
84 1 : func (d *DB) collectTableStats() bool {
85 1 : const maxTableStatsPerScan = 50
86 1 :
87 1 : d.mu.Lock()
88 1 : if !d.shouldCollectTableStatsLocked() {
89 1 : d.mu.Unlock()
90 1 : return false
91 1 : }
92 :
93 1 : pending := d.mu.tableStats.pending
94 1 : d.mu.tableStats.pending = nil
95 1 : d.mu.tableStats.loading = true
96 1 : jobID := d.newJobIDLocked()
97 1 : loadedInitial := d.mu.tableStats.loadedInitial
98 1 : // Drop DB.mu before performing IO.
99 1 : d.mu.Unlock()
100 1 :
101 1 : // Every run of collectTableStats either collects stats from the pending
102 1 : // list (if non-empty) or from scanning the version (loadedInitial is
103 1 : // false). This job only runs if at least one of those conditions holds.
104 1 :
105 1 : // Grab a read state to scan for tables.
106 1 : rs := d.loadReadState()
107 1 : var collected []collectedStats
108 1 : var hints []deleteCompactionHint
109 1 : if len(pending) > 0 {
110 1 : collected, hints = d.loadNewFileStats(rs, pending)
111 1 : } else {
112 1 : var moreRemain bool
113 1 : var buf [maxTableStatsPerScan]collectedStats
114 1 : collected, hints, moreRemain = d.scanReadStateTableStats(rs, buf[:0])
115 1 : loadedInitial = !moreRemain
116 1 : }
117 1 : rs.unref()
118 1 :
119 1 : // Update the TableMetadata with the loaded stats while holding d.mu.
120 1 : d.mu.Lock()
121 1 : defer d.mu.Unlock()
122 1 : d.mu.tableStats.loading = false
123 1 : if loadedInitial && !d.mu.tableStats.loadedInitial {
124 1 : d.mu.tableStats.loadedInitial = loadedInitial
125 1 : d.opts.EventListener.TableStatsLoaded(TableStatsInfo{
126 1 : JobID: int(jobID),
127 1 : })
128 1 : }
129 :
130 1 : maybeCompact := false
131 1 : for _, c := range collected {
132 1 : c.tableMetadata.Stats = c.TableStats
133 1 : maybeCompact = maybeCompact || fileCompensation(c.tableMetadata) > 0
134 1 : c.tableMetadata.StatsMarkValid()
135 1 : }
136 :
137 1 : d.mu.tableStats.cond.Broadcast()
138 1 : d.maybeCollectTableStatsLocked()
139 1 : if len(hints) > 0 && !d.opts.private.disableDeleteOnlyCompactions {
140 1 : // Verify that all of the hint tombstones' files still exist in the
141 1 : // current version. Otherwise, the tombstone itself may have been
142 1 : // compacted into L6 and more recent keys may have had their sequence
143 1 : // numbers zeroed.
144 1 : //
145 1 : // Note that it's possible that the tombstone file is being compacted
146 1 : // presently. In that case, the file will be present in v. When the
147 1 : // compaction finishes compacting the tombstone file, it will detect
148 1 : // and clear the hint.
149 1 : //
150 1 : // See DB.maybeUpdateDeleteCompactionHints.
151 1 : v := d.mu.versions.currentVersion()
152 1 : keepHints := hints[:0]
153 1 : for _, h := range hints {
154 1 : if v.Contains(h.tombstoneLevel, h.tombstoneFile) {
155 1 : keepHints = append(keepHints, h)
156 1 : }
157 : }
158 1 : d.mu.compact.deletionHints = append(d.mu.compact.deletionHints, keepHints...)
159 : }
160 1 : if maybeCompact {
161 1 : d.maybeScheduleCompaction()
162 1 : }
163 1 : return true
164 : }
165 :
166 : type collectedStats struct {
167 : *tableMetadata
168 : manifest.TableStats
169 : }
170 :
171 : func (d *DB) loadNewFileStats(
172 : rs *readState, pending []manifest.NewTableEntry,
173 1 : ) ([]collectedStats, []deleteCompactionHint) {
174 1 : var hints []deleteCompactionHint
175 1 : collected := make([]collectedStats, 0, len(pending))
176 1 : for _, nf := range pending {
177 1 : // A file's stats might have been populated by an earlier call to
178 1 : // loadNewFileStats if the file was moved.
179 1 : // NB: We're not holding d.mu which protects f.Stats, but only
180 1 : // collectTableStats updates f.Stats for active files, and we
181 1 : // ensure only one goroutine runs it at a time through
182 1 : // d.mu.tableStats.loading.
183 1 : if nf.Meta.StatsValid() {
184 1 : continue
185 : }
186 :
187 : // The file isn't guaranteed to still be live in the readState's
188 : // version. It may have been deleted or moved. Skip it if it's not in
189 : // the expected level.
190 1 : if !rs.current.Contains(nf.Level, nf.Meta) {
191 1 : continue
192 : }
193 :
194 1 : stats, newHints, err := d.loadTableStats(
195 1 : rs.current, nf.Level,
196 1 : nf.Meta,
197 1 : )
198 1 : if err != nil {
199 0 : d.opts.EventListener.BackgroundError(err)
200 0 : continue
201 : }
202 : // NB: We don't update the TableMetadata yet, because we aren't holding
203 : // DB.mu. We'll copy it to the TableMetadata after we're finished with
204 : // IO.
205 1 : collected = append(collected, collectedStats{
206 1 : tableMetadata: nf.Meta,
207 1 : TableStats: stats,
208 1 : })
209 1 : hints = append(hints, newHints...)
210 : }
211 1 : return collected, hints
212 : }
213 :
214 : // scanReadStateTableStats is run by an active stat collection job when there
215 : // are no pending new files, but there might be files that existed at Open for
216 : // which we haven't loaded table stats.
217 : func (d *DB) scanReadStateTableStats(
218 : rs *readState, fill []collectedStats,
219 1 : ) ([]collectedStats, []deleteCompactionHint, bool) {
220 1 : moreRemain := false
221 1 : var hints []deleteCompactionHint
222 1 : sizesChecked := make(map[base.DiskFileNum]struct{})
223 1 : for l, levelMetadata := range rs.current.Levels {
224 1 : iter := levelMetadata.Iter()
225 1 : for f := iter.First(); f != nil; f = iter.Next() {
226 1 : // NB: We're not holding d.mu which protects f.Stats, but only the
227 1 : // active stats collection job updates f.Stats for active files,
228 1 : // and we ensure only one goroutine runs it at a time through
229 1 : // d.mu.tableStats.loading. This makes it safe to read validity
230 1 : // through f.Stats.ValidLocked despite not holding d.mu.
231 1 : if f.StatsValid() {
232 1 : continue
233 : }
234 :
235 : // Limit how much work we do per read state. The older the read
236 : // state is, the higher the likelihood files are no longer being
237 : // used in the current version. If we've exhausted our allowance,
238 : // return true for the last return value to signal there's more
239 : // work to do.
240 1 : if len(fill) == cap(fill) {
241 1 : moreRemain = true
242 1 : return fill, hints, moreRemain
243 1 : }
244 :
245 : // If the file is remote and not SharedForeign, we should check if its size
246 : // matches. This is because checkConsistency skips over remote files.
247 : //
248 : // SharedForeign and External files are skipped as their sizes are allowed
249 : // to have a mismatch; the size stored in the FileBacking is just the part
250 : // of the file that is referenced by this Pebble instance, not the size of
251 : // the whole object.
252 1 : objMeta, err := d.objProvider.Lookup(base.FileTypeTable, f.FileBacking.DiskFileNum)
253 1 : if err != nil {
254 0 : // Set `moreRemain` so we'll try again.
255 0 : moreRemain = true
256 0 : d.opts.EventListener.BackgroundError(err)
257 0 : continue
258 : }
259 :
260 1 : shouldCheckSize := objMeta.IsRemote() &&
261 1 : !d.objProvider.IsSharedForeign(objMeta) &&
262 1 : !objMeta.IsExternal()
263 1 : if _, ok := sizesChecked[f.FileBacking.DiskFileNum]; !ok && shouldCheckSize {
264 1 : size, err := d.objProvider.Size(objMeta)
265 1 : fileSize := f.FileBacking.Size
266 1 : if err != nil {
267 0 : moreRemain = true
268 0 : d.opts.EventListener.BackgroundError(err)
269 0 : continue
270 : }
271 1 : if size != int64(fileSize) {
272 0 : err := errors.Errorf(
273 0 : "during consistency check in loadTableStats: L%d: %s: object size mismatch (%s): %d (provider) != %d (MANIFEST)",
274 0 : errors.Safe(l), f.FileNum, d.objProvider.Path(objMeta),
275 0 : errors.Safe(size), errors.Safe(fileSize))
276 0 : d.opts.EventListener.BackgroundError(err)
277 0 : d.opts.Logger.Fatalf("%s", err)
278 0 : }
279 :
280 1 : sizesChecked[f.FileBacking.DiskFileNum] = struct{}{}
281 : }
282 :
283 1 : stats, newHints, err := d.loadTableStats(
284 1 : rs.current, l, f,
285 1 : )
286 1 : if err != nil {
287 0 : // Set `moreRemain` so we'll try again.
288 0 : moreRemain = true
289 0 : d.opts.EventListener.BackgroundError(err)
290 0 : continue
291 : }
292 1 : fill = append(fill, collectedStats{
293 1 : tableMetadata: f,
294 1 : TableStats: stats,
295 1 : })
296 1 : hints = append(hints, newHints...)
297 : }
298 : }
299 1 : return fill, hints, moreRemain
300 : }
301 :
302 : func (d *DB) loadTableStats(
303 : v *version, level int, meta *tableMetadata,
304 1 : ) (manifest.TableStats, []deleteCompactionHint, error) {
305 1 : var stats manifest.TableStats
306 1 : var compactionHints []deleteCompactionHint
307 1 : err := d.fileCache.withCommonReader(
308 1 : meta, func(r sstable.CommonReader) (err error) {
309 1 : props := r.CommonProperties()
310 1 : stats.NumEntries = props.NumEntries
311 1 : stats.NumDeletions = props.NumDeletions
312 1 : stats.NumRangeKeySets = props.NumRangeKeySets
313 1 : stats.ValueBlocksSize = props.ValueBlocksSize
314 1 : stats.CompressionType = block.CompressionFromString(props.CompressionName)
315 1 : if props.NumDataBlocks > 0 {
316 1 : stats.TombstoneDenseBlocksRatio = float64(props.NumTombstoneDenseBlocks) / float64(props.NumDataBlocks)
317 1 : }
318 :
319 1 : if props.NumPointDeletions() > 0 {
320 1 : if err = d.loadTablePointKeyStats(props, v, level, meta, &stats); err != nil {
321 0 : return
322 0 : }
323 : }
324 1 : if props.NumRangeDeletions > 0 || props.NumRangeKeyDels > 0 {
325 1 : if compactionHints, err = d.loadTableRangeDelStats(
326 1 : r, v, level, meta, &stats,
327 1 : ); err != nil {
328 0 : return
329 0 : }
330 : }
331 1 : return
332 : })
333 1 : if err != nil {
334 0 : return stats, nil, err
335 0 : }
336 1 : return stats, compactionHints, nil
337 : }
338 :
339 : // loadTablePointKeyStats calculates the point key statistics for the given
340 : // table. The provided manifest.TableStats are updated.
341 : func (d *DB) loadTablePointKeyStats(
342 : props *sstable.CommonProperties,
343 : v *version,
344 : level int,
345 : meta *tableMetadata,
346 : stats *manifest.TableStats,
347 1 : ) error {
348 1 : // TODO(jackson): If the file has a wide keyspace, the average
349 1 : // value size beneath the entire file might not be representative
350 1 : // of the size of the keys beneath the point tombstones.
351 1 : // We could write the ranges of 'clusters' of point tombstones to
352 1 : // a sstable property and call averageValueSizeBeneath for each of
353 1 : // these narrower ranges to improve the estimate.
354 1 : avgValLogicalSize, compressionRatio, err := d.estimateSizesBeneath(v, level, meta, props)
355 1 : if err != nil {
356 0 : return err
357 0 : }
358 1 : stats.PointDeletionsBytesEstimate =
359 1 : pointDeletionsBytesEstimate(meta.Size, props, avgValLogicalSize, compressionRatio)
360 1 : return nil
361 : }
362 :
363 : // loadTableRangeDelStats calculates the range deletion and range key deletion
364 : // statistics for the given table.
365 : func (d *DB) loadTableRangeDelStats(
366 : r sstable.CommonReader, v *version, level int, meta *tableMetadata, stats *manifest.TableStats,
367 1 : ) ([]deleteCompactionHint, error) {
368 1 : iter, err := newCombinedDeletionKeyspanIter(d.opts.Comparer, r, meta)
369 1 : if err != nil {
370 0 : return nil, err
371 0 : }
372 1 : defer iter.Close()
373 1 : var compactionHints []deleteCompactionHint
374 1 : // We iterate over the defragmented range tombstones and range key deletions,
375 1 : // which ensures we don't double count ranges deleted at different sequence
376 1 : // numbers. Also, merging abutting tombstones reduces the number of calls to
377 1 : // estimateReclaimedSizeBeneath which is costly, and improves the accuracy of
378 1 : // our overall estimate.
379 1 : s, err := iter.First()
380 1 : for ; s != nil; s, err = iter.Next() {
381 1 : start, end := s.Start, s.End
382 1 : // We only need to consider deletion size estimates for tables that contain
383 1 : // RANGEDELs.
384 1 : var maxRangeDeleteSeqNum base.SeqNum
385 1 : for _, k := range s.Keys {
386 1 : if k.Kind() == base.InternalKeyKindRangeDelete && maxRangeDeleteSeqNum < k.SeqNum() {
387 1 : maxRangeDeleteSeqNum = k.SeqNum()
388 1 : break
389 : }
390 : }
391 :
392 : // If the file is in the last level of the LSM, there is no data beneath
393 : // it. The fact that there is still a range tombstone in a bottommost file
394 : // indicates two possibilites:
395 : // 1. an open snapshot kept the tombstone around, and the data the
396 : // tombstone deletes is contained within the file itself.
397 : // 2. the file was ingested.
398 : // In the first case, we'd like to estimate disk usage within the file
399 : // itself since compacting the file will drop that covered data. In the
400 : // second case, we expect that compacting the file will NOT drop any
401 : // data and rewriting the file is a waste of write bandwidth. We can
402 : // distinguish these cases by looking at the table metadata's sequence
403 : // numbers. A file's range deletions can only delete data within the
404 : // file at lower sequence numbers. All keys in an ingested sstable adopt
405 : // the same sequence number, preventing tombstones from deleting keys
406 : // within the same file. We check here if the largest RANGEDEL sequence
407 : // number is greater than the file's smallest sequence number. If it is,
408 : // the RANGEDEL could conceivably (although inconclusively) delete data
409 : // within the same file.
410 : //
411 : // Note that this heuristic is imperfect. If a table containing a range
412 : // deletion is ingested into L5 and subsequently compacted into L6 but
413 : // an open snapshot prevents elision of covered keys in L6, the
414 : // resulting RangeDeletionsBytesEstimate will incorrectly include all
415 : // covered keys.
416 : //
417 : // TODO(jackson): We could prevent the above error in the heuristic by
418 : // computing the file's RangeDeletionsBytesEstimate during the
419 : // compaction itself. It's unclear how common this is.
420 : //
421 : // NOTE: If the span `s` wholly contains a table containing range keys,
422 : // the returned size estimate will be slightly inflated by the range key
423 : // block. However, in practice, range keys are expected to be rare, and
424 : // the size of the range key block relative to the overall size of the
425 : // table is expected to be small.
426 1 : if level == numLevels-1 && meta.SmallestSeqNum < maxRangeDeleteSeqNum {
427 1 : size, err := r.EstimateDiskUsage(start, end)
428 1 : if err != nil {
429 0 : return nil, err
430 0 : }
431 1 : stats.RangeDeletionsBytesEstimate += size
432 1 :
433 1 : // As the file is in the bottommost level, there is no need to collect a
434 1 : // deletion hint.
435 1 : continue
436 : }
437 :
438 : // While the size estimates for point keys should only be updated if this
439 : // span contains a range del, the sequence numbers are required for the
440 : // hint. Unconditionally descend, but conditionally update the estimates.
441 1 : hintType := compactionHintFromKeys(s.Keys)
442 1 : estimate, hintSeqNum, err := d.estimateReclaimedSizeBeneath(v, level, start, end, hintType)
443 1 : if err != nil {
444 0 : return nil, err
445 0 : }
446 1 : stats.RangeDeletionsBytesEstimate += estimate
447 1 :
448 1 : // hintSeqNum is the smallest sequence number contained in any
449 1 : // file overlapping with the hint and in a level below it.
450 1 : if hintSeqNum == math.MaxUint64 {
451 1 : continue
452 : }
453 1 : hint := deleteCompactionHint{
454 1 : hintType: hintType,
455 1 : start: make([]byte, len(start)),
456 1 : end: make([]byte, len(end)),
457 1 : tombstoneFile: meta,
458 1 : tombstoneLevel: level,
459 1 : tombstoneLargestSeqNum: s.LargestSeqNum(),
460 1 : tombstoneSmallestSeqNum: s.SmallestSeqNum(),
461 1 : fileSmallestSeqNum: hintSeqNum,
462 1 : }
463 1 : copy(hint.start, start)
464 1 : copy(hint.end, end)
465 1 : compactionHints = append(compactionHints, hint)
466 : }
467 1 : if err != nil {
468 0 : return nil, err
469 0 : }
470 1 : return compactionHints, nil
471 : }
472 :
473 : func (d *DB) estimateSizesBeneath(
474 : v *version, level int, meta *tableMetadata, fileProps *sstable.CommonProperties,
475 1 : ) (avgValueLogicalSize, compressionRatio float64, err error) {
476 1 : // Find all files in lower levels that overlap with meta,
477 1 : // summing their value sizes and entry counts.
478 1 : file := meta
479 1 : var fileSum, keySum, valSum, entryCount uint64
480 1 : // Include the file itself. This is important because in some instances, the
481 1 : // computed compression ratio is applied to the tombstones contained within
482 1 : // `meta` itself. If there are no files beneath `meta` in the LSM, we would
483 1 : // calculate a compression ratio of 0 which is not accurate for the file's
484 1 : // own tombstones.
485 1 : fileSum += file.Size
486 1 : entryCount += fileProps.NumEntries
487 1 : keySum += fileProps.RawKeySize
488 1 : valSum += fileProps.RawValueSize
489 1 :
490 1 : addPhysicalTableStats := func(r *sstable.Reader) (err error) {
491 1 : fileSum += file.Size
492 1 : entryCount += r.Properties.NumEntries
493 1 : keySum += r.Properties.RawKeySize
494 1 : valSum += r.Properties.RawValueSize
495 1 : return nil
496 1 : }
497 1 : addVirtualTableStats := func(v sstable.VirtualReader) (err error) {
498 1 : fileSum += file.Size
499 1 : entryCount += file.Stats.NumEntries
500 1 : keySum += v.Properties.RawKeySize
501 1 : valSum += v.Properties.RawValueSize
502 1 : return nil
503 1 : }
504 :
505 1 : for l := level + 1; l < numLevels; l++ {
506 1 : overlaps := v.Overlaps(l, meta.UserKeyBounds())
507 1 : iter := overlaps.Iter()
508 1 : for file = iter.First(); file != nil; file = iter.Next() {
509 1 : var err error
510 1 : if file.Virtual {
511 1 : err = d.fileCache.withVirtualReader(file.VirtualMeta(), addVirtualTableStats)
512 1 : } else {
513 1 : err = d.fileCache.withReader(file.PhysicalMeta(), addPhysicalTableStats)
514 1 : }
515 1 : if err != nil {
516 0 : return 0, 0, err
517 0 : }
518 : }
519 : }
520 1 : if entryCount == 0 {
521 0 : return 0, 0, nil
522 0 : }
523 : // RawKeySize and RawValueSize are uncompressed totals. We'll need to scale
524 : // the value sum according to the data size to account for compression,
525 : // index blocks and metadata overhead. Eg:
526 : //
527 : // Compression rate × Average uncompressed value size
528 : //
529 : // ↓
530 : //
531 : // FileSize RawValueSize
532 : // ----------------------- × ------------
533 : // RawKeySize+RawValueSize NumEntries
534 : //
535 : // We return the average logical value size plus the compression ratio,
536 : // leaving the scaling to the caller. This allows the caller to perform
537 : // additional compression ratio scaling if necessary.
538 1 : uncompressedSum := float64(keySum + valSum)
539 1 : compressionRatio = float64(fileSum) / uncompressedSum
540 1 : avgValueLogicalSize = (float64(valSum) / float64(entryCount))
541 1 : return avgValueLogicalSize, compressionRatio, nil
542 : }
543 :
544 : func (d *DB) estimateReclaimedSizeBeneath(
545 : v *version, level int, start, end []byte, hintType deleteCompactionHintType,
546 1 : ) (estimate uint64, hintSeqNum base.SeqNum, err error) {
547 1 : // Find all files in lower levels that overlap with the deleted range
548 1 : // [start, end).
549 1 : //
550 1 : // An overlapping file might be completely contained by the range
551 1 : // tombstone, in which case we can count the entire file size in
552 1 : // our estimate without doing any additional I/O.
553 1 : //
554 1 : // Otherwise, estimating the range for the file requires
555 1 : // additional I/O to read the file's index blocks.
556 1 : hintSeqNum = math.MaxUint64
557 1 : for l := level + 1; l < numLevels; l++ {
558 1 : overlaps := v.Overlaps(l, base.UserKeyBoundsEndExclusive(start, end))
559 1 : iter := overlaps.Iter()
560 1 : for file := iter.First(); file != nil; file = iter.Next() {
561 1 : // Determine whether we need to update size estimates and hint seqnums
562 1 : // based on the type of hint and the type of keys in this file.
563 1 : var updateEstimates, updateHints bool
564 1 : switch hintType {
565 1 : case deleteCompactionHintTypePointKeyOnly:
566 1 : // The range deletion byte estimates should only be updated if this
567 1 : // table contains point keys. This ends up being an overestimate in
568 1 : // the case that table also has range keys, but such keys are expected
569 1 : // to contribute a negligible amount of the table's overall size,
570 1 : // relative to point keys.
571 1 : if file.HasPointKeys {
572 1 : updateEstimates = true
573 1 : }
574 : // As the initiating span contained only range dels, hints can only be
575 : // updated if this table does _not_ contain range keys.
576 1 : if !file.HasRangeKeys {
577 1 : updateHints = true
578 1 : }
579 1 : case deleteCompactionHintTypeRangeKeyOnly:
580 1 : // The initiating span contained only range key dels. The estimates
581 1 : // apply only to point keys, and are therefore not updated.
582 1 : updateEstimates = false
583 1 : // As the initiating span contained only range key dels, hints can
584 1 : // only be updated if this table does _not_ contain point keys.
585 1 : if !file.HasPointKeys {
586 1 : updateHints = true
587 1 : }
588 1 : case deleteCompactionHintTypePointAndRangeKey:
589 1 : // Always update the estimates and hints, as this hint type can drop a
590 1 : // file, irrespective of the mixture of keys. Similar to above, the
591 1 : // range del bytes estimates is an overestimate.
592 1 : updateEstimates, updateHints = true, true
593 0 : default:
594 0 : panic(fmt.Sprintf("pebble: unknown hint type %s", hintType))
595 : }
596 1 : startCmp := d.cmp(start, file.Smallest.UserKey)
597 1 : endCmp := d.cmp(file.Largest.UserKey, end)
598 1 : if startCmp <= 0 && (endCmp < 0 || endCmp == 0 && file.Largest.IsExclusiveSentinel()) {
599 1 : // The range fully contains the file, so skip looking it up in table
600 1 : // cache/looking at its indexes and add the full file size.
601 1 : if updateEstimates {
602 1 : estimate += file.Size
603 1 : }
604 1 : if updateHints && hintSeqNum > file.SmallestSeqNum {
605 1 : hintSeqNum = file.SmallestSeqNum
606 1 : }
607 1 : } else if d.cmp(file.Smallest.UserKey, end) <= 0 && d.cmp(start, file.Largest.UserKey) <= 0 {
608 1 : // Partial overlap.
609 1 : if hintType == deleteCompactionHintTypeRangeKeyOnly {
610 1 : // If the hint that generated this overlap contains only range keys,
611 1 : // there is no need to calculate disk usage, as the reclaimable space
612 1 : // is expected to be minimal relative to point keys.
613 1 : continue
614 : }
615 1 : var size uint64
616 1 : var err error
617 1 : if file.Virtual {
618 1 : err = d.fileCache.withVirtualReader(
619 1 : file.VirtualMeta(), func(r sstable.VirtualReader) (err error) {
620 1 : size, err = r.EstimateDiskUsage(start, end)
621 1 : return err
622 1 : })
623 1 : } else {
624 1 : err = d.fileCache.withReader(
625 1 : file.PhysicalMeta(), func(r *sstable.Reader) (err error) {
626 1 : size, err = r.EstimateDiskUsage(start, end)
627 1 : return err
628 1 : })
629 : }
630 :
631 1 : if err != nil {
632 0 : return 0, hintSeqNum, err
633 0 : }
634 1 : estimate += size
635 1 : if updateHints && hintSeqNum > file.SmallestSeqNum && d.FormatMajorVersion() >= FormatVirtualSSTables {
636 1 : // If the format major version is past Virtual SSTables, deletion only
637 1 : // hints can also apply to partial overlaps with sstables.
638 1 : hintSeqNum = file.SmallestSeqNum
639 1 : }
640 : }
641 : }
642 : }
643 1 : return estimate, hintSeqNum, nil
644 : }
645 :
646 1 : func maybeSetStatsFromProperties(meta physicalMeta, props *sstable.Properties) bool {
647 1 : // If a table contains range deletions or range key deletions, we defer the
648 1 : // stats collection. There are two main reasons for this:
649 1 : //
650 1 : // 1. Estimating the potential for reclaimed space due to a range deletion
651 1 : // tombstone requires scanning the LSM - a potentially expensive operation
652 1 : // that should be deferred.
653 1 : // 2. Range deletions and / or range key deletions present an opportunity to
654 1 : // compute "deletion hints", which also requires a scan of the LSM to
655 1 : // compute tables that would be eligible for deletion.
656 1 : //
657 1 : // These two tasks are deferred to the table stats collector goroutine.
658 1 : if props.NumRangeDeletions != 0 || props.NumRangeKeyDels != 0 {
659 1 : return false
660 1 : }
661 :
662 : // If a table is more than 10% point deletions without user-provided size
663 : // estimates, don't calculate the PointDeletionsBytesEstimate statistic
664 : // using our limited knowledge. The table stats collector can populate the
665 : // stats and calculate an average of value size of all the tables beneath
666 : // the table in the LSM, which will be more accurate.
667 1 : if unsizedDels := (props.NumDeletions - props.NumSizedDeletions); unsizedDels > props.NumEntries/10 {
668 1 : return false
669 1 : }
670 :
671 1 : var pointEstimate uint64
672 1 : if props.NumEntries > 0 {
673 1 : // Use the file's own average key and value sizes as an estimate. This
674 1 : // doesn't require any additional IO and since the number of point
675 1 : // deletions in the file is low, the error introduced by this crude
676 1 : // estimate is expected to be small.
677 1 : commonProps := &props.CommonProperties
678 1 : avgValSize, compressionRatio := estimatePhysicalSizes(meta.Size, commonProps)
679 1 : pointEstimate = pointDeletionsBytesEstimate(meta.Size, commonProps, avgValSize, compressionRatio)
680 1 : }
681 :
682 1 : meta.Stats.NumEntries = props.NumEntries
683 1 : meta.Stats.NumDeletions = props.NumDeletions
684 1 : meta.Stats.NumRangeKeySets = props.NumRangeKeySets
685 1 : meta.Stats.PointDeletionsBytesEstimate = pointEstimate
686 1 : meta.Stats.RangeDeletionsBytesEstimate = 0
687 1 : meta.Stats.ValueBlocksSize = props.ValueBlocksSize
688 1 : meta.Stats.CompressionType = block.CompressionFromString(props.CompressionName)
689 1 : meta.StatsMarkValid()
690 1 : return true
691 : }
692 :
693 : func pointDeletionsBytesEstimate(
694 : fileSize uint64, props *sstable.CommonProperties, avgValLogicalSize, compressionRatio float64,
695 1 : ) (estimate uint64) {
696 1 : if props.NumEntries == 0 {
697 0 : return 0
698 0 : }
699 1 : numPointDels := props.NumPointDeletions()
700 1 : if numPointDels == 0 {
701 1 : return 0
702 1 : }
703 : // Estimate the potential space to reclaim using the table's own properties.
704 : // There may or may not be keys covered by any individual point tombstone.
705 : // If not, compacting the point tombstone into L6 will at least allow us to
706 : // drop the point deletion key and will reclaim the tombstone's key bytes.
707 : // If there are covered key(s), we also get to drop key and value bytes for
708 : // each covered key.
709 : //
710 : // Some point tombstones (DELSIZEDs) carry a user-provided estimate of the
711 : // uncompressed size of entries that will be elided by fully compacting the
712 : // tombstone. For these tombstones, there's no guesswork—we use the
713 : // RawPointTombstoneValueSizeHint property which is the sum of all these
714 : // tombstones' encoded values.
715 : //
716 : // For un-sized point tombstones (DELs), we estimate assuming that each
717 : // point tombstone on average covers 1 key and using average value sizes.
718 : // This is almost certainly an overestimate, but that's probably okay
719 : // because point tombstones can slow range iterations even when they don't
720 : // cover a key.
721 : //
722 : // TODO(jackson): This logic doesn't directly incorporate fixed per-key
723 : // overhead (8-byte trailer, plus at least 1 byte encoding the length of the
724 : // key and 1 byte encoding the length of the value). This overhead is
725 : // indirectly incorporated through the compression ratios, but that results
726 : // in the overhead being smeared per key-byte and value-byte, rather than
727 : // per-entry. This per-key fixed overhead can be nontrivial, especially for
728 : // dense swaths of point tombstones. Give some thought as to whether we
729 : // should directly include fixed per-key overhead in the calculations.
730 :
731 : // Below, we calculate the tombstone contributions and the shadowed keys'
732 : // contributions separately.
733 1 : var tombstonesLogicalSize float64
734 1 : var shadowedLogicalSize float64
735 1 :
736 1 : // 1. Calculate the contribution of the tombstone keys themselves.
737 1 : if props.RawPointTombstoneKeySize > 0 {
738 1 : tombstonesLogicalSize += float64(props.RawPointTombstoneKeySize)
739 1 : } else {
740 0 : // This sstable predates the existence of the RawPointTombstoneKeySize
741 0 : // property. We can use the average key size within the file itself and
742 0 : // the count of point deletions to estimate the size.
743 0 : tombstonesLogicalSize += float64(numPointDels * props.RawKeySize / props.NumEntries)
744 0 : }
745 :
746 : // 2. Calculate the contribution of the keys shadowed by tombstones.
747 : //
748 : // 2a. First account for keys shadowed by DELSIZED tombstones. THE DELSIZED
749 : // tombstones encode the size of both the key and value of the shadowed KV
750 : // entries. These sizes are aggregated into a sstable property.
751 1 : shadowedLogicalSize += float64(props.RawPointTombstoneValueSize)
752 1 :
753 1 : // 2b. Calculate the contribution of the KV entries shadowed by ordinary DEL
754 1 : // keys.
755 1 : numUnsizedDels := numPointDels - props.NumSizedDeletions
756 1 : {
757 1 : // The shadowed keys have the same exact user keys as the tombstones
758 1 : // themselves, so we can use the `tombstonesLogicalSize` we computed
759 1 : // earlier as an estimate. There's a complication that
760 1 : // `tombstonesLogicalSize` may include DELSIZED keys we already
761 1 : // accounted for.
762 1 : shadowedLogicalSize += float64(tombstonesLogicalSize) / float64(numPointDels) * float64(numUnsizedDels)
763 1 :
764 1 : // Calculate the contribution of the deleted values. The caller has
765 1 : // already computed an average logical size (possibly computed across
766 1 : // many sstables).
767 1 : shadowedLogicalSize += float64(numUnsizedDels) * avgValLogicalSize
768 1 : }
769 :
770 : // Scale both tombstone and shadowed totals by logical:physical ratios to
771 : // account for compression, metadata overhead, etc.
772 : //
773 : // Physical FileSize
774 : // ----------- = -----------------------
775 : // Logical RawKeySize+RawValueSize
776 : //
777 1 : return uint64((tombstonesLogicalSize + shadowedLogicalSize) * compressionRatio)
778 : }
779 :
780 : func estimatePhysicalSizes(
781 : fileSize uint64, props *sstable.CommonProperties,
782 1 : ) (avgValLogicalSize, compressionRatio float64) {
783 1 : // RawKeySize and RawValueSize are uncompressed totals. Scale according to
784 1 : // the data size to account for compression, index blocks and metadata
785 1 : // overhead. Eg:
786 1 : //
787 1 : // Compression rate × Average uncompressed value size
788 1 : //
789 1 : // ↓
790 1 : //
791 1 : // FileSize RawValSize
792 1 : // ----------------------- × ----------
793 1 : // RawKeySize+RawValueSize NumEntries
794 1 : //
795 1 : uncompressedSum := props.RawKeySize + props.RawValueSize
796 1 : compressionRatio = float64(fileSize) / float64(uncompressedSum)
797 1 : avgValLogicalSize = (float64(props.RawValueSize) / float64(props.NumEntries))
798 1 : return avgValLogicalSize, compressionRatio
799 1 : }
800 :
801 : // newCombinedDeletionKeyspanIter returns a keyspan.FragmentIterator that
802 : // returns "ranged deletion" spans for a single table, providing a combined view
803 : // of both range deletion and range key deletion spans. The
804 : // tableRangedDeletionIter is intended for use in the specific case of computing
805 : // the statistics and deleteCompactionHints for a single table.
806 : //
807 : // As an example, consider the following set of spans from the range deletion
808 : // and range key blocks of a table:
809 : //
810 : // |---------| |---------| |-------| RANGEKEYDELs
811 : // |-----------|-------------| |-----| RANGEDELs
812 : // __________________________________________________________
813 : // 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
814 : //
815 : // The tableRangedDeletionIter produces the following set of output spans, where
816 : // '1' indicates a span containing only range deletions, '2' is a span
817 : // containing only range key deletions, and '3' is a span containing a mixture
818 : // of both range deletions and range key deletions.
819 : //
820 : // 1 3 1 3 2 1 3 2
821 : // |-----|---------|-----|---|-----| |---|-|-----|
822 : // __________________________________________________________
823 : // 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
824 : //
825 : // Algorithm.
826 : //
827 : // The iterator first defragments the range deletion and range key blocks
828 : // separately. During this defragmentation, the range key block is also filtered
829 : // so that keys other than range key deletes are ignored. The range delete and
830 : // range key delete keyspaces are then merged.
831 : //
832 : // Note that the only fragmentation introduced by merging is from where a range
833 : // del span overlaps with a range key del span. Within the bounds of any overlap
834 : // there is guaranteed to be no further fragmentation, as the constituent spans
835 : // have already been defragmented. To the left and right of any overlap, the
836 : // same reasoning applies. For example,
837 : //
838 : // |--------| |-------| RANGEKEYDEL
839 : // |---------------------------| RANGEDEL
840 : // |----1---|----3---|----1----|---2---| Merged, fragmented spans.
841 : // __________________________________________________________
842 : // 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
843 : //
844 : // Any fragmented abutting spans produced by the merging iter will be of
845 : // differing types (i.e. a transition from a span with homogenous key kinds to a
846 : // heterogeneous span, or a transition from a span with exclusively range dels
847 : // to a span with exclusively range key dels). Therefore, further
848 : // defragmentation is not required.
849 : //
850 : // Each span returned by the tableRangeDeletionIter will have at most four keys,
851 : // corresponding to the largest and smallest sequence numbers encountered across
852 : // the range deletes and range keys deletes that comprised the merged spans.
853 : func newCombinedDeletionKeyspanIter(
854 : comparer *base.Comparer, cr sstable.CommonReader, m *tableMetadata,
855 1 : ) (keyspan.FragmentIterator, error) {
856 1 : // The range del iter and range key iter are each wrapped in their own
857 1 : // defragmenting iter. For each iter, abutting spans can always be merged.
858 1 : var equal = keyspan.DefragmentMethodFunc(func(_ base.CompareRangeSuffixes, a, b *keyspan.Span) bool { return true })
859 : // Reduce keys by maintaining a slice of at most length two, corresponding to
860 : // the largest and smallest keys in the defragmented span. This maintains the
861 : // contract that the emitted slice is sorted by (SeqNum, Kind) descending.
862 1 : reducer := func(current, incoming []keyspan.Key) []keyspan.Key {
863 1 : if len(current) == 0 && len(incoming) == 0 {
864 0 : // While this should never occur in practice, a defensive return is used
865 0 : // here to preserve correctness.
866 0 : return current
867 0 : }
868 1 : var largest, smallest keyspan.Key
869 1 : var set bool
870 1 : for _, keys := range [2][]keyspan.Key{current, incoming} {
871 1 : if len(keys) == 0 {
872 0 : continue
873 : }
874 1 : first, last := keys[0], keys[len(keys)-1]
875 1 : if !set {
876 1 : largest, smallest = first, last
877 1 : set = true
878 1 : continue
879 : }
880 1 : if first.Trailer > largest.Trailer {
881 1 : largest = first
882 1 : }
883 1 : if last.Trailer < smallest.Trailer {
884 1 : smallest = last
885 1 : }
886 : }
887 1 : if largest.Equal(comparer.CompareRangeSuffixes, smallest) {
888 1 : current = append(current[:0], largest)
889 1 : } else {
890 1 : current = append(current[:0], largest, smallest)
891 1 : }
892 1 : return current
893 : }
894 :
895 : // The separate iters for the range dels and range keys are wrapped in a
896 : // merging iter to join the keyspaces into a single keyspace. The separate
897 : // iters are only added if the particular key kind is present.
898 1 : mIter := &keyspanimpl.MergingIter{}
899 1 : var transform = keyspan.TransformerFunc(func(_ base.CompareRangeSuffixes, in keyspan.Span, out *keyspan.Span) error {
900 1 : if in.KeysOrder != keyspan.ByTrailerDesc {
901 0 : panic("pebble: combined deletion iter encountered keys in non-trailer descending order")
902 : }
903 1 : out.Start, out.End = in.Start, in.End
904 1 : out.Keys = append(out.Keys[:0], in.Keys...)
905 1 : out.KeysOrder = keyspan.ByTrailerDesc
906 1 : // NB: The order of by-trailer descending may have been violated,
907 1 : // because we've layered rangekey and rangedel iterators from the same
908 1 : // sstable into the same keyspanimpl.MergingIter. The MergingIter will
909 1 : // return the keys in the order that the child iterators were provided.
910 1 : // Sort the keys to ensure they're sorted by trailer descending.
911 1 : keyspan.SortKeysByTrailer(out.Keys)
912 1 : return nil
913 : })
914 1 : mIter.Init(comparer, transform, new(keyspanimpl.MergingBuffers))
915 1 :
916 1 : iter, err := cr.NewRawRangeDelIter(context.TODO(), m.FragmentIterTransforms(), block.NoReadEnv)
917 1 : if err != nil {
918 0 : return nil, err
919 0 : }
920 1 : if iter != nil {
921 1 : // Assert expected bounds. In previous versions of Pebble, range
922 1 : // deletions persisted to sstables could exceed the bounds of the
923 1 : // containing files due to "split user keys." This required readers to
924 1 : // constrain the tombstones' bounds to the containing file at read time.
925 1 : // See docs/range_deletions.md for an extended discussion of the design
926 1 : // and invariants at that time.
927 1 : //
928 1 : // We've since compacted away all 'split user-keys' and in the process
929 1 : // eliminated all "untruncated range tombstones" for physical sstables.
930 1 : // We no longer need to perform truncation at read time for these
931 1 : // sstables.
932 1 : //
933 1 : // At the same time, we've also introduced the concept of "virtual
934 1 : // SSTables" where the table metadata's effective bounds can again be
935 1 : // reduced to be narrower than the contained tombstones. These virtual
936 1 : // SSTables handle truncation differently, performing it using
937 1 : // keyspan.Truncate when the sstable's range deletion iterator is
938 1 : // opened.
939 1 : //
940 1 : // Together, these mean that we should never see untruncated range
941 1 : // tombstones any more—and the merging iterator no longer accounts for
942 1 : // their existence. Since there's abundant subtlety that we're relying
943 1 : // on, we choose to be conservative and assert that these invariants
944 1 : // hold. We could (and previously did) choose to only validate these
945 1 : // bounds in invariants builds, but the most likely avenue for these
946 1 : // tombstones' existence is through a bug in a migration and old data
947 1 : // sitting around in an old store from long ago.
948 1 : //
949 1 : // The table stats collector will read all files range deletions
950 1 : // asynchronously after Open, and provides a perfect opportunity to
951 1 : // validate our invariants without harming user latency. We also
952 1 : // previously performed truncation here which similarly required key
953 1 : // comparisons, so replacing those key comparisons with assertions
954 1 : // should be roughly similar in performance.
955 1 : //
956 1 : // TODO(jackson): Only use AssertBounds in invariants builds in the
957 1 : // following release.
958 1 : iter = keyspan.AssertBounds(
959 1 : iter, m.SmallestPointKey, m.LargestPointKey.UserKey, comparer.Compare,
960 1 : )
961 1 : dIter := &keyspan.DefragmentingIter{}
962 1 : dIter.Init(comparer, iter, equal, reducer, new(keyspan.DefragmentingBuffers))
963 1 : iter = dIter
964 1 : mIter.AddLevel(iter)
965 1 : }
966 :
967 1 : iter, err = cr.NewRawRangeKeyIter(context.TODO(), m.FragmentIterTransforms(), block.NoReadEnv)
968 1 : if err != nil {
969 0 : return nil, err
970 0 : }
971 1 : if iter != nil {
972 1 : // Assert expected bounds in tests.
973 1 : if invariants.Sometimes(50) {
974 1 : iter = keyspan.AssertBounds(
975 1 : iter, m.SmallestRangeKey, m.LargestRangeKey.UserKey, comparer.Compare,
976 1 : )
977 1 : }
978 : // Wrap the range key iterator in a filter that elides keys other than range
979 : // key deletions.
980 1 : iter = keyspan.Filter(iter, func(in *keyspan.Span, buf []keyspan.Key) []keyspan.Key {
981 1 : keys := buf[:0]
982 1 : for _, k := range in.Keys {
983 1 : if k.Kind() != base.InternalKeyKindRangeKeyDelete {
984 1 : continue
985 : }
986 1 : keys = append(keys, k)
987 : }
988 1 : return keys
989 : }, comparer.Compare)
990 1 : dIter := &keyspan.DefragmentingIter{}
991 1 : dIter.Init(comparer, iter, equal, reducer, new(keyspan.DefragmentingBuffers))
992 1 : iter = dIter
993 1 : mIter.AddLevel(iter)
994 : }
995 :
996 1 : return mIter, nil
997 : }
998 :
999 : // rangeKeySetsAnnotator is a manifest.Annotator that annotates B-Tree nodes
1000 : // with the sum of the files' counts of range key fragments. The count of range
1001 : // key sets may change once a table's stats are loaded asynchronously, so its
1002 : // values are marked as cacheable only if a file's stats have been loaded.
1003 1 : var rangeKeySetsAnnotator = manifest.SumAnnotator(func(f *manifest.TableMetadata) (uint64, bool) {
1004 1 : return f.Stats.NumRangeKeySets, f.StatsValid()
1005 1 : })
1006 :
1007 : // tombstonesAnnotator is a manifest.Annotator that annotates B-Tree nodes
1008 : // with the sum of the files' counts of tombstones (DEL, SINGLEDEL and RANGEDEL
1009 : // keys). The count of tombstones may change once a table's stats are loaded
1010 : // asynchronously, so its values are marked as cacheable only if a file's stats
1011 : // have been loaded.
1012 1 : var tombstonesAnnotator = manifest.SumAnnotator(func(f *manifest.TableMetadata) (uint64, bool) {
1013 1 : return f.Stats.NumDeletions, f.StatsValid()
1014 1 : })
1015 :
1016 : // valueBlocksSizeAnnotator is a manifest.Annotator that annotates B-Tree
1017 : // nodes with the sum of the files' Properties.ValueBlocksSize. The value block
1018 : // size may change once a table's stats are loaded asynchronously, so its
1019 : // values are marked as cacheable only if a file's stats have been loaded.
1020 1 : var valueBlockSizeAnnotator = manifest.SumAnnotator(func(f *tableMetadata) (uint64, bool) {
1021 1 : return f.Stats.ValueBlocksSize, f.StatsValid()
1022 1 : })
1023 :
1024 : // compressionTypeAnnotator is a manifest.Annotator that annotates B-tree
1025 : // nodes with the compression type of the file. Its annotation type is
1026 : // compressionTypes. The compression type may change once a table's stats are
1027 : // loaded asynchronously, so its values are marked as cacheable only if a file's
1028 : // stats have been loaded.
1029 : var compressionTypeAnnotator = manifest.Annotator[compressionTypes]{
1030 : Aggregator: compressionTypeAggregator{},
1031 : }
1032 :
1033 : type compressionTypeAggregator struct{}
1034 :
1035 : type compressionTypes struct {
1036 : snappy, zstd, none, unknown uint64
1037 : }
1038 :
1039 1 : func (a compressionTypeAggregator) Zero(dst *compressionTypes) *compressionTypes {
1040 1 : if dst == nil {
1041 1 : return new(compressionTypes)
1042 1 : }
1043 0 : *dst = compressionTypes{}
1044 0 : return dst
1045 : }
1046 :
1047 : func (a compressionTypeAggregator) Accumulate(
1048 : f *tableMetadata, dst *compressionTypes,
1049 1 : ) (v *compressionTypes, cacheOK bool) {
1050 1 : switch f.Stats.CompressionType {
1051 1 : case SnappyCompression:
1052 1 : dst.snappy++
1053 1 : case ZstdCompression:
1054 1 : dst.zstd++
1055 1 : case NoCompression:
1056 1 : dst.none++
1057 1 : default:
1058 1 : dst.unknown++
1059 : }
1060 1 : return dst, f.StatsValid()
1061 : }
1062 :
1063 : func (a compressionTypeAggregator) Merge(
1064 : src *compressionTypes, dst *compressionTypes,
1065 1 : ) *compressionTypes {
1066 1 : dst.snappy += src.snappy
1067 1 : dst.zstd += src.zstd
1068 1 : dst.none += src.none
1069 1 : dst.unknown += src.unknown
1070 1 : return dst
1071 1 : }
|
