Line data Source code
1 : // Copyright 2018 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "context"
9 : "fmt"
10 : "slices"
11 : "sort"
12 : "time"
13 :
14 : "github.com/cockroachdb/errors"
15 : "github.com/cockroachdb/pebble/internal/base"
16 : "github.com/cockroachdb/pebble/internal/cache"
17 : "github.com/cockroachdb/pebble/internal/invariants"
18 : "github.com/cockroachdb/pebble/internal/keyspan"
19 : "github.com/cockroachdb/pebble/internal/manifest"
20 : "github.com/cockroachdb/pebble/internal/overlap"
21 : "github.com/cockroachdb/pebble/internal/sstableinternal"
22 : "github.com/cockroachdb/pebble/objstorage"
23 : "github.com/cockroachdb/pebble/objstorage/remote"
24 : "github.com/cockroachdb/pebble/sstable"
25 : )
26 :
27 2 : func sstableKeyCompare(userCmp Compare, a, b InternalKey) int {
28 2 : c := userCmp(a.UserKey, b.UserKey)
29 2 : if c != 0 {
30 2 : return c
31 2 : }
32 2 : if a.IsExclusiveSentinel() {
33 2 : if !b.IsExclusiveSentinel() {
34 2 : return -1
35 2 : }
36 2 : } else if b.IsExclusiveSentinel() {
37 2 : return +1
38 2 : }
39 2 : return 0
40 : }
41 :
42 : // KeyRange encodes a key range in user key space. A KeyRange's Start is
43 : // inclusive while its End is exclusive.
44 : //
45 : // KeyRange is equivalent to base.UserKeyBounds with exclusive end.
46 : type KeyRange struct {
47 : Start, End []byte
48 : }
49 :
50 : // Valid returns true if the KeyRange is defined.
51 2 : func (k *KeyRange) Valid() bool {
52 2 : return k.Start != nil && k.End != nil
53 2 : }
54 :
55 : // Contains returns whether the specified key exists in the KeyRange.
56 2 : func (k *KeyRange) Contains(cmp base.Compare, key InternalKey) bool {
57 2 : v := cmp(key.UserKey, k.End)
58 2 : return (v < 0 || (v == 0 && key.IsExclusiveSentinel())) && cmp(k.Start, key.UserKey) <= 0
59 2 : }
60 :
61 : // UserKeyBounds returns the KeyRange as UserKeyBounds. Also implements the internal `bounded` interface.
62 2 : func (k KeyRange) UserKeyBounds() base.UserKeyBounds {
63 2 : return base.UserKeyBoundsEndExclusive(k.Start, k.End)
64 2 : }
65 :
66 : // OverlapsInternalKeyRange checks if the specified internal key range has an
67 : // overlap with the KeyRange. Note that we aren't checking for full containment
68 : // of smallest-largest within k, rather just that there's some intersection
69 : // between the two ranges.
70 2 : func (k *KeyRange) OverlapsInternalKeyRange(cmp base.Compare, smallest, largest InternalKey) bool {
71 2 : ukb := k.UserKeyBounds()
72 2 : b := base.UserKeyBoundsFromInternal(smallest, largest)
73 2 : return ukb.Overlaps(cmp, &b)
74 2 : }
75 :
76 : // Overlaps checks if the specified file has an overlap with the KeyRange.
77 : // Note that we aren't checking for full containment of m within k, rather just
78 : // that there's some intersection between m and k's bounds.
79 1 : func (k *KeyRange) Overlaps(cmp base.Compare, m *fileMetadata) bool {
80 1 : b := k.UserKeyBounds()
81 1 : return m.Overlaps(cmp, &b)
82 1 : }
83 :
84 : // OverlapsKeyRange checks if this span overlaps with the provided KeyRange.
85 : // Note that we aren't checking for full containment of either span in the other,
86 : // just that there's a key x that is in both key ranges.
87 2 : func (k *KeyRange) OverlapsKeyRange(cmp Compare, span KeyRange) bool {
88 2 : return cmp(k.Start, span.End) < 0 && cmp(k.End, span.Start) > 0
89 2 : }
90 :
91 2 : func ingestValidateKey(opts *Options, key *InternalKey) error {
92 2 : if key.Kind() == InternalKeyKindInvalid {
93 1 : return base.CorruptionErrorf("pebble: external sstable has corrupted key: %s",
94 1 : key.Pretty(opts.Comparer.FormatKey))
95 1 : }
96 2 : if key.SeqNum() != 0 {
97 1 : return base.CorruptionErrorf("pebble: external sstable has non-zero seqnum: %s",
98 1 : key.Pretty(opts.Comparer.FormatKey))
99 1 : }
100 2 : return nil
101 : }
102 :
103 : // ingestSynthesizeShared constructs a fileMetadata for one shared sstable owned
104 : // or shared by another node.
105 : func ingestSynthesizeShared(
106 : opts *Options, sm SharedSSTMeta, fileNum base.FileNum,
107 2 : ) (*fileMetadata, error) {
108 2 : if sm.Size == 0 {
109 0 : // Disallow 0 file sizes
110 0 : return nil, errors.New("pebble: cannot ingest shared file with size 0")
111 0 : }
112 : // Don't load table stats. Doing a round trip to shared storage, one SST
113 : // at a time is not worth it as it slows down ingestion.
114 2 : meta := &fileMetadata{
115 2 : FileNum: fileNum,
116 2 : CreationTime: time.Now().Unix(),
117 2 : Virtual: true,
118 2 : Size: sm.Size,
119 2 : }
120 2 : // For simplicity, we use the same number for both the FileNum and the
121 2 : // DiskFileNum (even though this is a virtual sstable). Pass the underlying
122 2 : // FileBacking's size to the same size as the virtualized view of the sstable.
123 2 : // This ensures that we don't over-prioritize this sstable for compaction just
124 2 : // yet, as we do not have a clear sense of what parts of this sstable are
125 2 : // referenced by other nodes.
126 2 : meta.InitProviderBacking(base.DiskFileNum(fileNum), sm.Size)
127 2 :
128 2 : if sm.LargestPointKey.Valid() && sm.LargestPointKey.UserKey != nil {
129 2 : // Initialize meta.{HasPointKeys,Smallest,Largest}, etc.
130 2 : //
131 2 : // NB: We create new internal keys and pass them into ExtendPointKeyBounds
132 2 : // so that we can sub a zero sequence number into the bounds. We can set
133 2 : // the sequence number to anything here; it'll be reset in ingestUpdateSeqNum
134 2 : // anyway. However, we do need to use the same sequence number across all
135 2 : // bound keys at this step so that we end up with bounds that are consistent
136 2 : // across point/range keys.
137 2 : //
138 2 : // Because of the sequence number rewriting, we cannot use the Kind of
139 2 : // sm.SmallestPointKey. For example, the original SST might start with
140 2 : // a.SET.2 and a.RANGEDEL.1 (with a.SET.2 being the smallest key); after
141 2 : // rewriting the sequence numbers, these keys become a.SET.100 and
142 2 : // a.RANGEDEL.100, with a.RANGEDEL.100 being the smallest key. To create a
143 2 : // correct bound, we just use the maximum key kind (which sorts first).
144 2 : // Similarly, we use the smallest key kind for the largest key.
145 2 : smallestPointKey := base.MakeInternalKey(sm.SmallestPointKey.UserKey, 0, base.InternalKeyKindMaxForSSTable)
146 2 : largestPointKey := base.MakeInternalKey(sm.LargestPointKey.UserKey, 0, 0)
147 2 : if sm.LargestPointKey.IsExclusiveSentinel() {
148 2 : largestPointKey = base.MakeRangeDeleteSentinelKey(sm.LargestPointKey.UserKey)
149 2 : }
150 2 : if opts.Comparer.Equal(smallestPointKey.UserKey, largestPointKey.UserKey) &&
151 2 : smallestPointKey.Trailer < largestPointKey.Trailer {
152 0 : // We get kinds from the sender, however we substitute our own sequence
153 0 : // numbers. This can result in cases where an sstable [b#5,SET-b#4,DELSIZED]
154 0 : // becomes [b#0,SET-b#0,DELSIZED] when we synthesize it here, but the
155 0 : // kinds need to be reversed now because DelSized > Set.
156 0 : smallestPointKey, largestPointKey = largestPointKey, smallestPointKey
157 0 : }
158 2 : meta.ExtendPointKeyBounds(opts.Comparer.Compare, smallestPointKey, largestPointKey)
159 : }
160 2 : if sm.LargestRangeKey.Valid() && sm.LargestRangeKey.UserKey != nil {
161 2 : // Initialize meta.{HasRangeKeys,Smallest,Largest}, etc.
162 2 : //
163 2 : // See comment above on why we use a zero sequence number and these key
164 2 : // kinds here.
165 2 : smallestRangeKey := base.MakeInternalKey(sm.SmallestRangeKey.UserKey, 0, base.InternalKeyKindRangeKeyMax)
166 2 : largestRangeKey := base.MakeExclusiveSentinelKey(base.InternalKeyKindRangeKeyMin, sm.LargestRangeKey.UserKey)
167 2 : meta.ExtendRangeKeyBounds(opts.Comparer.Compare, smallestRangeKey, largestRangeKey)
168 2 : }
169 2 : if err := meta.Validate(opts.Comparer.Compare, opts.Comparer.FormatKey); err != nil {
170 0 : return nil, err
171 0 : }
172 2 : return meta, nil
173 : }
174 :
175 : // ingestLoad1External loads the fileMetadata for one external sstable.
176 : // Sequence number and target level calculation happens during prepare/apply.
177 : func ingestLoad1External(
178 : opts *Options, e ExternalFile, fileNum base.FileNum,
179 2 : ) (*fileMetadata, error) {
180 2 : if e.Size == 0 {
181 0 : return nil, errors.New("pebble: cannot ingest external file with size 0")
182 0 : }
183 2 : if !e.HasRangeKey && !e.HasPointKey {
184 0 : return nil, errors.New("pebble: cannot ingest external file with no point or range keys")
185 0 : }
186 :
187 2 : if opts.Comparer.Compare(e.StartKey, e.EndKey) > 0 {
188 1 : return nil, errors.Newf("pebble: external file bounds [%q, %q) are invalid", e.StartKey, e.EndKey)
189 1 : }
190 2 : if opts.Comparer.Compare(e.StartKey, e.EndKey) == 0 && !e.EndKeyIsInclusive {
191 0 : return nil, errors.Newf("pebble: external file bounds [%q, %q) are invalid", e.StartKey, e.EndKey)
192 0 : }
193 2 : if n := opts.Comparer.Split(e.StartKey); n != len(e.StartKey) {
194 1 : return nil, errors.Newf("pebble: external file bounds start key %q has suffix", e.StartKey)
195 1 : }
196 2 : if n := opts.Comparer.Split(e.EndKey); n != len(e.EndKey) {
197 1 : return nil, errors.Newf("pebble: external file bounds end key %q has suffix", e.EndKey)
198 1 : }
199 :
200 : // Don't load table stats. Doing a round trip to shared storage, one SST
201 : // at a time is not worth it as it slows down ingestion.
202 2 : meta := &fileMetadata{
203 2 : FileNum: fileNum,
204 2 : CreationTime: time.Now().Unix(),
205 2 : Virtual: true,
206 2 : Size: e.Size,
207 2 : }
208 2 :
209 2 : // In the name of keeping this ingestion as fast as possible, we avoid
210 2 : // *all* existence checks and synthesize a file metadata with smallest/largest
211 2 : // keys that overlap whatever the passed-in span was.
212 2 : smallestCopy := slices.Clone(e.StartKey)
213 2 : largestCopy := slices.Clone(e.EndKey)
214 2 : if e.HasPointKey {
215 2 : // Sequence numbers are updated later by
216 2 : // ingestUpdateSeqNum, applying a squence number that
217 2 : // is applied to all keys in the sstable.
218 2 : if e.EndKeyIsInclusive {
219 1 : meta.ExtendPointKeyBounds(
220 1 : opts.Comparer.Compare,
221 1 : base.MakeInternalKey(smallestCopy, 0, base.InternalKeyKindMaxForSSTable),
222 1 : base.MakeInternalKey(largestCopy, 0, 0))
223 2 : } else {
224 2 : meta.ExtendPointKeyBounds(
225 2 : opts.Comparer.Compare,
226 2 : base.MakeInternalKey(smallestCopy, 0, base.InternalKeyKindMaxForSSTable),
227 2 : base.MakeRangeDeleteSentinelKey(largestCopy))
228 2 : }
229 : }
230 2 : if e.HasRangeKey {
231 2 : meta.ExtendRangeKeyBounds(
232 2 : opts.Comparer.Compare,
233 2 : base.MakeInternalKey(smallestCopy, 0, InternalKeyKindRangeKeyMax),
234 2 : base.MakeExclusiveSentinelKey(InternalKeyKindRangeKeyMin, largestCopy),
235 2 : )
236 2 : }
237 :
238 2 : meta.SyntheticPrefixAndSuffix = sstable.MakeSyntheticPrefixAndSuffix(e.SyntheticPrefix, e.SyntheticSuffix)
239 2 :
240 2 : return meta, nil
241 : }
242 :
243 : type rangeKeyIngestValidator struct {
244 : // lastRangeKey is the last range key seen in the previous file.
245 : lastRangeKey keyspan.Span
246 : // comparer, if unset, disables range key validation.
247 : comparer *base.Comparer
248 : }
249 :
250 2 : func disableRangeKeyChecks() rangeKeyIngestValidator {
251 2 : return rangeKeyIngestValidator{}
252 2 : }
253 :
254 : func validateSuffixedBoundaries(
255 : cmp *base.Comparer, lastRangeKey keyspan.Span,
256 2 : ) rangeKeyIngestValidator {
257 2 : return rangeKeyIngestValidator{
258 2 : lastRangeKey: lastRangeKey,
259 2 : comparer: cmp,
260 2 : }
261 2 : }
262 :
263 : // Validate valides if the stored state of this rangeKeyIngestValidator allows for
264 : // a file with the given nextFileSmallestKey to be ingested, such that the stored
265 : // last file's largest range key defragments cleanly with the next file's smallest
266 : // key if it was suffixed. If a value of nil is passed in for nextFileSmallestKey,
267 : // that denotes the next file does not have a range key or there is no next file.
268 2 : func (r *rangeKeyIngestValidator) Validate(nextFileSmallestKey *keyspan.Span) error {
269 2 : if r.comparer == nil {
270 2 : return nil
271 2 : }
272 2 : if r.lastRangeKey.Valid() {
273 2 : if r.comparer.Split.HasSuffix(r.lastRangeKey.End) {
274 1 : if nextFileSmallestKey == nil || !r.comparer.Equal(r.lastRangeKey.End, nextFileSmallestKey.Start) {
275 1 : // The last range key has a suffix, and it doesn't defragment cleanly with this range key.
276 1 : return errors.AssertionFailedf("pebble: ingest sstable has suffixed largest range key that does not match the start key of the next sstable: %s",
277 1 : r.comparer.FormatKey(r.lastRangeKey.End))
278 1 : } else if !keyspan.DefragmentInternal.ShouldDefragment(r.comparer.CompareRangeSuffixes, &r.lastRangeKey, nextFileSmallestKey) {
279 0 : // The last range key has a suffix, and it doesn't defragment cleanly with this range key.
280 0 : return errors.AssertionFailedf("pebble: ingest sstable has suffixed range key that won't defragment with next sstable: %s",
281 0 : r.comparer.FormatKey(r.lastRangeKey.End))
282 0 : }
283 : }
284 2 : } else if nextFileSmallestKey != nil && r.comparer.Split.HasSuffix(nextFileSmallestKey.Start) {
285 0 : return errors.Newf("pebble: ingest sstable has suffixed range key start that won't defragment: %s",
286 0 : r.comparer.FormatKey(nextFileSmallestKey.Start))
287 0 : }
288 2 : return nil
289 : }
290 :
291 : // ingestLoad1 creates the FileMetadata for one file. This file will be owned
292 : // by this store.
293 : //
294 : // prevLastRangeKey is the last range key from the previous file. It is used to
295 : // ensure that the range keys defragment cleanly across files. These checks
296 : // are disabled if disableRangeKeyChecks is true.
297 : func ingestLoad1(
298 : ctx context.Context,
299 : opts *Options,
300 : fmv FormatMajorVersion,
301 : readable objstorage.Readable,
302 : cacheID cache.ID,
303 : fileNum base.FileNum,
304 : rangeKeyValidator rangeKeyIngestValidator,
305 2 : ) (meta *fileMetadata, lastRangeKey keyspan.Span, err error) {
306 2 : o := opts.MakeReaderOptions()
307 2 : o.SetInternalCacheOpts(sstableinternal.CacheOptions{
308 2 : Cache: opts.Cache,
309 2 : CacheID: cacheID,
310 2 : FileNum: base.PhysicalTableDiskFileNum(fileNum),
311 2 : })
312 2 : r, err := sstable.NewReader(ctx, readable, o)
313 2 : if err != nil {
314 1 : return nil, keyspan.Span{}, err
315 1 : }
316 2 : defer r.Close()
317 2 :
318 2 : // Avoid ingesting tables with format versions this DB doesn't support.
319 2 : tf, err := r.TableFormat()
320 2 : if err != nil {
321 0 : return nil, keyspan.Span{}, err
322 0 : }
323 2 : if tf < fmv.MinTableFormat() || tf > fmv.MaxTableFormat() {
324 1 : return nil, keyspan.Span{}, errors.Newf(
325 1 : "pebble: table format %s is not within range supported at DB format major version %d, (%s,%s)",
326 1 : tf, fmv, fmv.MinTableFormat(), fmv.MaxTableFormat(),
327 1 : )
328 1 : }
329 2 : if tf.BlockColumnar() {
330 2 : if _, ok := opts.KeySchemas[r.Properties.KeySchemaName]; !ok {
331 0 : return nil, keyspan.Span{}, errors.Newf(
332 0 : "pebble: table uses key schema %q unknown to the database",
333 0 : r.Properties.KeySchemaName)
334 0 : }
335 : }
336 :
337 2 : meta = &fileMetadata{}
338 2 : meta.FileNum = fileNum
339 2 : meta.Size = uint64(readable.Size())
340 2 : meta.CreationTime = time.Now().Unix()
341 2 : meta.InitPhysicalBacking()
342 2 :
343 2 : // Avoid loading into the file cache for collecting stats if we
344 2 : // don't need to. If there are no range deletions, we have all the
345 2 : // information to compute the stats here.
346 2 : //
347 2 : // This is helpful in tests for avoiding awkwardness around deletion of
348 2 : // ingested files from MemFS. MemFS implements the Windows semantics of
349 2 : // disallowing removal of an open file. Under MemFS, if we don't populate
350 2 : // meta.Stats here, the file will be loaded into the file cache for
351 2 : // calculating stats before we can remove the original link.
352 2 : maybeSetStatsFromProperties(meta.PhysicalMeta(), &r.Properties)
353 2 :
354 2 : {
355 2 : iter, err := r.NewIter(sstable.NoTransforms, nil /* lower */, nil /* upper */)
356 2 : if err != nil {
357 1 : return nil, keyspan.Span{}, err
358 1 : }
359 2 : defer iter.Close()
360 2 : var smallest InternalKey
361 2 : if kv := iter.First(); kv != nil {
362 2 : if err := ingestValidateKey(opts, &kv.K); err != nil {
363 1 : return nil, keyspan.Span{}, err
364 1 : }
365 2 : smallest = kv.K.Clone()
366 : }
367 2 : if err := iter.Error(); err != nil {
368 1 : return nil, keyspan.Span{}, err
369 1 : }
370 2 : if kv := iter.Last(); kv != nil {
371 2 : if err := ingestValidateKey(opts, &kv.K); err != nil {
372 0 : return nil, keyspan.Span{}, err
373 0 : }
374 2 : meta.ExtendPointKeyBounds(opts.Comparer.Compare, smallest, kv.K.Clone())
375 : }
376 2 : if err := iter.Error(); err != nil {
377 1 : return nil, keyspan.Span{}, err
378 1 : }
379 : }
380 :
381 2 : iter, err := r.NewRawRangeDelIter(ctx, sstable.NoFragmentTransforms)
382 2 : if err != nil {
383 0 : return nil, keyspan.Span{}, err
384 0 : }
385 2 : if iter != nil {
386 2 : defer iter.Close()
387 2 : var smallest InternalKey
388 2 : if s, err := iter.First(); err != nil {
389 0 : return nil, keyspan.Span{}, err
390 2 : } else if s != nil {
391 2 : key := s.SmallestKey()
392 2 : if err := ingestValidateKey(opts, &key); err != nil {
393 0 : return nil, keyspan.Span{}, err
394 0 : }
395 2 : smallest = key.Clone()
396 : }
397 2 : if s, err := iter.Last(); err != nil {
398 0 : return nil, keyspan.Span{}, err
399 2 : } else if s != nil {
400 2 : k := s.SmallestKey()
401 2 : if err := ingestValidateKey(opts, &k); err != nil {
402 0 : return nil, keyspan.Span{}, err
403 0 : }
404 2 : largest := s.LargestKey().Clone()
405 2 : meta.ExtendPointKeyBounds(opts.Comparer.Compare, smallest, largest)
406 : }
407 : }
408 :
409 : // Update the range-key bounds for the table.
410 2 : {
411 2 : iter, err := r.NewRawRangeKeyIter(ctx, sstable.NoFragmentTransforms)
412 2 : if err != nil {
413 0 : return nil, keyspan.Span{}, err
414 0 : }
415 2 : if iter != nil {
416 2 : defer iter.Close()
417 2 : var smallest InternalKey
418 2 : if s, err := iter.First(); err != nil {
419 0 : return nil, keyspan.Span{}, err
420 2 : } else if s != nil {
421 2 : key := s.SmallestKey()
422 2 : if err := ingestValidateKey(opts, &key); err != nil {
423 0 : return nil, keyspan.Span{}, err
424 0 : }
425 2 : smallest = key.Clone()
426 2 : // Range keys need some additional validation as we need to ensure they
427 2 : // defragment cleanly with the lastRangeKey from the previous file.
428 2 : if err := rangeKeyValidator.Validate(s); err != nil {
429 0 : return nil, keyspan.Span{}, err
430 0 : }
431 : }
432 2 : lastRangeKey = keyspan.Span{}
433 2 : if s, err := iter.Last(); err != nil {
434 0 : return nil, keyspan.Span{}, err
435 2 : } else if s != nil {
436 2 : k := s.SmallestKey()
437 2 : if err := ingestValidateKey(opts, &k); err != nil {
438 0 : return nil, keyspan.Span{}, err
439 0 : }
440 : // As range keys are fragmented, the end key of the last range key in
441 : // the table provides the upper bound for the table.
442 2 : largest := s.LargestKey().Clone()
443 2 : meta.ExtendRangeKeyBounds(opts.Comparer.Compare, smallest, largest)
444 2 : lastRangeKey = s.Clone()
445 0 : } else {
446 0 : // s == nil.
447 0 : rangeKeyValidator.Validate(nil /* nextFileSmallestKey */)
448 0 : }
449 2 : } else {
450 2 : rangeKeyValidator.Validate(nil /* nextFileSmallestKey */)
451 2 : lastRangeKey = keyspan.Span{}
452 2 : }
453 : }
454 :
455 2 : if !meta.HasPointKeys && !meta.HasRangeKeys {
456 2 : return nil, keyspan.Span{}, nil
457 2 : }
458 :
459 : // Sanity check that the various bounds on the file were set consistently.
460 2 : if err := meta.Validate(opts.Comparer.Compare, opts.Comparer.FormatKey); err != nil {
461 0 : return nil, keyspan.Span{}, err
462 0 : }
463 :
464 2 : return meta, lastRangeKey, nil
465 : }
466 :
467 : type ingestLoadResult struct {
468 : local []ingestLocalMeta
469 : shared []ingestSharedMeta
470 : external []ingestExternalMeta
471 :
472 : externalFilesHaveLevel bool
473 : }
474 :
475 : type ingestLocalMeta struct {
476 : *fileMetadata
477 : path string
478 : }
479 :
480 : type ingestSharedMeta struct {
481 : *fileMetadata
482 : shared SharedSSTMeta
483 : }
484 :
485 : type ingestExternalMeta struct {
486 : *fileMetadata
487 : external ExternalFile
488 : // usedExistingBacking is true if the external file is reusing a backing
489 : // that existed before this ingestion. In this case, we called
490 : // VirtualBackings.Protect() on that backing; we will need to call
491 : // Unprotect() after the ingestion.
492 : usedExistingBacking bool
493 : }
494 :
495 2 : func (r *ingestLoadResult) fileCount() int {
496 2 : return len(r.local) + len(r.shared) + len(r.external)
497 2 : }
498 :
499 : func ingestLoad(
500 : ctx context.Context,
501 : opts *Options,
502 : fmv FormatMajorVersion,
503 : paths []string,
504 : shared []SharedSSTMeta,
505 : external []ExternalFile,
506 : cacheID cache.ID,
507 : pending []base.FileNum,
508 2 : ) (ingestLoadResult, error) {
509 2 : localFileNums := pending[:len(paths)]
510 2 : sharedFileNums := pending[len(paths) : len(paths)+len(shared)]
511 2 : externalFileNums := pending[len(paths)+len(shared) : len(paths)+len(shared)+len(external)]
512 2 :
513 2 : var result ingestLoadResult
514 2 : result.local = make([]ingestLocalMeta, 0, len(paths))
515 2 : var lastRangeKey keyspan.Span
516 2 : // NB: we disable range key boundary assertions if we have shared or external files
517 2 : // present in this ingestion. This is because a suffixed range key in a local file
518 2 : // can possibly defragment with a suffixed range key in a shared or external file.
519 2 : // We also disable range key boundary assertions if we have CreateOnShared set to
520 2 : // true, as that means we could have suffixed RangeKeyDels or Unsets in the local
521 2 : // files that won't ever be surfaced, even if there are no shared or external files
522 2 : // in the ingestion.
523 2 : shouldDisableRangeKeyChecks := len(shared) > 0 || len(external) > 0 || opts.Experimental.CreateOnShared != remote.CreateOnSharedNone
524 2 : for i := range paths {
525 2 : f, err := opts.FS.Open(paths[i])
526 2 : if err != nil {
527 1 : return ingestLoadResult{}, err
528 1 : }
529 :
530 2 : readable, err := sstable.NewSimpleReadable(f)
531 2 : if err != nil {
532 1 : return ingestLoadResult{}, err
533 1 : }
534 2 : var m *fileMetadata
535 2 : rangeKeyValidator := disableRangeKeyChecks()
536 2 : if !shouldDisableRangeKeyChecks {
537 2 : rangeKeyValidator = validateSuffixedBoundaries(opts.Comparer, lastRangeKey)
538 2 : }
539 2 : m, lastRangeKey, err = ingestLoad1(ctx, opts, fmv, readable, cacheID, localFileNums[i], rangeKeyValidator)
540 2 : if err != nil {
541 1 : return ingestLoadResult{}, err
542 1 : }
543 2 : if m != nil {
544 2 : result.local = append(result.local, ingestLocalMeta{
545 2 : fileMetadata: m,
546 2 : path: paths[i],
547 2 : })
548 2 : }
549 : }
550 :
551 2 : if !shouldDisableRangeKeyChecks {
552 2 : rangeKeyValidator := validateSuffixedBoundaries(opts.Comparer, lastRangeKey)
553 2 : if err := rangeKeyValidator.Validate(nil /* nextFileSmallestKey */); err != nil {
554 1 : return ingestLoadResult{}, err
555 1 : }
556 : }
557 :
558 : // Sort the shared files according to level.
559 2 : sort.Sort(sharedByLevel(shared))
560 2 :
561 2 : result.shared = make([]ingestSharedMeta, 0, len(shared))
562 2 : for i := range shared {
563 2 : m, err := ingestSynthesizeShared(opts, shared[i], sharedFileNums[i])
564 2 : if err != nil {
565 0 : return ingestLoadResult{}, err
566 0 : }
567 2 : if shared[i].Level < sharedLevelsStart {
568 0 : return ingestLoadResult{}, errors.New("cannot ingest shared file in level below sharedLevelsStart")
569 0 : }
570 2 : result.shared = append(result.shared, ingestSharedMeta{
571 2 : fileMetadata: m,
572 2 : shared: shared[i],
573 2 : })
574 : }
575 2 : result.external = make([]ingestExternalMeta, 0, len(external))
576 2 : for i := range external {
577 2 : m, err := ingestLoad1External(opts, external[i], externalFileNums[i])
578 2 : if err != nil {
579 1 : return ingestLoadResult{}, err
580 1 : }
581 2 : result.external = append(result.external, ingestExternalMeta{
582 2 : fileMetadata: m,
583 2 : external: external[i],
584 2 : })
585 2 : if external[i].Level > 0 {
586 1 : if i != 0 && !result.externalFilesHaveLevel {
587 0 : return ingestLoadResult{}, base.AssertionFailedf("pebble: external sstables must all have level set or unset")
588 0 : }
589 1 : result.externalFilesHaveLevel = true
590 2 : } else if result.externalFilesHaveLevel {
591 0 : return ingestLoadResult{}, base.AssertionFailedf("pebble: external sstables must all have level set or unset")
592 0 : }
593 : }
594 2 : return result, nil
595 : }
596 :
597 2 : func ingestSortAndVerify(cmp Compare, lr ingestLoadResult, exciseSpan KeyRange) error {
598 2 : // Verify that all the shared files (i.e. files in sharedMeta)
599 2 : // fit within the exciseSpan.
600 2 : for _, f := range lr.shared {
601 2 : if !exciseSpan.Contains(cmp, f.Smallest) || !exciseSpan.Contains(cmp, f.Largest) {
602 0 : return errors.Newf("pebble: shared file outside of excise span, span [%s-%s), file = %s", exciseSpan.Start, exciseSpan.End, f.String())
603 0 : }
604 : }
605 :
606 2 : if lr.externalFilesHaveLevel {
607 1 : for _, f := range lr.external {
608 1 : if !exciseSpan.Contains(cmp, f.Smallest) || !exciseSpan.Contains(cmp, f.Largest) {
609 0 : return base.AssertionFailedf("pebble: external file outside of excise span, span [%s-%s), file = %s", exciseSpan.Start, exciseSpan.End, f.String())
610 0 : }
611 : }
612 : }
613 :
614 2 : if len(lr.external) > 0 {
615 2 : if len(lr.shared) > 0 {
616 0 : // If external files are present alongside shared files,
617 0 : // return an error.
618 0 : return base.AssertionFailedf("pebble: external files cannot be ingested atomically alongside shared files")
619 0 : }
620 :
621 : // Sort according to the smallest key.
622 2 : slices.SortFunc(lr.external, func(a, b ingestExternalMeta) int {
623 2 : return cmp(a.Smallest.UserKey, b.Smallest.UserKey)
624 2 : })
625 2 : for i := 1; i < len(lr.external); i++ {
626 2 : if sstableKeyCompare(cmp, lr.external[i-1].Largest, lr.external[i].Smallest) >= 0 {
627 1 : return errors.Newf("pebble: external sstables have overlapping ranges")
628 1 : }
629 : }
630 2 : return nil
631 : }
632 2 : if len(lr.local) <= 1 {
633 2 : return nil
634 2 : }
635 :
636 : // Sort according to the smallest key.
637 2 : slices.SortFunc(lr.local, func(a, b ingestLocalMeta) int {
638 2 : return cmp(a.Smallest.UserKey, b.Smallest.UserKey)
639 2 : })
640 :
641 2 : for i := 1; i < len(lr.local); i++ {
642 2 : if sstableKeyCompare(cmp, lr.local[i-1].Largest, lr.local[i].Smallest) >= 0 {
643 2 : return errors.Newf("pebble: local ingestion sstables have overlapping ranges")
644 2 : }
645 : }
646 2 : if len(lr.shared) == 0 {
647 2 : return nil
648 2 : }
649 0 : filesInLevel := make([]*fileMetadata, 0, len(lr.shared))
650 0 : for l := sharedLevelsStart; l < numLevels; l++ {
651 0 : filesInLevel = filesInLevel[:0]
652 0 : for i := range lr.shared {
653 0 : if lr.shared[i].shared.Level == uint8(l) {
654 0 : filesInLevel = append(filesInLevel, lr.shared[i].fileMetadata)
655 0 : }
656 : }
657 0 : for i := range lr.external {
658 0 : if lr.external[i].external.Level == uint8(l) {
659 0 : filesInLevel = append(filesInLevel, lr.external[i].fileMetadata)
660 0 : }
661 : }
662 0 : slices.SortFunc(filesInLevel, func(a, b *fileMetadata) int {
663 0 : return cmp(a.Smallest.UserKey, b.Smallest.UserKey)
664 0 : })
665 0 : for i := 1; i < len(filesInLevel); i++ {
666 0 : if sstableKeyCompare(cmp, filesInLevel[i-1].Largest, filesInLevel[i].Smallest) >= 0 {
667 0 : return base.AssertionFailedf("pebble: external shared sstables have overlapping ranges")
668 0 : }
669 : }
670 : }
671 0 : return nil
672 : }
673 :
674 1 : func ingestCleanup(objProvider objstorage.Provider, meta []ingestLocalMeta) error {
675 1 : var firstErr error
676 1 : for i := range meta {
677 1 : if err := objProvider.Remove(fileTypeTable, meta[i].FileBacking.DiskFileNum); err != nil {
678 1 : firstErr = firstError(firstErr, err)
679 1 : }
680 : }
681 1 : return firstErr
682 : }
683 :
684 : // ingestLinkLocal creates new objects which are backed by either hardlinks to or
685 : // copies of the ingested files.
686 : func ingestLinkLocal(
687 : ctx context.Context,
688 : jobID JobID,
689 : opts *Options,
690 : objProvider objstorage.Provider,
691 : localMetas []ingestLocalMeta,
692 2 : ) error {
693 2 : for i := range localMetas {
694 2 : objMeta, err := objProvider.LinkOrCopyFromLocal(
695 2 : ctx, opts.FS, localMetas[i].path, fileTypeTable, localMetas[i].FileBacking.DiskFileNum,
696 2 : objstorage.CreateOptions{PreferSharedStorage: true},
697 2 : )
698 2 : if err != nil {
699 1 : if err2 := ingestCleanup(objProvider, localMetas[:i]); err2 != nil {
700 0 : opts.Logger.Errorf("ingest cleanup failed: %v", err2)
701 0 : }
702 1 : return err
703 : }
704 2 : if opts.EventListener.TableCreated != nil {
705 2 : opts.EventListener.TableCreated(TableCreateInfo{
706 2 : JobID: int(jobID),
707 2 : Reason: "ingesting",
708 2 : Path: objProvider.Path(objMeta),
709 2 : FileNum: base.PhysicalTableDiskFileNum(localMetas[i].FileNum),
710 2 : })
711 2 : }
712 : }
713 2 : return nil
714 : }
715 :
716 : // ingestAttachRemote attaches remote objects to the storage provider.
717 : //
718 : // For external objects, we reuse existing FileBackings from the current version
719 : // when possible.
720 : //
721 : // ingestUnprotectExternalBackings() must be called after this function (even in
722 : // error cases).
723 2 : func (d *DB) ingestAttachRemote(jobID JobID, lr ingestLoadResult) error {
724 2 : remoteObjs := make([]objstorage.RemoteObjectToAttach, 0, len(lr.shared)+len(lr.external))
725 2 : for i := range lr.shared {
726 2 : backing, err := lr.shared[i].shared.Backing.Get()
727 2 : if err != nil {
728 0 : return err
729 0 : }
730 2 : remoteObjs = append(remoteObjs, objstorage.RemoteObjectToAttach{
731 2 : FileNum: lr.shared[i].FileBacking.DiskFileNum,
732 2 : FileType: fileTypeTable,
733 2 : Backing: backing,
734 2 : })
735 : }
736 :
737 2 : d.findExistingBackingsForExternalObjects(lr.external)
738 2 :
739 2 : newFileBackings := make(map[remote.ObjectKey]*fileBacking, len(lr.external))
740 2 : for i := range lr.external {
741 2 : meta := lr.external[i].fileMetadata
742 2 : if meta.FileBacking != nil {
743 2 : // The backing was filled in by findExistingBackingsForExternalObjects().
744 2 : continue
745 : }
746 2 : key := remote.MakeObjectKey(lr.external[i].external.Locator, lr.external[i].external.ObjName)
747 2 : if backing, ok := newFileBackings[key]; ok {
748 2 : // We already created the same backing in this loop.
749 2 : meta.FileBacking = backing
750 2 : continue
751 : }
752 2 : providerBacking, err := d.objProvider.CreateExternalObjectBacking(key.Locator, key.ObjectName)
753 2 : if err != nil {
754 0 : return err
755 0 : }
756 : // We have to attach the remote object (and assign it a DiskFileNum). For
757 : // simplicity, we use the same number for both the FileNum and the
758 : // DiskFileNum (even though this is a virtual sstable).
759 2 : meta.InitProviderBacking(base.DiskFileNum(meta.FileNum), lr.external[i].external.Size)
760 2 :
761 2 : // Set the underlying FileBacking's size to the same size as the virtualized
762 2 : // view of the sstable. This ensures that we don't over-prioritize this
763 2 : // sstable for compaction just yet, as we do not have a clear sense of
764 2 : // what parts of this sstable are referenced by other nodes.
765 2 : meta.FileBacking.Size = lr.external[i].external.Size
766 2 : newFileBackings[key] = meta.FileBacking
767 2 :
768 2 : remoteObjs = append(remoteObjs, objstorage.RemoteObjectToAttach{
769 2 : FileNum: meta.FileBacking.DiskFileNum,
770 2 : FileType: fileTypeTable,
771 2 : Backing: providerBacking,
772 2 : })
773 : }
774 :
775 2 : for i := range lr.external {
776 2 : if err := lr.external[i].Validate(d.opts.Comparer.Compare, d.opts.Comparer.FormatKey); err != nil {
777 0 : return err
778 0 : }
779 : }
780 :
781 2 : remoteObjMetas, err := d.objProvider.AttachRemoteObjects(remoteObjs)
782 2 : if err != nil {
783 0 : return err
784 0 : }
785 :
786 2 : for i := range lr.shared {
787 2 : // One corner case around file sizes we need to be mindful of, is that
788 2 : // if one of the shareObjs was initially created by us (and has boomeranged
789 2 : // back from another node), we'll need to update the FileBacking's size
790 2 : // to be the true underlying size. Otherwise, we could hit errors when we
791 2 : // open the db again after a crash/restart (see checkConsistency in open.go),
792 2 : // plus it more accurately allows us to prioritize compactions of files
793 2 : // that were originally created by us.
794 2 : if remoteObjMetas[i].IsShared() && !d.objProvider.IsSharedForeign(remoteObjMetas[i]) {
795 2 : size, err := d.objProvider.Size(remoteObjMetas[i])
796 2 : if err != nil {
797 0 : return err
798 0 : }
799 2 : lr.shared[i].FileBacking.Size = uint64(size)
800 : }
801 : }
802 :
803 2 : if d.opts.EventListener.TableCreated != nil {
804 2 : for i := range remoteObjMetas {
805 2 : d.opts.EventListener.TableCreated(TableCreateInfo{
806 2 : JobID: int(jobID),
807 2 : Reason: "ingesting",
808 2 : Path: d.objProvider.Path(remoteObjMetas[i]),
809 2 : FileNum: remoteObjMetas[i].DiskFileNum,
810 2 : })
811 2 : }
812 : }
813 :
814 2 : return nil
815 : }
816 :
817 : // findExistingBackingsForExternalObjects populates the FileBacking for external
818 : // files which are already in use by the current version.
819 : //
820 : // We take a Ref and LatestRef on populated backings.
821 2 : func (d *DB) findExistingBackingsForExternalObjects(metas []ingestExternalMeta) {
822 2 : d.mu.Lock()
823 2 : defer d.mu.Unlock()
824 2 :
825 2 : for i := range metas {
826 2 : diskFileNums := d.objProvider.GetExternalObjects(metas[i].external.Locator, metas[i].external.ObjName)
827 2 : // We cross-check against fileBackings in the current version because it is
828 2 : // possible that the external object is referenced by an sstable which only
829 2 : // exists in a previous version. In that case, that object could be removed
830 2 : // at any time so we cannot reuse it.
831 2 : for _, n := range diskFileNums {
832 2 : if backing, ok := d.mu.versions.virtualBackings.Get(n); ok {
833 2 : // Protect this backing from being removed from the latest version. We
834 2 : // will unprotect in ingestUnprotectExternalBackings.
835 2 : d.mu.versions.virtualBackings.Protect(n)
836 2 : metas[i].usedExistingBacking = true
837 2 : metas[i].FileBacking = backing
838 2 : break
839 : }
840 : }
841 : }
842 : }
843 :
844 : // ingestUnprotectExternalBackings unprotects the file backings that were reused
845 : // for external objects when the ingestion fails.
846 2 : func (d *DB) ingestUnprotectExternalBackings(lr ingestLoadResult) {
847 2 : d.mu.Lock()
848 2 : defer d.mu.Unlock()
849 2 :
850 2 : for _, meta := range lr.external {
851 2 : if meta.usedExistingBacking {
852 2 : // If the backing is not use anywhere else and the ingest failed (or the
853 2 : // ingested tables were already compacted away), this call will cause in
854 2 : // the next version update to remove the backing.
855 2 : d.mu.versions.virtualBackings.Unprotect(meta.FileBacking.DiskFileNum)
856 2 : }
857 : }
858 : }
859 :
860 : func setSeqNumInMetadata(
861 : m *fileMetadata, seqNum base.SeqNum, cmp Compare, format base.FormatKey,
862 2 : ) error {
863 2 : setSeqFn := func(k base.InternalKey) base.InternalKey {
864 2 : return base.MakeInternalKey(k.UserKey, seqNum, k.Kind())
865 2 : }
866 : // NB: we set the fields directly here, rather than via their Extend*
867 : // methods, as we are updating sequence numbers.
868 2 : if m.HasPointKeys {
869 2 : m.SmallestPointKey = setSeqFn(m.SmallestPointKey)
870 2 : }
871 2 : if m.HasRangeKeys {
872 2 : m.SmallestRangeKey = setSeqFn(m.SmallestRangeKey)
873 2 : }
874 2 : m.Smallest = setSeqFn(m.Smallest)
875 2 : // Only update the seqnum for the largest key if that key is not an
876 2 : // "exclusive sentinel" (i.e. a range deletion sentinel or a range key
877 2 : // boundary), as doing so effectively drops the exclusive sentinel (by
878 2 : // lowering the seqnum from the max value), and extends the bounds of the
879 2 : // table.
880 2 : // NB: as the largest range key is always an exclusive sentinel, it is never
881 2 : // updated.
882 2 : if m.HasPointKeys && !m.LargestPointKey.IsExclusiveSentinel() {
883 2 : m.LargestPointKey = setSeqFn(m.LargestPointKey)
884 2 : }
885 2 : if !m.Largest.IsExclusiveSentinel() {
886 2 : m.Largest = setSeqFn(m.Largest)
887 2 : }
888 : // Setting smallestSeqNum == largestSeqNum triggers the setting of
889 : // Properties.GlobalSeqNum when an sstable is loaded.
890 2 : m.SmallestSeqNum = seqNum
891 2 : m.LargestSeqNum = seqNum
892 2 : m.LargestSeqNumAbsolute = seqNum
893 2 : // Ensure the new bounds are consistent.
894 2 : if err := m.Validate(cmp, format); err != nil {
895 0 : return err
896 0 : }
897 2 : return nil
898 : }
899 :
900 : func ingestUpdateSeqNum(
901 : cmp Compare, format base.FormatKey, seqNum base.SeqNum, loadResult ingestLoadResult,
902 2 : ) error {
903 2 : // Shared sstables are required to be sorted by level ascending. We then
904 2 : // iterate the shared sstables in reverse, assigning the lower sequence
905 2 : // numbers to the shared sstables that will be ingested into the lower
906 2 : // (larger numbered) levels first. This ensures sequence number shadowing is
907 2 : // correct.
908 2 : for i := len(loadResult.shared) - 1; i >= 0; i-- {
909 2 : if i-1 >= 0 && loadResult.shared[i-1].shared.Level > loadResult.shared[i].shared.Level {
910 0 : panic(errors.AssertionFailedf("shared files %s, %s out of order", loadResult.shared[i-1], loadResult.shared[i]))
911 : }
912 2 : if err := setSeqNumInMetadata(loadResult.shared[i].fileMetadata, seqNum, cmp, format); err != nil {
913 0 : return err
914 0 : }
915 2 : seqNum++
916 : }
917 2 : for i := range loadResult.external {
918 2 : if err := setSeqNumInMetadata(loadResult.external[i].fileMetadata, seqNum, cmp, format); err != nil {
919 0 : return err
920 0 : }
921 2 : seqNum++
922 : }
923 2 : for i := range loadResult.local {
924 2 : if err := setSeqNumInMetadata(loadResult.local[i].fileMetadata, seqNum, cmp, format); err != nil {
925 0 : return err
926 0 : }
927 2 : seqNum++
928 : }
929 2 : return nil
930 : }
931 :
932 : // ingestTargetLevel returns the target level for a file being ingested.
933 : // If suggestSplit is true, it accounts for ingest-time splitting as part of
934 : // its target level calculation, and if a split candidate is found, that file
935 : // is returned as the splitFile.
936 : func ingestTargetLevel(
937 : ctx context.Context,
938 : cmp base.Compare,
939 : lsmOverlap overlap.WithLSM,
940 : baseLevel int,
941 : compactions map[*compaction]struct{},
942 : meta *fileMetadata,
943 : suggestSplit bool,
944 2 : ) (targetLevel int, splitFile *fileMetadata, err error) {
945 2 : // Find the lowest level which does not have any files which overlap meta. We
946 2 : // search from L0 to L6 looking for whether there are any files in the level
947 2 : // which overlap meta. We want the "lowest" level (where lower means
948 2 : // increasing level number) in order to reduce write amplification.
949 2 : //
950 2 : // There are 2 kinds of overlap we need to check for: file boundary overlap
951 2 : // and data overlap. Data overlap implies file boundary overlap. Note that it
952 2 : // is always possible to ingest into L0.
953 2 : //
954 2 : // To place meta at level i where i > 0:
955 2 : // - there must not be any data overlap with levels <= i, since that will
956 2 : // violate the sequence number invariant.
957 2 : // - no file boundary overlap with level i, since that will violate the
958 2 : // invariant that files do not overlap in levels i > 0.
959 2 : // - if there is only a file overlap at a given level, and no data overlap,
960 2 : // we can still slot a file at that level. We return the fileMetadata with
961 2 : // which we have file boundary overlap (must be only one file, as sstable
962 2 : // bounds are usually tight on user keys) and the caller is expected to split
963 2 : // that sstable into two virtual sstables, allowing this file to go into that
964 2 : // level. Note that if we have file boundary overlap with two files, which
965 2 : // should only happen on rare occasions, we treat it as data overlap and
966 2 : // don't use this optimization.
967 2 : //
968 2 : // The file boundary overlap check is simpler to conceptualize. Consider the
969 2 : // following example, in which the ingested file lies completely before or
970 2 : // after the file being considered.
971 2 : //
972 2 : // |--| |--| ingested file: [a,b] or [f,g]
973 2 : // |-----| existing file: [c,e]
974 2 : // _____________________
975 2 : // a b c d e f g
976 2 : //
977 2 : // In both cases the ingested file can move to considering the next level.
978 2 : //
979 2 : // File boundary overlap does not necessarily imply data overlap. The check
980 2 : // for data overlap is a little more nuanced. Consider the following examples:
981 2 : //
982 2 : // 1. No data overlap:
983 2 : //
984 2 : // |-| |--| ingested file: [cc-d] or [ee-ff]
985 2 : // |*--*--*----*------*| existing file: [a-g], points: [a, b, c, dd, g]
986 2 : // _____________________
987 2 : // a b c d e f g
988 2 : //
989 2 : // In this case the ingested files can "fall through" this level. The checks
990 2 : // continue at the next level.
991 2 : //
992 2 : // 2. Data overlap:
993 2 : //
994 2 : // |--| ingested file: [d-e]
995 2 : // |*--*--*----*------*| existing file: [a-g], points: [a, b, c, dd, g]
996 2 : // _____________________
997 2 : // a b c d e f g
998 2 : //
999 2 : // In this case the file cannot be ingested into this level as the point 'dd'
1000 2 : // is in the way.
1001 2 : //
1002 2 : // It is worth noting that the check for data overlap is only approximate. In
1003 2 : // the previous example, the ingested table [d-e] could contain only the
1004 2 : // points 'd' and 'e', in which case the table would be eligible for
1005 2 : // considering lower levels. However, such a fine-grained check would need to
1006 2 : // be exhaustive (comparing points and ranges in both the ingested existing
1007 2 : // tables) and such a check is prohibitively expensive. Thus Pebble treats any
1008 2 : // existing point that falls within the ingested table bounds as being "data
1009 2 : // overlap".
1010 2 :
1011 2 : if lsmOverlap[0].Result == overlap.Data {
1012 2 : return 0, nil, nil
1013 2 : }
1014 2 : targetLevel = 0
1015 2 : splitFile = nil
1016 2 : for level := baseLevel; level < numLevels; level++ {
1017 2 : var candidateSplitFile *fileMetadata
1018 2 : switch lsmOverlap[level].Result {
1019 2 : case overlap.Data:
1020 2 : // We cannot ingest into or under this level; return the best target level
1021 2 : // so far.
1022 2 : return targetLevel, splitFile, nil
1023 :
1024 2 : case overlap.OnlyBoundary:
1025 2 : if !suggestSplit || lsmOverlap[level].SplitFile == nil {
1026 2 : // We can ingest under this level, but not into this level.
1027 2 : continue
1028 : }
1029 : // We can ingest into this level if we split this file.
1030 2 : candidateSplitFile = lsmOverlap[level].SplitFile
1031 :
1032 2 : case overlap.None:
1033 : // We can ingest into this level.
1034 :
1035 0 : default:
1036 0 : return 0, nil, base.AssertionFailedf("unexpected WithLevel.Result: %v", lsmOverlap[level].Result)
1037 : }
1038 :
1039 : // Check boundary overlap with any ongoing compactions. We consider an
1040 : // overlapping compaction that's writing files to an output level as
1041 : // equivalent to boundary overlap with files in that output level.
1042 : //
1043 : // We cannot check for data overlap with the new SSTs compaction will produce
1044 : // since compaction hasn't been done yet. However, there's no need to check
1045 : // since all keys in them will be from levels in [c.startLevel,
1046 : // c.outputLevel], and all those levels have already had their data overlap
1047 : // tested negative (else we'd have returned earlier).
1048 : //
1049 : // An alternative approach would be to cancel these compactions and proceed
1050 : // with an ingest-time split on this level if necessary. However, compaction
1051 : // cancellation can result in significant wasted effort and is best avoided
1052 : // unless necessary.
1053 2 : overlaps := false
1054 2 : for c := range compactions {
1055 2 : if c.outputLevel == nil || level != c.outputLevel.level {
1056 2 : continue
1057 : }
1058 2 : if cmp(meta.Smallest.UserKey, c.largest.UserKey) <= 0 &&
1059 2 : cmp(meta.Largest.UserKey, c.smallest.UserKey) >= 0 {
1060 2 : overlaps = true
1061 2 : break
1062 : }
1063 : }
1064 2 : if !overlaps {
1065 2 : targetLevel = level
1066 2 : splitFile = candidateSplitFile
1067 2 : }
1068 : }
1069 2 : return targetLevel, splitFile, nil
1070 : }
1071 :
1072 : // Ingest ingests a set of sstables into the DB. Ingestion of the files is
1073 : // atomic and semantically equivalent to creating a single batch containing all
1074 : // of the mutations in the sstables. Ingestion may require the memtable to be
1075 : // flushed. The ingested sstable files are moved into the DB and must reside on
1076 : // the same filesystem as the DB. Sstables can be created for ingestion using
1077 : // sstable.Writer. On success, Ingest removes the input paths.
1078 : //
1079 : // Two types of sstables are accepted for ingestion(s): one is sstables present
1080 : // in the instance's vfs.FS and can be referenced locally. The other is sstables
1081 : // present in remote.Storage, referred to as shared or foreign sstables. These
1082 : // shared sstables can be linked through objstorageprovider.Provider, and do not
1083 : // need to already be present on the local vfs.FS. Foreign sstables must all fit
1084 : // in an excise span, and are destined for a level specified in SharedSSTMeta.
1085 : //
1086 : // All sstables *must* be Sync()'d by the caller after all bytes are written
1087 : // and before its file handle is closed; failure to do so could violate
1088 : // durability or lead to corrupted on-disk state. This method cannot, in a
1089 : // platform-and-FS-agnostic way, ensure that all sstables in the input are
1090 : // properly synced to disk. Opening new file handles and Sync()-ing them
1091 : // does not always guarantee durability; see the discussion here on that:
1092 : // https://github.com/cockroachdb/pebble/pull/835#issuecomment-663075379
1093 : //
1094 : // Ingestion loads each sstable into the lowest level of the LSM which it
1095 : // doesn't overlap (see ingestTargetLevel). If an sstable overlaps a memtable,
1096 : // ingestion forces the memtable to flush, and then waits for the flush to
1097 : // occur. In some cases, such as with no foreign sstables and no excise span,
1098 : // ingestion that gets blocked on a memtable can join the flushable queue and
1099 : // finish even before the memtable has been flushed.
1100 : //
1101 : // The steps for ingestion are:
1102 : //
1103 : // 1. Allocate file numbers for every sstable being ingested.
1104 : // 2. Load the metadata for all sstables being ingested.
1105 : // 3. Sort the sstables by smallest key, verifying non overlap (for local
1106 : // sstables).
1107 : // 4. Hard link (or copy) the local sstables into the DB directory.
1108 : // 5. Allocate a sequence number to use for all of the entries in the
1109 : // local sstables. This is the step where overlap with memtables is
1110 : // determined. If there is overlap, we remember the most recent memtable
1111 : // that overlaps.
1112 : // 6. Update the sequence number in the ingested local sstables. (Remote
1113 : // sstables get fixed sequence numbers that were determined at load time.)
1114 : // 7. Wait for the most recent memtable that overlaps to flush (if any).
1115 : // 8. Add the ingested sstables to the version (DB.ingestApply).
1116 : // 8.1. If an excise span was specified, figure out what sstables in the
1117 : // current version overlap with the excise span, and create new virtual
1118 : // sstables out of those sstables that exclude the excised span (DB.excise).
1119 : // 9. Publish the ingestion sequence number.
1120 : //
1121 : // Note that if the mutable memtable overlaps with ingestion, a flush of the
1122 : // memtable is forced equivalent to DB.Flush. Additionally, subsequent
1123 : // mutations that get sequence numbers larger than the ingestion sequence
1124 : // number get queued up behind the ingestion waiting for it to complete. This
1125 : // can produce a noticeable hiccup in performance. See
1126 : // https://github.com/cockroachdb/pebble/issues/25 for an idea for how to fix
1127 : // this hiccup.
1128 2 : func (d *DB) Ingest(ctx context.Context, paths []string) error {
1129 2 : if err := d.closed.Load(); err != nil {
1130 1 : panic(err)
1131 : }
1132 2 : if d.opts.ReadOnly {
1133 1 : return ErrReadOnly
1134 1 : }
1135 2 : _, err := d.ingest(ctx, paths, nil /* shared */, KeyRange{}, nil /* external */)
1136 2 : return err
1137 : }
1138 :
1139 : // IngestOperationStats provides some information about where in the LSM the
1140 : // bytes were ingested.
1141 : type IngestOperationStats struct {
1142 : // Bytes is the total bytes in the ingested sstables.
1143 : Bytes uint64
1144 : // ApproxIngestedIntoL0Bytes is the approximate number of bytes ingested
1145 : // into L0. This value is approximate when flushable ingests are active and
1146 : // an ingest overlaps an entry in the flushable queue. Currently, this
1147 : // approximation is very rough, only including tables that overlapped the
1148 : // memtable. This estimate may be improved with #2112.
1149 : ApproxIngestedIntoL0Bytes uint64
1150 : // MemtableOverlappingFiles is the count of ingested sstables
1151 : // that overlapped keys in the memtables.
1152 : MemtableOverlappingFiles int
1153 : }
1154 :
1155 : // ExternalFile are external sstables that can be referenced through
1156 : // objprovider and ingested as remote files that will not be refcounted or
1157 : // cleaned up. For use with online restore. Note that the underlying sstable
1158 : // could contain keys outside the [Smallest,Largest) bounds; however Pebble
1159 : // is expected to only read the keys within those bounds.
1160 : type ExternalFile struct {
1161 : // Locator is the shared.Locator that can be used with objProvider to
1162 : // resolve a reference to this external sstable.
1163 : Locator remote.Locator
1164 :
1165 : // ObjName is the unique name of this sstable on Locator.
1166 : ObjName string
1167 :
1168 : // Size of the referenced proportion of the virtualized sstable. An estimate
1169 : // is acceptable in lieu of the backing file size.
1170 : Size uint64
1171 :
1172 : // StartKey and EndKey define the bounds of the sstable; the ingestion
1173 : // of this file will only result in keys within [StartKey, EndKey) if
1174 : // EndKeyIsInclusive is false or [StartKey, EndKey] if it is true.
1175 : // These bounds are loose i.e. it's possible for keys to not span the
1176 : // entirety of this range.
1177 : //
1178 : // StartKey and EndKey user keys must not have suffixes.
1179 : //
1180 : // Multiple ExternalFiles in one ingestion must all have non-overlapping
1181 : // bounds.
1182 : StartKey, EndKey []byte
1183 :
1184 : // EndKeyIsInclusive is true if EndKey should be treated as inclusive.
1185 : EndKeyIsInclusive bool
1186 :
1187 : // HasPointKey and HasRangeKey denote whether this file contains point keys
1188 : // or range keys. If both structs are false, an error is returned during
1189 : // ingestion.
1190 : HasPointKey, HasRangeKey bool
1191 :
1192 : // SyntheticPrefix will prepend this suffix to all keys in the file during
1193 : // iteration. Note that the backing file itself is not modified.
1194 : //
1195 : // SyntheticPrefix must be a prefix of both Bounds.Start and Bounds.End.
1196 : SyntheticPrefix []byte
1197 :
1198 : // SyntheticSuffix will replace the suffix of every key in the file during
1199 : // iteration. Note that the file itself is not modified, rather, every key
1200 : // returned by an iterator will have the synthetic suffix.
1201 : //
1202 : // SyntheticSuffix can only be used under the following conditions:
1203 : // - the synthetic suffix must sort before any non-empty suffixes in the
1204 : // backing sst (the entire sst, not just the part restricted to Bounds).
1205 : // - the backing sst must not contain multiple keys with the same prefix.
1206 : SyntheticSuffix []byte
1207 :
1208 : // Level denotes the level at which this file was present at read time
1209 : // if the external file was returned by a scan of an existing Pebble
1210 : // instance. If Level is 0, this field is ignored.
1211 : Level uint8
1212 : }
1213 :
1214 : // IngestWithStats does the same as Ingest, and additionally returns
1215 : // IngestOperationStats.
1216 1 : func (d *DB) IngestWithStats(ctx context.Context, paths []string) (IngestOperationStats, error) {
1217 1 : if err := d.closed.Load(); err != nil {
1218 0 : panic(err)
1219 : }
1220 1 : if d.opts.ReadOnly {
1221 0 : return IngestOperationStats{}, ErrReadOnly
1222 0 : }
1223 1 : return d.ingest(ctx, paths, nil, KeyRange{}, nil)
1224 : }
1225 :
1226 : // IngestExternalFiles does the same as IngestWithStats, and additionally
1227 : // accepts external files (with locator info that can be resolved using
1228 : // d.opts.SharedStorage). These files must also be non-overlapping with
1229 : // each other, and must be resolvable through d.objProvider.
1230 : func (d *DB) IngestExternalFiles(
1231 : ctx context.Context, external []ExternalFile,
1232 2 : ) (IngestOperationStats, error) {
1233 2 : if err := d.closed.Load(); err != nil {
1234 0 : panic(err)
1235 : }
1236 :
1237 2 : if d.opts.ReadOnly {
1238 0 : return IngestOperationStats{}, ErrReadOnly
1239 0 : }
1240 2 : if d.opts.Experimental.RemoteStorage == nil {
1241 0 : return IngestOperationStats{}, errors.New("pebble: cannot ingest external files without shared storage configured")
1242 0 : }
1243 2 : return d.ingest(ctx, nil, nil, KeyRange{}, external)
1244 : }
1245 :
1246 : // IngestAndExcise does the same as IngestWithStats, and additionally accepts a
1247 : // list of shared files to ingest that can be read from a remote.Storage through
1248 : // a Provider. All the shared files must live within exciseSpan, and any existing
1249 : // keys in exciseSpan are deleted by turning existing sstables into virtual
1250 : // sstables (if not virtual already) and shrinking their spans to exclude
1251 : // exciseSpan. See the comment at Ingest for a more complete picture of the
1252 : // ingestion process.
1253 : //
1254 : // Panics if this DB instance was not instantiated with a remote.Storage and
1255 : // shared sstables are present.
1256 : func (d *DB) IngestAndExcise(
1257 : ctx context.Context,
1258 : paths []string,
1259 : shared []SharedSSTMeta,
1260 : external []ExternalFile,
1261 : exciseSpan KeyRange,
1262 2 : ) (IngestOperationStats, error) {
1263 2 : if err := d.closed.Load(); err != nil {
1264 0 : panic(err)
1265 : }
1266 2 : if d.opts.ReadOnly {
1267 0 : return IngestOperationStats{}, ErrReadOnly
1268 0 : }
1269 2 : if invariants.Enabled {
1270 2 : // Excise is only supported on prefix keys.
1271 2 : if d.opts.Comparer.Split(exciseSpan.Start) != len(exciseSpan.Start) {
1272 0 : panic("IngestAndExcise called with suffixed start key")
1273 : }
1274 2 : if d.opts.Comparer.Split(exciseSpan.End) != len(exciseSpan.End) {
1275 0 : panic("IngestAndExcise called with suffixed end key")
1276 : }
1277 : }
1278 2 : if v := d.FormatMajorVersion(); v < FormatMinForSharedObjects {
1279 0 : return IngestOperationStats{}, errors.Errorf(
1280 0 : "store has format major version %d; IngestAndExcise requires at least %d",
1281 0 : v, FormatMinForSharedObjects,
1282 0 : )
1283 0 : }
1284 2 : return d.ingest(ctx, paths, shared, exciseSpan, external)
1285 : }
1286 :
1287 : // Both DB.mu and commitPipeline.mu must be held while this is called.
1288 : func (d *DB) newIngestedFlushableEntry(
1289 : meta []*fileMetadata, seqNum base.SeqNum, logNum base.DiskFileNum, exciseSpan KeyRange,
1290 2 : ) (*flushableEntry, error) {
1291 2 : // If there's an excise being done atomically with the same ingest, we
1292 2 : // assign the lowest sequence number in the set of sequence numbers for this
1293 2 : // ingestion to the excise. Note that we've already allocated fileCount+1
1294 2 : // sequence numbers in this case.
1295 2 : //
1296 2 : // This mimics the behaviour in the non-flushable ingest case (see the callsite
1297 2 : // for ingestUpdateSeqNum).
1298 2 : fileSeqNumStart := seqNum
1299 2 : if exciseSpan.Valid() {
1300 2 : fileSeqNumStart = seqNum + 1 // the first seqNum is reserved for the excise.
1301 2 : }
1302 : // Update the sequence number for all of the sstables in the
1303 : // metadata. Writing the metadata to the manifest when the
1304 : // version edit is applied is the mechanism that persists the
1305 : // sequence number. The sstables themselves are left unmodified.
1306 : // In this case, a version edit will only be written to the manifest
1307 : // when the flushable is eventually flushed. If Pebble restarts in that
1308 : // time, then we'll lose the ingest sequence number information. But this
1309 : // information will also be reconstructed on node restart.
1310 2 : for i, m := range meta {
1311 2 : if err := setSeqNumInMetadata(m, fileSeqNumStart+base.SeqNum(i), d.cmp, d.opts.Comparer.FormatKey); err != nil {
1312 0 : return nil, err
1313 0 : }
1314 : }
1315 :
1316 2 : f := newIngestedFlushable(meta, d.opts.Comparer, d.newIters, d.tableNewRangeKeyIter, exciseSpan, seqNum)
1317 2 :
1318 2 : // NB: The logNum/seqNum are the WAL number which we're writing this entry
1319 2 : // to and the sequence number within the WAL which we'll write this entry
1320 2 : // to.
1321 2 : entry := d.newFlushableEntry(f, logNum, seqNum)
1322 2 : // The flushable entry starts off with a single reader ref, so increment
1323 2 : // the FileMetadata.Refs.
1324 2 : for _, file := range f.files {
1325 2 : file.FileBacking.Ref()
1326 2 : }
1327 2 : entry.unrefFiles = func() []*fileBacking {
1328 2 : var obsolete []*fileBacking
1329 2 : for _, file := range f.files {
1330 2 : if file.FileBacking.Unref() == 0 {
1331 2 : obsolete = append(obsolete, file.FileMetadata.FileBacking)
1332 2 : }
1333 : }
1334 2 : return obsolete
1335 : }
1336 :
1337 2 : entry.flushForced = true
1338 2 : entry.releaseMemAccounting = func() {}
1339 2 : return entry, nil
1340 : }
1341 :
1342 : // Both DB.mu and commitPipeline.mu must be held while this is called. Since
1343 : // we're holding both locks, the order in which we rotate the memtable or
1344 : // recycle the WAL in this function is irrelevant as long as the correct log
1345 : // numbers are assigned to the appropriate flushable.
1346 : func (d *DB) handleIngestAsFlushable(
1347 : meta []*fileMetadata, seqNum base.SeqNum, exciseSpan KeyRange,
1348 2 : ) error {
1349 2 : b := d.NewBatch()
1350 2 : if exciseSpan.Valid() {
1351 2 : b.excise(exciseSpan.Start, exciseSpan.End)
1352 2 : }
1353 2 : for _, m := range meta {
1354 2 : b.ingestSST(m.FileNum)
1355 2 : }
1356 2 : b.setSeqNum(seqNum)
1357 2 :
1358 2 : // If the WAL is disabled, then the logNum used to create the flushable
1359 2 : // entry doesn't matter. We just use the logNum assigned to the current
1360 2 : // mutable memtable. If the WAL is enabled, then this logNum will be
1361 2 : // overwritten by the logNum of the log which will contain the log entry
1362 2 : // for the ingestedFlushable.
1363 2 : logNum := d.mu.mem.queue[len(d.mu.mem.queue)-1].logNum
1364 2 : if !d.opts.DisableWAL {
1365 2 : // We create a new WAL for the flushable instead of reusing the end of
1366 2 : // the previous WAL. This simplifies the increment of the minimum
1367 2 : // unflushed log number, and also simplifies WAL replay.
1368 2 : var prevLogSize uint64
1369 2 : logNum, prevLogSize = d.rotateWAL()
1370 2 : // As the rotator of the WAL, we're responsible for updating the
1371 2 : // previous flushable queue tail's log size.
1372 2 : d.mu.mem.queue[len(d.mu.mem.queue)-1].logSize = prevLogSize
1373 2 :
1374 2 : d.mu.Unlock()
1375 2 : err := d.commit.directWrite(b)
1376 2 : if err != nil {
1377 0 : d.opts.Logger.Fatalf("%v", err)
1378 0 : }
1379 2 : d.mu.Lock()
1380 : }
1381 :
1382 : // The excise span is going to outlive this ingestion call. Copy it.
1383 2 : exciseSpan = KeyRange{
1384 2 : Start: slices.Clone(exciseSpan.Start),
1385 2 : End: slices.Clone(exciseSpan.End),
1386 2 : }
1387 2 : entry, err := d.newIngestedFlushableEntry(meta, seqNum, logNum, exciseSpan)
1388 2 : if err != nil {
1389 0 : return err
1390 0 : }
1391 2 : nextSeqNum := seqNum + base.SeqNum(b.Count())
1392 2 :
1393 2 : // Set newLogNum to the logNum of the previous flushable. This value is
1394 2 : // irrelevant if the WAL is disabled. If the WAL is enabled, then we set
1395 2 : // the appropriate value below.
1396 2 : newLogNum := d.mu.mem.queue[len(d.mu.mem.queue)-1].logNum
1397 2 : if !d.opts.DisableWAL {
1398 2 : // newLogNum will be the WAL num of the next mutable memtable which
1399 2 : // comes after the ingestedFlushable in the flushable queue. The mutable
1400 2 : // memtable will be created below.
1401 2 : //
1402 2 : // The prevLogSize returned by rotateWAL is the WAL to which the
1403 2 : // flushable ingest keys were appended. This intermediary WAL is only
1404 2 : // used to record the flushable ingest and nothing else.
1405 2 : newLogNum, entry.logSize = d.rotateWAL()
1406 2 : }
1407 :
1408 2 : d.mu.versions.metrics.Ingest.Count++
1409 2 : currMem := d.mu.mem.mutable
1410 2 : // NB: Placing ingested sstables above the current memtables
1411 2 : // requires rotating of the existing memtables/WAL. There is
1412 2 : // some concern of churning through tiny memtables due to
1413 2 : // ingested sstables being placed on top of them, but those
1414 2 : // memtables would have to be flushed anyways.
1415 2 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
1416 2 : d.rotateMemtable(newLogNum, nextSeqNum, currMem, 0 /* minSize */)
1417 2 : d.updateReadStateLocked(d.opts.DebugCheck)
1418 2 : // TODO(aaditya): is this necessary? we call this already in rotateMemtable above
1419 2 : d.maybeScheduleFlush()
1420 2 : return nil
1421 : }
1422 :
1423 : // See comment at Ingest() for details on how this works.
1424 : func (d *DB) ingest(
1425 : ctx context.Context,
1426 : paths []string,
1427 : shared []SharedSSTMeta,
1428 : exciseSpan KeyRange,
1429 : external []ExternalFile,
1430 2 : ) (IngestOperationStats, error) {
1431 2 : if len(shared) > 0 && d.opts.Experimental.RemoteStorage == nil {
1432 0 : panic("cannot ingest shared sstables with nil SharedStorage")
1433 : }
1434 2 : if (exciseSpan.Valid() || len(shared) > 0 || len(external) > 0) && d.FormatMajorVersion() < FormatVirtualSSTables {
1435 0 : return IngestOperationStats{}, errors.New("pebble: format major version too old for excise, shared or external sstable ingestion")
1436 0 : }
1437 2 : if len(external) > 0 && d.FormatMajorVersion() < FormatSyntheticPrefixSuffix {
1438 1 : for i := range external {
1439 1 : if len(external[i].SyntheticPrefix) > 0 {
1440 1 : return IngestOperationStats{}, errors.New("pebble: format major version too old for synthetic prefix ingestion")
1441 1 : }
1442 1 : if len(external[i].SyntheticSuffix) > 0 {
1443 1 : return IngestOperationStats{}, errors.New("pebble: format major version too old for synthetic suffix ingestion")
1444 1 : }
1445 : }
1446 : }
1447 : // Allocate file numbers for all of the files being ingested and mark them as
1448 : // pending in order to prevent them from being deleted. Note that this causes
1449 : // the file number ordering to be out of alignment with sequence number
1450 : // ordering. The sorting of L0 tables by sequence number avoids relying on
1451 : // that (busted) invariant.
1452 2 : pendingOutputs := make([]base.FileNum, len(paths)+len(shared)+len(external))
1453 2 : for i := 0; i < len(paths)+len(shared)+len(external); i++ {
1454 2 : pendingOutputs[i] = d.mu.versions.getNextFileNum()
1455 2 : }
1456 :
1457 2 : jobID := d.newJobID()
1458 2 :
1459 2 : // Load the metadata for all the files being ingested. This step detects
1460 2 : // and elides empty sstables.
1461 2 : loadResult, err := ingestLoad(ctx, d.opts, d.FormatMajorVersion(), paths, shared, external, d.cacheID, pendingOutputs)
1462 2 : if err != nil {
1463 1 : return IngestOperationStats{}, err
1464 1 : }
1465 :
1466 2 : if loadResult.fileCount() == 0 {
1467 2 : // All of the sstables to be ingested were empty. Nothing to do.
1468 2 : return IngestOperationStats{}, nil
1469 2 : }
1470 :
1471 : // Verify the sstables do not overlap.
1472 2 : if err := ingestSortAndVerify(d.cmp, loadResult, exciseSpan); err != nil {
1473 2 : return IngestOperationStats{}, err
1474 2 : }
1475 :
1476 : // Hard link the sstables into the DB directory. Since the sstables aren't
1477 : // referenced by a version, they won't be used. If the hard linking fails
1478 : // (e.g. because the files reside on a different filesystem), ingestLinkLocal
1479 : // will fall back to copying, and if that fails we undo our work and return an
1480 : // error.
1481 2 : if err := ingestLinkLocal(ctx, jobID, d.opts, d.objProvider, loadResult.local); err != nil {
1482 0 : return IngestOperationStats{}, err
1483 0 : }
1484 :
1485 2 : err = d.ingestAttachRemote(jobID, loadResult)
1486 2 : defer d.ingestUnprotectExternalBackings(loadResult)
1487 2 : if err != nil {
1488 0 : return IngestOperationStats{}, err
1489 0 : }
1490 :
1491 : // Make the new tables durable. We need to do this at some point before we
1492 : // update the MANIFEST (via logAndApply), otherwise a crash can have the
1493 : // tables referenced in the MANIFEST, but not present in the provider.
1494 2 : if err := d.objProvider.Sync(); err != nil {
1495 1 : return IngestOperationStats{}, err
1496 1 : }
1497 :
1498 : // metaFlushableOverlaps is a map indicating which of the ingested sstables
1499 : // overlap some table in the flushable queue. It's used to approximate
1500 : // ingest-into-L0 stats when using flushable ingests.
1501 2 : metaFlushableOverlaps := make(map[FileNum]bool, loadResult.fileCount())
1502 2 : var mem *flushableEntry
1503 2 : var mut *memTable
1504 2 : // asFlushable indicates whether the sstable was ingested as a flushable.
1505 2 : var asFlushable bool
1506 2 : prepare := func(seqNum base.SeqNum) {
1507 2 : // Note that d.commit.mu is held by commitPipeline when calling prepare.
1508 2 :
1509 2 : // Determine the set of bounds we care about for the purpose of checking
1510 2 : // for overlap among the flushables. If there's an excise span, we need
1511 2 : // to check for overlap with its bounds as well.
1512 2 : overlapBounds := make([]bounded, 0, loadResult.fileCount()+1)
1513 2 : for _, m := range loadResult.local {
1514 2 : overlapBounds = append(overlapBounds, m.fileMetadata)
1515 2 : }
1516 2 : for _, m := range loadResult.shared {
1517 2 : overlapBounds = append(overlapBounds, m.fileMetadata)
1518 2 : }
1519 2 : for _, m := range loadResult.external {
1520 2 : overlapBounds = append(overlapBounds, m.fileMetadata)
1521 2 : }
1522 2 : if exciseSpan.Valid() {
1523 2 : overlapBounds = append(overlapBounds, &exciseSpan)
1524 2 : }
1525 :
1526 2 : d.mu.Lock()
1527 2 : defer d.mu.Unlock()
1528 2 :
1529 2 : // Check if any of the currently-open EventuallyFileOnlySnapshots overlap
1530 2 : // in key ranges with the excise span. If so, we need to check for memtable
1531 2 : // overlaps with all bounds of that EventuallyFileOnlySnapshot in addition
1532 2 : // to the ingestion's own bounds too.
1533 2 :
1534 2 : if exciseSpan.Valid() {
1535 2 : for s := d.mu.snapshots.root.next; s != &d.mu.snapshots.root; s = s.next {
1536 2 : if s.efos == nil {
1537 0 : continue
1538 : }
1539 2 : if base.Visible(seqNum, s.efos.seqNum, base.SeqNumMax) {
1540 0 : // We only worry about snapshots older than the excise. Any snapshots
1541 0 : // created after the excise should see the excised view of the LSM
1542 0 : // anyway.
1543 0 : //
1544 0 : // Since we delay publishing the excise seqnum as visible until after
1545 0 : // the apply step, this case will never be hit in practice until we
1546 0 : // make excises flushable ingests.
1547 0 : continue
1548 : }
1549 2 : if invariants.Enabled {
1550 2 : if s.efos.hasTransitioned() {
1551 0 : panic("unexpected transitioned EFOS in snapshots list")
1552 : }
1553 : }
1554 2 : for i := range s.efos.protectedRanges {
1555 2 : if !s.efos.protectedRanges[i].OverlapsKeyRange(d.cmp, exciseSpan) {
1556 2 : continue
1557 : }
1558 : // Our excise conflicts with this EFOS. We need to add its protected
1559 : // ranges to our overlapBounds. Grow overlapBounds in one allocation
1560 : // if necesary.
1561 2 : prs := s.efos.protectedRanges
1562 2 : if cap(overlapBounds) < len(overlapBounds)+len(prs) {
1563 2 : oldOverlapBounds := overlapBounds
1564 2 : overlapBounds = make([]bounded, len(oldOverlapBounds), len(oldOverlapBounds)+len(prs))
1565 2 : copy(overlapBounds, oldOverlapBounds)
1566 2 : }
1567 2 : for i := range prs {
1568 2 : overlapBounds = append(overlapBounds, &prs[i])
1569 2 : }
1570 2 : break
1571 : }
1572 : }
1573 : }
1574 :
1575 : // Check to see if any files overlap with any of the memtables. The queue
1576 : // is ordered from oldest to newest with the mutable memtable being the
1577 : // last element in the slice. We want to wait for the newest table that
1578 : // overlaps.
1579 :
1580 2 : for i := len(d.mu.mem.queue) - 1; i >= 0; i-- {
1581 2 : m := d.mu.mem.queue[i]
1582 2 : m.computePossibleOverlaps(func(b bounded) shouldContinue {
1583 2 : // If this is the first table to overlap a flushable, save
1584 2 : // the flushable. This ingest must be ingested or flushed
1585 2 : // after it.
1586 2 : if mem == nil {
1587 2 : mem = m
1588 2 : }
1589 :
1590 2 : switch v := b.(type) {
1591 2 : case *fileMetadata:
1592 2 : // NB: False positives are possible if `m` is a flushable
1593 2 : // ingest that overlaps the file `v` in bounds but doesn't
1594 2 : // contain overlapping data. This is considered acceptable
1595 2 : // because it's rare (in CockroachDB a bound overlap likely
1596 2 : // indicates a data overlap), and blocking the commit
1597 2 : // pipeline while we perform I/O to check for overlap may be
1598 2 : // more disruptive than enqueueing this ingestion on the
1599 2 : // flushable queue and switching to a new memtable.
1600 2 : metaFlushableOverlaps[v.FileNum] = true
1601 2 : case *KeyRange:
1602 : // An excise span or an EventuallyFileOnlySnapshot protected range;
1603 : // not a file.
1604 0 : default:
1605 0 : panic("unreachable")
1606 : }
1607 2 : return continueIteration
1608 : }, overlapBounds...)
1609 : }
1610 :
1611 2 : if mem == nil {
1612 2 : // No overlap with any of the queued flushables, so no need to queue
1613 2 : // after them.
1614 2 :
1615 2 : // New writes with higher sequence numbers may be concurrently
1616 2 : // committed. We must ensure they don't flush before this ingest
1617 2 : // completes. To do that, we ref the mutable memtable as a writer,
1618 2 : // preventing its flushing (and the flushing of all subsequent
1619 2 : // flushables in the queue). Once we've acquired the manifest lock
1620 2 : // to add the ingested sstables to the LSM, we can unref as we're
1621 2 : // guaranteed that the flush won't edit the LSM before this ingest.
1622 2 : mut = d.mu.mem.mutable
1623 2 : mut.writerRef()
1624 2 : return
1625 2 : }
1626 :
1627 : // The ingestion overlaps with some entry in the flushable queue. If the
1628 : // pre-conditions are met below, we can treat this ingestion as a flushable
1629 : // ingest, otherwise we wait on the memtable flush before ingestion.
1630 : //
1631 : // TODO(aaditya): We should make flushableIngest compatible with remote
1632 : // files.
1633 2 : hasRemoteFiles := len(shared) > 0 || len(external) > 0
1634 2 : canIngestFlushable := d.FormatMajorVersion() >= FormatFlushableIngest &&
1635 2 : (len(d.mu.mem.queue) < d.opts.MemTableStopWritesThreshold) &&
1636 2 : !d.opts.Experimental.DisableIngestAsFlushable() && !hasRemoteFiles &&
1637 2 : (!exciseSpan.Valid() || d.FormatMajorVersion() >= FormatFlushableIngestExcises)
1638 2 :
1639 2 : if !canIngestFlushable {
1640 2 : // We're not able to ingest as a flushable,
1641 2 : // so we must synchronously flush.
1642 2 : //
1643 2 : // TODO(bilal): Currently, if any of the files being ingested are shared,
1644 2 : // we cannot use flushable ingests and need
1645 2 : // to wait synchronously.
1646 2 : if mem.flushable == d.mu.mem.mutable {
1647 2 : err = d.makeRoomForWrite(nil)
1648 2 : }
1649 : // New writes with higher sequence numbers may be concurrently
1650 : // committed. We must ensure they don't flush before this ingest
1651 : // completes. To do that, we ref the mutable memtable as a writer,
1652 : // preventing its flushing (and the flushing of all subsequent
1653 : // flushables in the queue). Once we've acquired the manifest lock
1654 : // to add the ingested sstables to the LSM, we can unref as we're
1655 : // guaranteed that the flush won't edit the LSM before this ingest.
1656 2 : mut = d.mu.mem.mutable
1657 2 : mut.writerRef()
1658 2 : mem.flushForced = true
1659 2 : d.maybeScheduleFlush()
1660 2 : return
1661 : }
1662 : // Since there aren't too many memtables already queued up, we can
1663 : // slide the ingested sstables on top of the existing memtables.
1664 2 : asFlushable = true
1665 2 : fileMetas := make([]*fileMetadata, len(loadResult.local))
1666 2 : for i := range fileMetas {
1667 2 : fileMetas[i] = loadResult.local[i].fileMetadata
1668 2 : }
1669 2 : err = d.handleIngestAsFlushable(fileMetas, seqNum, exciseSpan)
1670 : }
1671 :
1672 2 : var ve *versionEdit
1673 2 : apply := func(seqNum base.SeqNum) {
1674 2 : if err != nil || asFlushable {
1675 2 : // An error occurred during prepare.
1676 2 : if mut != nil {
1677 0 : if mut.writerUnref() {
1678 0 : d.mu.Lock()
1679 0 : d.maybeScheduleFlush()
1680 0 : d.mu.Unlock()
1681 0 : }
1682 : }
1683 2 : return
1684 : }
1685 :
1686 : // If there's an excise being done atomically with the same ingest, we
1687 : // assign the lowest sequence number in the set of sequence numbers for this
1688 : // ingestion to the excise. Note that we've already allocated fileCount+1
1689 : // sequence numbers in this case.
1690 2 : if exciseSpan.Valid() {
1691 2 : seqNum++ // the first seqNum is reserved for the excise.
1692 2 : }
1693 : // Update the sequence numbers for all ingested sstables'
1694 : // metadata. When the version edit is applied, the metadata is
1695 : // written to the manifest, persisting the sequence number.
1696 : // The sstables themselves are left unmodified.
1697 2 : if err = ingestUpdateSeqNum(
1698 2 : d.cmp, d.opts.Comparer.FormatKey, seqNum, loadResult,
1699 2 : ); err != nil {
1700 0 : if mut != nil {
1701 0 : if mut.writerUnref() {
1702 0 : d.mu.Lock()
1703 0 : d.maybeScheduleFlush()
1704 0 : d.mu.Unlock()
1705 0 : }
1706 : }
1707 0 : return
1708 : }
1709 :
1710 : // If we overlapped with a memtable in prepare wait for the flush to
1711 : // finish.
1712 2 : if mem != nil {
1713 2 : <-mem.flushed
1714 2 : }
1715 :
1716 : // Assign the sstables to the correct level in the LSM and apply the
1717 : // version edit.
1718 2 : ve, err = d.ingestApply(ctx, jobID, loadResult, mut, exciseSpan, seqNum)
1719 : }
1720 :
1721 : // Only one ingest can occur at a time because if not, one would block waiting
1722 : // for the other to finish applying. This blocking would happen while holding
1723 : // the commit mutex which would prevent unrelated batches from writing their
1724 : // changes to the WAL and memtable. This will cause a bigger commit hiccup
1725 : // during ingestion.
1726 2 : seqNumCount := loadResult.fileCount()
1727 2 : if exciseSpan.Valid() {
1728 2 : seqNumCount++
1729 2 : }
1730 2 : d.commit.ingestSem <- struct{}{}
1731 2 : d.commit.AllocateSeqNum(seqNumCount, prepare, apply)
1732 2 : <-d.commit.ingestSem
1733 2 :
1734 2 : if err != nil {
1735 1 : if err2 := ingestCleanup(d.objProvider, loadResult.local); err2 != nil {
1736 0 : d.opts.Logger.Errorf("ingest cleanup failed: %v", err2)
1737 0 : }
1738 2 : } else {
1739 2 : // Since we either created a hard link to the ingesting files, or copied
1740 2 : // them over, it is safe to remove the originals paths.
1741 2 : for i := range loadResult.local {
1742 2 : path := loadResult.local[i].path
1743 2 : if err2 := d.opts.FS.Remove(path); err2 != nil {
1744 1 : d.opts.Logger.Errorf("ingest failed to remove original file: %s", err2)
1745 1 : }
1746 : }
1747 : }
1748 :
1749 2 : info := TableIngestInfo{
1750 2 : JobID: int(jobID),
1751 2 : Err: err,
1752 2 : flushable: asFlushable,
1753 2 : }
1754 2 : if len(loadResult.local) > 0 {
1755 2 : info.GlobalSeqNum = loadResult.local[0].SmallestSeqNum
1756 2 : } else if len(loadResult.shared) > 0 {
1757 2 : info.GlobalSeqNum = loadResult.shared[0].SmallestSeqNum
1758 2 : } else {
1759 2 : info.GlobalSeqNum = loadResult.external[0].SmallestSeqNum
1760 2 : }
1761 2 : var stats IngestOperationStats
1762 2 : if ve != nil {
1763 2 : info.Tables = make([]struct {
1764 2 : TableInfo
1765 2 : Level int
1766 2 : }, len(ve.NewFiles))
1767 2 : for i := range ve.NewFiles {
1768 2 : e := &ve.NewFiles[i]
1769 2 : info.Tables[i].Level = e.Level
1770 2 : info.Tables[i].TableInfo = e.Meta.TableInfo()
1771 2 : stats.Bytes += e.Meta.Size
1772 2 : if e.Level == 0 {
1773 2 : stats.ApproxIngestedIntoL0Bytes += e.Meta.Size
1774 2 : }
1775 2 : if metaFlushableOverlaps[e.Meta.FileNum] {
1776 2 : stats.MemtableOverlappingFiles++
1777 2 : }
1778 : }
1779 2 : } else if asFlushable {
1780 2 : // NB: If asFlushable == true, there are no shared sstables.
1781 2 : info.Tables = make([]struct {
1782 2 : TableInfo
1783 2 : Level int
1784 2 : }, len(loadResult.local))
1785 2 : for i, f := range loadResult.local {
1786 2 : info.Tables[i].Level = -1
1787 2 : info.Tables[i].TableInfo = f.TableInfo()
1788 2 : stats.Bytes += f.Size
1789 2 : // We don't have exact stats on which files will be ingested into
1790 2 : // L0, because actual ingestion into the LSM has been deferred until
1791 2 : // flush time. Instead, we infer based on memtable overlap.
1792 2 : //
1793 2 : // TODO(jackson): If we optimistically compute data overlap (#2112)
1794 2 : // before entering the commit pipeline, we can use that overlap to
1795 2 : // improve our approximation by incorporating overlap with L0, not
1796 2 : // just memtables.
1797 2 : if metaFlushableOverlaps[f.FileNum] {
1798 2 : stats.ApproxIngestedIntoL0Bytes += f.Size
1799 2 : stats.MemtableOverlappingFiles++
1800 2 : }
1801 : }
1802 : }
1803 2 : d.opts.EventListener.TableIngested(info)
1804 2 :
1805 2 : return stats, err
1806 : }
1807 :
1808 : // excise updates ve to include a replacement of the file m with new virtual
1809 : // sstables that exclude exciseSpan, returning a slice of newly-created files if
1810 : // any. If the entirety of m is deleted by exciseSpan, no new sstables are added
1811 : // and m is deleted. Note that ve is updated in-place.
1812 : //
1813 : // This method is agnostic to whether d.mu is held or not. Some cases call it with
1814 : // the db mutex held (eg. ingest-time excises), while in the case of compactions
1815 : // the mutex is not held.
1816 : func (d *DB) excise(
1817 : ctx context.Context, exciseSpan base.UserKeyBounds, m *fileMetadata, ve *versionEdit, level int,
1818 2 : ) ([]manifest.NewFileEntry, error) {
1819 2 : numCreatedFiles := 0
1820 2 : // Check if there's actually an overlap between m and exciseSpan.
1821 2 : mBounds := base.UserKeyBoundsFromInternal(m.Smallest, m.Largest)
1822 2 : if !exciseSpan.Overlaps(d.cmp, &mBounds) {
1823 2 : return nil, nil
1824 2 : }
1825 2 : ve.DeletedFiles[deletedFileEntry{
1826 2 : Level: level,
1827 2 : FileNum: m.FileNum,
1828 2 : }] = m
1829 2 : // Fast path: m sits entirely within the exciseSpan, so just delete it.
1830 2 : if exciseSpan.ContainsInternalKey(d.cmp, m.Smallest) && exciseSpan.ContainsInternalKey(d.cmp, m.Largest) {
1831 2 : return nil, nil
1832 2 : }
1833 :
1834 2 : var iters iterSet
1835 2 : var itersLoaded bool
1836 2 : defer iters.CloseAll()
1837 2 : loadItersIfNecessary := func() error {
1838 2 : if itersLoaded {
1839 2 : return nil
1840 2 : }
1841 2 : var err error
1842 2 : iters, err = d.newIters(ctx, m, &IterOptions{
1843 2 : Category: categoryIngest,
1844 2 : layer: manifest.Level(level),
1845 2 : }, internalIterOpts{}, iterPointKeys|iterRangeDeletions|iterRangeKeys)
1846 2 : itersLoaded = true
1847 2 : return err
1848 : }
1849 :
1850 2 : needsBacking := false
1851 2 : // Create a file to the left of the excise span, if necessary.
1852 2 : // The bounds of this file will be [m.Smallest, lastKeyBefore(exciseSpan.Start)].
1853 2 : //
1854 2 : // We create bounds that are tight on user keys, and we make the effort to find
1855 2 : // the last key in the original sstable that's smaller than exciseSpan.Start
1856 2 : // even though it requires some sstable reads. We could choose to create
1857 2 : // virtual sstables on loose userKey bounds, in which case we could just set
1858 2 : // leftFile.Largest to an exclusive sentinel at exciseSpan.Start. The biggest
1859 2 : // issue with that approach would be that it'd lead to lots of small virtual
1860 2 : // sstables in the LSM that have no guarantee on containing even a single user
1861 2 : // key within the file bounds. This has the potential to increase both read and
1862 2 : // write-amp as we will be opening up these sstables only to find no relevant
1863 2 : // keys in the read path, and compacting sstables on top of them instead of
1864 2 : // directly into the space occupied by them. We choose to incur the cost of
1865 2 : // calculating tight bounds at this time instead of creating more work in the
1866 2 : // future.
1867 2 : //
1868 2 : // TODO(bilal): Some of this work can happen without grabbing the manifest
1869 2 : // lock; we could grab one currentVersion, release the lock, calculate excised
1870 2 : // files, then grab the lock again and recalculate for just the files that
1871 2 : // have changed since our previous calculation. Do this optimiaztino as part of
1872 2 : // https://github.com/cockroachdb/pebble/issues/2112 .
1873 2 : if d.cmp(m.Smallest.UserKey, exciseSpan.Start) < 0 {
1874 2 : leftFile := &fileMetadata{
1875 2 : Virtual: true,
1876 2 : FileBacking: m.FileBacking,
1877 2 : FileNum: d.mu.versions.getNextFileNum(),
1878 2 : // Note that these are loose bounds for smallest/largest seqnums, but they're
1879 2 : // sufficient for maintaining correctness.
1880 2 : SmallestSeqNum: m.SmallestSeqNum,
1881 2 : LargestSeqNum: m.LargestSeqNum,
1882 2 : LargestSeqNumAbsolute: m.LargestSeqNumAbsolute,
1883 2 : SyntheticPrefixAndSuffix: m.SyntheticPrefixAndSuffix,
1884 2 : }
1885 2 : if m.HasPointKeys && !exciseSpan.ContainsInternalKey(d.cmp, m.SmallestPointKey) {
1886 2 : // This file will probably contain point keys.
1887 2 : if err := loadItersIfNecessary(); err != nil {
1888 0 : return nil, err
1889 0 : }
1890 2 : smallestPointKey := m.SmallestPointKey
1891 2 : if kv := iters.Point().SeekLT(exciseSpan.Start, base.SeekLTFlagsNone); kv != nil {
1892 2 : leftFile.ExtendPointKeyBounds(d.cmp, smallestPointKey, kv.K.Clone())
1893 2 : }
1894 : // Store the min of (exciseSpan.Start, rdel.End) in lastRangeDel. This
1895 : // needs to be a copy if the key is owned by the range del iter.
1896 2 : var lastRangeDel []byte
1897 2 : if rdel, err := iters.RangeDeletion().SeekLT(exciseSpan.Start); err != nil {
1898 0 : return nil, err
1899 2 : } else if rdel != nil {
1900 2 : lastRangeDel = append(lastRangeDel[:0], rdel.End...)
1901 2 : if d.cmp(lastRangeDel, exciseSpan.Start) > 0 {
1902 2 : lastRangeDel = exciseSpan.Start
1903 2 : }
1904 : }
1905 2 : if lastRangeDel != nil {
1906 2 : leftFile.ExtendPointKeyBounds(d.cmp, smallestPointKey, base.MakeExclusiveSentinelKey(InternalKeyKindRangeDelete, lastRangeDel))
1907 2 : }
1908 : }
1909 2 : if m.HasRangeKeys && !exciseSpan.ContainsInternalKey(d.cmp, m.SmallestRangeKey) {
1910 2 : // This file will probably contain range keys.
1911 2 : if err := loadItersIfNecessary(); err != nil {
1912 0 : return nil, err
1913 0 : }
1914 2 : smallestRangeKey := m.SmallestRangeKey
1915 2 : // Store the min of (exciseSpan.Start, rkey.End) in lastRangeKey. This
1916 2 : // needs to be a copy if the key is owned by the range key iter.
1917 2 : var lastRangeKey []byte
1918 2 : var lastRangeKeyKind InternalKeyKind
1919 2 : if rkey, err := iters.RangeKey().SeekLT(exciseSpan.Start); err != nil {
1920 0 : return nil, err
1921 2 : } else if rkey != nil {
1922 2 : lastRangeKey = append(lastRangeKey[:0], rkey.End...)
1923 2 : if d.cmp(lastRangeKey, exciseSpan.Start) > 0 {
1924 2 : lastRangeKey = exciseSpan.Start
1925 2 : }
1926 2 : lastRangeKeyKind = rkey.Keys[0].Kind()
1927 : }
1928 2 : if lastRangeKey != nil {
1929 2 : leftFile.ExtendRangeKeyBounds(d.cmp, smallestRangeKey, base.MakeExclusiveSentinelKey(lastRangeKeyKind, lastRangeKey))
1930 2 : }
1931 : }
1932 2 : if leftFile.HasRangeKeys || leftFile.HasPointKeys {
1933 2 : var err error
1934 2 : leftFile.Size, err = d.fileCache.estimateSize(m, leftFile.Smallest.UserKey, leftFile.Largest.UserKey)
1935 2 : if err != nil {
1936 0 : return nil, err
1937 0 : }
1938 2 : if leftFile.Size == 0 {
1939 2 : // On occasion, estimateSize gives us a low estimate, i.e. a 0 file size,
1940 2 : // such as if the excised file only has range keys/dels and no point
1941 2 : // keys. This can cause panics in places where we divide by file sizes.
1942 2 : // Correct for it here.
1943 2 : leftFile.Size = 1
1944 2 : }
1945 2 : if err := leftFile.Validate(d.cmp, d.opts.Comparer.FormatKey); err != nil {
1946 0 : return nil, err
1947 0 : }
1948 2 : leftFile.ValidateVirtual(m)
1949 2 : ve.NewFiles = append(ve.NewFiles, newFileEntry{Level: level, Meta: leftFile})
1950 2 : needsBacking = true
1951 2 : numCreatedFiles++
1952 : }
1953 : }
1954 : // Create a file to the right, if necessary.
1955 2 : if exciseSpan.ContainsInternalKey(d.cmp, m.Largest) {
1956 2 : // No key exists to the right of the excise span in this file.
1957 2 : if needsBacking && !m.Virtual {
1958 2 : // If m is virtual, then its file backing is already known to the manifest.
1959 2 : // We don't need to create another file backing. Note that there must be
1960 2 : // only one CreatedBackingTables entry per backing sstable. This is
1961 2 : // indicated by the VersionEdit.CreatedBackingTables invariant.
1962 2 : ve.CreatedBackingTables = append(ve.CreatedBackingTables, m.FileBacking)
1963 2 : }
1964 2 : return ve.NewFiles[len(ve.NewFiles)-numCreatedFiles:], nil
1965 : }
1966 : // Create a new file, rightFile, between [firstKeyAfter(exciseSpan.End), m.Largest].
1967 : //
1968 : // See comment before the definition of leftFile for the motivation behind
1969 : // calculating tight user-key bounds.
1970 2 : rightFile := &fileMetadata{
1971 2 : Virtual: true,
1972 2 : FileBacking: m.FileBacking,
1973 2 : FileNum: d.mu.versions.getNextFileNum(),
1974 2 : // Note that these are loose bounds for smallest/largest seqnums, but they're
1975 2 : // sufficient for maintaining correctness.
1976 2 : SmallestSeqNum: m.SmallestSeqNum,
1977 2 : LargestSeqNum: m.LargestSeqNum,
1978 2 : LargestSeqNumAbsolute: m.LargestSeqNumAbsolute,
1979 2 : SyntheticPrefixAndSuffix: m.SyntheticPrefixAndSuffix,
1980 2 : }
1981 2 : if m.HasPointKeys && !exciseSpan.ContainsInternalKey(d.cmp, m.LargestPointKey) {
1982 2 : // This file will probably contain point keys
1983 2 : if err := loadItersIfNecessary(); err != nil {
1984 0 : return nil, err
1985 0 : }
1986 2 : largestPointKey := m.LargestPointKey
1987 2 : if kv := iters.Point().SeekGE(exciseSpan.End.Key, base.SeekGEFlagsNone); kv != nil {
1988 2 : if exciseSpan.End.Kind == base.Inclusive && d.equal(exciseSpan.End.Key, kv.K.UserKey) {
1989 0 : return nil, base.AssertionFailedf("cannot excise with an inclusive end key and data overlap at end key")
1990 0 : }
1991 2 : rightFile.ExtendPointKeyBounds(d.cmp, kv.K.Clone(), largestPointKey)
1992 : }
1993 : // Store the max of (exciseSpan.End, rdel.Start) in firstRangeDel. This
1994 : // needs to be a copy if the key is owned by the range del iter.
1995 2 : var firstRangeDel []byte
1996 2 : rdel, err := iters.RangeDeletion().SeekGE(exciseSpan.End.Key)
1997 2 : if err != nil {
1998 0 : return nil, err
1999 2 : } else if rdel != nil {
2000 2 : firstRangeDel = append(firstRangeDel[:0], rdel.Start...)
2001 2 : if d.cmp(firstRangeDel, exciseSpan.End.Key) < 0 {
2002 2 : // NB: This can only be done if the end bound is exclusive.
2003 2 : if exciseSpan.End.Kind != base.Exclusive {
2004 0 : return nil, base.AssertionFailedf("cannot truncate rangedel during excise with an inclusive upper bound")
2005 0 : }
2006 2 : firstRangeDel = exciseSpan.End.Key
2007 : }
2008 : }
2009 2 : if firstRangeDel != nil {
2010 2 : smallestPointKey := rdel.SmallestKey()
2011 2 : smallestPointKey.UserKey = firstRangeDel
2012 2 : rightFile.ExtendPointKeyBounds(d.cmp, smallestPointKey, largestPointKey)
2013 2 : }
2014 : }
2015 2 : if m.HasRangeKeys && !exciseSpan.ContainsInternalKey(d.cmp, m.LargestRangeKey) {
2016 2 : // This file will probably contain range keys.
2017 2 : if err := loadItersIfNecessary(); err != nil {
2018 0 : return nil, err
2019 0 : }
2020 2 : largestRangeKey := m.LargestRangeKey
2021 2 : // Store the max of (exciseSpan.End, rkey.Start) in firstRangeKey. This
2022 2 : // needs to be a copy if the key is owned by the range key iter.
2023 2 : var firstRangeKey []byte
2024 2 : rkey, err := iters.RangeKey().SeekGE(exciseSpan.End.Key)
2025 2 : if err != nil {
2026 0 : return nil, err
2027 2 : } else if rkey != nil {
2028 2 : firstRangeKey = append(firstRangeKey[:0], rkey.Start...)
2029 2 : if d.cmp(firstRangeKey, exciseSpan.End.Key) < 0 {
2030 2 : if exciseSpan.End.Kind != base.Exclusive {
2031 0 : return nil, base.AssertionFailedf("cannot truncate range key during excise with an inclusive upper bound")
2032 0 : }
2033 2 : firstRangeKey = exciseSpan.End.Key
2034 : }
2035 : }
2036 2 : if firstRangeKey != nil {
2037 2 : smallestRangeKey := rkey.SmallestKey()
2038 2 : smallestRangeKey.UserKey = firstRangeKey
2039 2 : // We call ExtendRangeKeyBounds so any internal boundType fields are
2040 2 : // set correctly. Note that this is mildly wasteful as we'll be comparing
2041 2 : // rightFile.{Smallest,Largest}RangeKey with themselves, which can be
2042 2 : // avoided if we exported ExtendOverallKeyBounds or so.
2043 2 : rightFile.ExtendRangeKeyBounds(d.cmp, smallestRangeKey, largestRangeKey)
2044 2 : }
2045 : }
2046 2 : if rightFile.HasRangeKeys || rightFile.HasPointKeys {
2047 2 : var err error
2048 2 : rightFile.Size, err = d.fileCache.estimateSize(m, rightFile.Smallest.UserKey, rightFile.Largest.UserKey)
2049 2 : if err != nil {
2050 0 : return nil, err
2051 0 : }
2052 2 : if rightFile.Size == 0 {
2053 2 : // On occasion, estimateSize gives us a low estimate, i.e. a 0 file size,
2054 2 : // such as if the excised file only has range keys/dels and no point keys.
2055 2 : // This can cause panics in places where we divide by file sizes. Correct
2056 2 : // for it here.
2057 2 : rightFile.Size = 1
2058 2 : }
2059 2 : if err := rightFile.Validate(d.cmp, d.opts.Comparer.FormatKey); err != nil {
2060 0 : return nil, err
2061 0 : }
2062 2 : rightFile.ValidateVirtual(m)
2063 2 : ve.NewFiles = append(ve.NewFiles, newFileEntry{Level: level, Meta: rightFile})
2064 2 : needsBacking = true
2065 2 : numCreatedFiles++
2066 : }
2067 :
2068 2 : if needsBacking && !m.Virtual {
2069 2 : // If m is virtual, then its file backing is already known to the manifest.
2070 2 : // We don't need to create another file backing. Note that there must be
2071 2 : // only one CreatedBackingTables entry per backing sstable. This is
2072 2 : // indicated by the VersionEdit.CreatedBackingTables invariant.
2073 2 : ve.CreatedBackingTables = append(ve.CreatedBackingTables, m.FileBacking)
2074 2 : }
2075 :
2076 2 : return ve.NewFiles[len(ve.NewFiles)-numCreatedFiles:], nil
2077 : }
2078 :
2079 : type ingestSplitFile struct {
2080 : // ingestFile is the file being ingested.
2081 : ingestFile *fileMetadata
2082 : // splitFile is the file that needs to be split to allow ingestFile to slot
2083 : // into `level` level.
2084 : splitFile *fileMetadata
2085 : // The level where ingestFile will go (and where splitFile already is).
2086 : level int
2087 : }
2088 :
2089 : // ingestSplit splits files specified in `files` and updates ve in-place to
2090 : // account for existing files getting split into two virtual sstables. The map
2091 : // `replacedFiles` contains an in-progress map of all files that have been
2092 : // replaced with new virtual sstables in this version edit so far, which is also
2093 : // updated in-place.
2094 : //
2095 : // d.mu as well as the manifest lock must be held when calling this method.
2096 : func (d *DB) ingestSplit(
2097 : ctx context.Context,
2098 : ve *versionEdit,
2099 : updateMetrics func(*fileMetadata, int, []newFileEntry),
2100 : files []ingestSplitFile,
2101 : replacedFiles map[base.FileNum][]newFileEntry,
2102 2 : ) error {
2103 2 : for _, s := range files {
2104 2 : ingestFileBounds := s.ingestFile.UserKeyBounds()
2105 2 : // replacedFiles can be thought of as a tree, where we start iterating with
2106 2 : // s.splitFile and run its fileNum through replacedFiles, then find which of
2107 2 : // the replaced files overlaps with s.ingestFile, which becomes the new
2108 2 : // splitFile, then we check splitFile's replacements in replacedFiles again
2109 2 : // for overlap with s.ingestFile, and so on until we either can't find the
2110 2 : // current splitFile in replacedFiles (i.e. that's the file that now needs to
2111 2 : // be split), or we don't find a file that overlaps with s.ingestFile, which
2112 2 : // means a prior ingest split already produced enough room for s.ingestFile
2113 2 : // to go into this level without necessitating another ingest split.
2114 2 : splitFile := s.splitFile
2115 2 : for splitFile != nil {
2116 2 : replaced, ok := replacedFiles[splitFile.FileNum]
2117 2 : if !ok {
2118 2 : break
2119 : }
2120 2 : updatedSplitFile := false
2121 2 : for i := range replaced {
2122 2 : if replaced[i].Meta.Overlaps(d.cmp, &ingestFileBounds) {
2123 2 : if updatedSplitFile {
2124 0 : // This should never happen because the earlier ingestTargetLevel
2125 0 : // function only finds split file candidates that are guaranteed to
2126 0 : // have no data overlap, only boundary overlap. See the comments
2127 0 : // in that method to see the definitions of data vs boundary
2128 0 : // overlap. That, plus the fact that files in `replaced` are
2129 0 : // guaranteed to have file bounds that are tight on user keys
2130 0 : // (as that's what `d.excise` produces), means that the only case
2131 0 : // where we overlap with two or more files in `replaced` is if we
2132 0 : // actually had data overlap all along, or if the ingestion files
2133 0 : // were overlapping, either of which is an invariant violation.
2134 0 : panic("updated with two files in ingestSplit")
2135 : }
2136 2 : splitFile = replaced[i].Meta
2137 2 : updatedSplitFile = true
2138 : }
2139 : }
2140 2 : if !updatedSplitFile {
2141 2 : // None of the replaced files overlapped with the file being ingested.
2142 2 : // This can happen if we've already excised a span overlapping with
2143 2 : // this file, or if we have consecutive ingested files that can slide
2144 2 : // within the same gap between keys in an existing file. For instance,
2145 2 : // if an existing file has keys a and g and we're ingesting b-c, d-e,
2146 2 : // the first loop iteration will split the existing file into one that
2147 2 : // ends in a and another that starts at g, and the second iteration will
2148 2 : // fall into this case and require no splitting.
2149 2 : //
2150 2 : // No splitting necessary.
2151 2 : splitFile = nil
2152 2 : }
2153 : }
2154 2 : if splitFile == nil {
2155 2 : continue
2156 : }
2157 : // NB: excise operates on [start, end). We're splitting at [start, end]
2158 : // (assuming !s.ingestFile.Largest.IsExclusiveSentinel()). The conflation
2159 : // of exclusive vs inclusive end bounds should not make a difference here
2160 : // as we're guaranteed to not have any data overlap between splitFile and
2161 : // s.ingestFile. d.excise will return an error if we pass an inclusive user
2162 : // key bound _and_ we end up seeing data overlap at the end key.
2163 2 : added, err := d.excise(ctx, base.UserKeyBoundsFromInternal(s.ingestFile.Smallest, s.ingestFile.Largest), splitFile, ve, s.level)
2164 2 : if err != nil {
2165 0 : return err
2166 0 : }
2167 2 : if _, ok := ve.DeletedFiles[deletedFileEntry{
2168 2 : Level: s.level,
2169 2 : FileNum: splitFile.FileNum,
2170 2 : }]; !ok {
2171 0 : panic("did not split file that was expected to be split")
2172 : }
2173 2 : replacedFiles[splitFile.FileNum] = added
2174 2 : for i := range added {
2175 2 : addedBounds := added[i].Meta.UserKeyBounds()
2176 2 : if s.ingestFile.Overlaps(d.cmp, &addedBounds) {
2177 0 : panic("ingest-time split produced a file that overlaps with ingested file")
2178 : }
2179 : }
2180 2 : updateMetrics(splitFile, s.level, added)
2181 : }
2182 : // Flatten the version edit by removing any entries from ve.NewFiles that
2183 : // are also in ve.DeletedFiles.
2184 2 : newNewFiles := ve.NewFiles[:0]
2185 2 : for i := range ve.NewFiles {
2186 2 : fn := ve.NewFiles[i].Meta.FileNum
2187 2 : deEntry := deletedFileEntry{Level: ve.NewFiles[i].Level, FileNum: fn}
2188 2 : if _, ok := ve.DeletedFiles[deEntry]; ok {
2189 2 : delete(ve.DeletedFiles, deEntry)
2190 2 : } else {
2191 2 : newNewFiles = append(newNewFiles, ve.NewFiles[i])
2192 2 : }
2193 : }
2194 2 : ve.NewFiles = newNewFiles
2195 2 : return nil
2196 : }
2197 :
2198 : func (d *DB) ingestApply(
2199 : ctx context.Context,
2200 : jobID JobID,
2201 : lr ingestLoadResult,
2202 : mut *memTable,
2203 : exciseSpan KeyRange,
2204 : exciseSeqNum base.SeqNum,
2205 2 : ) (*versionEdit, error) {
2206 2 : d.mu.Lock()
2207 2 : defer d.mu.Unlock()
2208 2 :
2209 2 : ve := &versionEdit{
2210 2 : NewFiles: make([]newFileEntry, lr.fileCount()),
2211 2 : }
2212 2 : if exciseSpan.Valid() || (d.opts.Experimental.IngestSplit != nil && d.opts.Experimental.IngestSplit()) {
2213 2 : ve.DeletedFiles = map[manifest.DeletedFileEntry]*manifest.FileMetadata{}
2214 2 : }
2215 2 : metrics := make(map[int]*LevelMetrics)
2216 2 :
2217 2 : // Lock the manifest for writing before we use the current version to
2218 2 : // determine the target level. This prevents two concurrent ingestion jobs
2219 2 : // from using the same version to determine the target level, and also
2220 2 : // provides serialization with concurrent compaction and flush jobs.
2221 2 : // logAndApply unconditionally releases the manifest lock, but any earlier
2222 2 : // returns must unlock the manifest.
2223 2 : d.mu.versions.logLock()
2224 2 :
2225 2 : if mut != nil {
2226 2 : // Unref the mutable memtable to allows its flush to proceed. Now that we've
2227 2 : // acquired the manifest lock, we can be certain that if the mutable
2228 2 : // memtable has received more recent conflicting writes, the flush won't
2229 2 : // beat us to applying to the manifest resulting in sequence number
2230 2 : // inversion. Even though we call maybeScheduleFlush right now, this flush
2231 2 : // will apply after our ingestion.
2232 2 : if mut.writerUnref() {
2233 2 : d.maybeScheduleFlush()
2234 2 : }
2235 : }
2236 :
2237 2 : current := d.mu.versions.currentVersion()
2238 2 : overlapChecker := &overlapChecker{
2239 2 : comparer: d.opts.Comparer,
2240 2 : newIters: d.newIters,
2241 2 : opts: IterOptions{
2242 2 : logger: d.opts.Logger,
2243 2 : Category: categoryIngest,
2244 2 : },
2245 2 : v: current,
2246 2 : }
2247 2 : shouldIngestSplit := d.opts.Experimental.IngestSplit != nil &&
2248 2 : d.opts.Experimental.IngestSplit() && d.FormatMajorVersion() >= FormatVirtualSSTables
2249 2 : baseLevel := d.mu.versions.picker.getBaseLevel()
2250 2 : // filesToSplit is a list where each element is a pair consisting of a file
2251 2 : // being ingested and a file being split to make room for an ingestion into
2252 2 : // that level. Each ingested file will appear at most once in this list. It
2253 2 : // is possible for split files to appear twice in this list.
2254 2 : filesToSplit := make([]ingestSplitFile, 0)
2255 2 : checkCompactions := false
2256 2 : for i := 0; i < lr.fileCount(); i++ {
2257 2 : // Determine the lowest level in the LSM for which the sstable doesn't
2258 2 : // overlap any existing files in the level.
2259 2 : var m *fileMetadata
2260 2 : specifiedLevel := -1
2261 2 : isShared := false
2262 2 : isExternal := false
2263 2 : if i < len(lr.local) {
2264 2 : // local file.
2265 2 : m = lr.local[i].fileMetadata
2266 2 : } else if (i - len(lr.local)) < len(lr.shared) {
2267 2 : // shared file.
2268 2 : isShared = true
2269 2 : sharedIdx := i - len(lr.local)
2270 2 : m = lr.shared[sharedIdx].fileMetadata
2271 2 : specifiedLevel = int(lr.shared[sharedIdx].shared.Level)
2272 2 : } else {
2273 2 : // external file.
2274 2 : isExternal = true
2275 2 : externalIdx := i - (len(lr.local) + len(lr.shared))
2276 2 : m = lr.external[externalIdx].fileMetadata
2277 2 : if lr.externalFilesHaveLevel {
2278 1 : specifiedLevel = int(lr.external[externalIdx].external.Level)
2279 1 : }
2280 : }
2281 :
2282 : // Add to CreatedBackingTables if this is a new backing.
2283 : //
2284 : // Shared files always have a new backing. External files have new backings
2285 : // iff the backing disk file num and the file num match (see ingestAttachRemote).
2286 2 : if isShared || (isExternal && m.FileBacking.DiskFileNum == base.DiskFileNum(m.FileNum)) {
2287 2 : ve.CreatedBackingTables = append(ve.CreatedBackingTables, m.FileBacking)
2288 2 : }
2289 :
2290 2 : f := &ve.NewFiles[i]
2291 2 : var err error
2292 2 : if specifiedLevel != -1 {
2293 2 : f.Level = specifiedLevel
2294 2 : } else {
2295 2 : var splitFile *fileMetadata
2296 2 : if exciseSpan.Valid() && exciseSpan.Contains(d.cmp, m.Smallest) && exciseSpan.Contains(d.cmp, m.Largest) {
2297 2 : // This file fits perfectly within the excise span. We can slot it at
2298 2 : // L6, or sharedLevelsStart - 1 if we have shared files.
2299 2 : if len(lr.shared) > 0 || lr.externalFilesHaveLevel {
2300 2 : f.Level = sharedLevelsStart - 1
2301 2 : if baseLevel > f.Level {
2302 2 : f.Level = 0
2303 2 : }
2304 2 : } else {
2305 2 : f.Level = 6
2306 2 : }
2307 2 : } else {
2308 2 : // We check overlap against the LSM without holding DB.mu. Note that we
2309 2 : // are still holding the log lock, so the version cannot change.
2310 2 : // TODO(radu): perform this check optimistically outside of the log lock.
2311 2 : var lsmOverlap overlap.WithLSM
2312 2 : lsmOverlap, err = func() (overlap.WithLSM, error) {
2313 2 : d.mu.Unlock()
2314 2 : defer d.mu.Lock()
2315 2 : return overlapChecker.DetermineLSMOverlap(ctx, m.UserKeyBounds())
2316 2 : }()
2317 2 : if err == nil {
2318 2 : f.Level, splitFile, err = ingestTargetLevel(
2319 2 : ctx, d.cmp, lsmOverlap, baseLevel, d.mu.compact.inProgress, m, shouldIngestSplit,
2320 2 : )
2321 2 : }
2322 : }
2323 :
2324 2 : if splitFile != nil {
2325 2 : if invariants.Enabled {
2326 2 : if lf := current.Levels[f.Level].Find(d.cmp, splitFile); lf.Empty() {
2327 0 : panic("splitFile returned is not in level it should be")
2328 : }
2329 : }
2330 : // We take advantage of the fact that we won't drop the db mutex
2331 : // between now and the call to logAndApply. So, no files should
2332 : // get added to a new in-progress compaction at this point. We can
2333 : // avoid having to iterate on in-progress compactions to cancel them
2334 : // if none of the files being split have a compacting state.
2335 2 : if splitFile.IsCompacting() {
2336 1 : checkCompactions = true
2337 1 : }
2338 2 : filesToSplit = append(filesToSplit, ingestSplitFile{ingestFile: m, splitFile: splitFile, level: f.Level})
2339 : }
2340 : }
2341 2 : if err != nil {
2342 0 : d.mu.versions.logUnlock()
2343 0 : return nil, err
2344 0 : }
2345 2 : if isShared && f.Level < sharedLevelsStart {
2346 0 : panic(fmt.Sprintf("cannot slot a shared file higher than the highest shared level: %d < %d",
2347 0 : f.Level, sharedLevelsStart))
2348 : }
2349 2 : f.Meta = m
2350 2 : levelMetrics := metrics[f.Level]
2351 2 : if levelMetrics == nil {
2352 2 : levelMetrics = &LevelMetrics{}
2353 2 : metrics[f.Level] = levelMetrics
2354 2 : }
2355 2 : levelMetrics.NumFiles++
2356 2 : levelMetrics.Size += int64(m.Size)
2357 2 : levelMetrics.BytesIngested += m.Size
2358 2 : levelMetrics.TablesIngested++
2359 : }
2360 : // replacedFiles maps files excised due to exciseSpan (or splitFiles returned
2361 : // by ingestTargetLevel), to files that were created to replace it. This map
2362 : // is used to resolve references to split files in filesToSplit, as it is
2363 : // possible for a file that we want to split to no longer exist or have a
2364 : // newer fileMetadata due to a split induced by another ingestion file, or an
2365 : // excise.
2366 2 : replacedFiles := make(map[base.FileNum][]newFileEntry)
2367 2 : updateLevelMetricsOnExcise := func(m *fileMetadata, level int, added []newFileEntry) {
2368 2 : levelMetrics := metrics[level]
2369 2 : if levelMetrics == nil {
2370 2 : levelMetrics = &LevelMetrics{}
2371 2 : metrics[level] = levelMetrics
2372 2 : }
2373 2 : levelMetrics.NumFiles--
2374 2 : levelMetrics.Size -= int64(m.Size)
2375 2 : for i := range added {
2376 2 : levelMetrics.NumFiles++
2377 2 : levelMetrics.Size += int64(added[i].Meta.Size)
2378 2 : }
2379 : }
2380 2 : if exciseSpan.Valid() {
2381 2 : // Iterate through all levels and find files that intersect with exciseSpan.
2382 2 : //
2383 2 : // TODO(bilal): We could drop the DB mutex here as we don't need it for
2384 2 : // excises; we only need to hold the version lock which we already are
2385 2 : // holding. However releasing the DB mutex could mess with the
2386 2 : // ingestTargetLevel calculation that happened above, as it assumed that it
2387 2 : // had a complete view of in-progress compactions that wouldn't change
2388 2 : // until logAndApply is called. If we were to drop the mutex now, we could
2389 2 : // schedule another in-progress compaction that would go into the chosen target
2390 2 : // level and lead to file overlap within level (which would panic in
2391 2 : // logAndApply). We should drop the db mutex here, do the excise, then
2392 2 : // re-grab the DB mutex and rerun just the in-progress compaction check to
2393 2 : // see if any new compactions are conflicting with our chosen target levels
2394 2 : // for files, and if they are, we should signal those compactions to error
2395 2 : // out.
2396 2 : for level := range current.Levels {
2397 2 : overlaps := current.Overlaps(level, exciseSpan.UserKeyBounds())
2398 2 : iter := overlaps.Iter()
2399 2 :
2400 2 : for m := iter.First(); m != nil; m = iter.Next() {
2401 2 : newFiles, err := d.excise(ctx, exciseSpan.UserKeyBounds(), m, ve, level)
2402 2 : if err != nil {
2403 0 : return nil, err
2404 0 : }
2405 :
2406 2 : if _, ok := ve.DeletedFiles[deletedFileEntry{
2407 2 : Level: level,
2408 2 : FileNum: m.FileNum,
2409 2 : }]; !ok {
2410 2 : // We did not excise this file.
2411 2 : continue
2412 : }
2413 2 : replacedFiles[m.FileNum] = newFiles
2414 2 : updateLevelMetricsOnExcise(m, level, newFiles)
2415 : }
2416 : }
2417 : }
2418 2 : if len(filesToSplit) > 0 {
2419 2 : // For the same reasons as the above call to excise, we hold the db mutex
2420 2 : // while calling this method.
2421 2 : if err := d.ingestSplit(ctx, ve, updateLevelMetricsOnExcise, filesToSplit, replacedFiles); err != nil {
2422 0 : return nil, err
2423 0 : }
2424 : }
2425 2 : if len(filesToSplit) > 0 || exciseSpan.Valid() {
2426 2 : for c := range d.mu.compact.inProgress {
2427 2 : if c.versionEditApplied {
2428 1 : continue
2429 : }
2430 : // Check if this compaction overlaps with the excise span. Note that just
2431 : // checking if the inputs individually overlap with the excise span
2432 : // isn't sufficient; for instance, a compaction could have [a,b] and [e,f]
2433 : // as inputs and write it all out as [a,b,e,f] in one sstable. If we're
2434 : // doing a [c,d) excise at the same time as this compaction, we will have
2435 : // to error out the whole compaction as we can't guarantee it hasn't/won't
2436 : // write a file overlapping with the excise span.
2437 2 : if exciseSpan.OverlapsInternalKeyRange(d.cmp, c.smallest, c.largest) {
2438 2 : c.cancel.Store(true)
2439 2 : }
2440 : // Check if this compaction's inputs have been replaced due to an
2441 : // ingest-time split. In that case, cancel the compaction as a newly picked
2442 : // compaction would need to include any new files that slid in between
2443 : // previously-existing files. Note that we cancel any compaction that has a
2444 : // file that was ingest-split as an input, even if it started before this
2445 : // ingestion.
2446 2 : if checkCompactions {
2447 1 : for i := range c.inputs {
2448 1 : iter := c.inputs[i].files.Iter()
2449 1 : for f := iter.First(); f != nil; f = iter.Next() {
2450 1 : if _, ok := replacedFiles[f.FileNum]; ok {
2451 1 : c.cancel.Store(true)
2452 1 : break
2453 : }
2454 : }
2455 : }
2456 : }
2457 : }
2458 : }
2459 :
2460 2 : if err := d.mu.versions.logAndApply(jobID, ve, metrics, false /* forceRotation */, func() []compactionInfo {
2461 2 : return d.getInProgressCompactionInfoLocked(nil)
2462 2 : }); err != nil {
2463 1 : // Note: any error during logAndApply is fatal; this won't be reachable in production.
2464 1 : return nil, err
2465 1 : }
2466 :
2467 : // Check for any EventuallyFileOnlySnapshots that could be watching for
2468 : // an excise on this span. There should be none as the
2469 : // computePossibleOverlaps steps should have forced these EFOS to transition
2470 : // to file-only snapshots by now. If we see any that conflict with this
2471 : // excise, panic.
2472 2 : if exciseSpan.Valid() {
2473 2 : for s := d.mu.snapshots.root.next; s != &d.mu.snapshots.root; s = s.next {
2474 1 : // Skip non-EFOS snapshots, and also skip any EFOS that were created
2475 1 : // *after* the excise.
2476 1 : if s.efos == nil || base.Visible(exciseSeqNum, s.efos.seqNum, base.SeqNumMax) {
2477 0 : continue
2478 : }
2479 1 : efos := s.efos
2480 1 : // TODO(bilal): We can make this faster by taking advantage of the sorted
2481 1 : // nature of protectedRanges to do a sort.Search, or even maintaining a
2482 1 : // global list of all protected ranges instead of having to peer into every
2483 1 : // snapshot.
2484 1 : for i := range efos.protectedRanges {
2485 1 : if efos.protectedRanges[i].OverlapsKeyRange(d.cmp, exciseSpan) {
2486 0 : panic("unexpected excise of an EventuallyFileOnlySnapshot's bounds")
2487 : }
2488 : }
2489 : }
2490 : }
2491 :
2492 2 : d.mu.versions.metrics.Ingest.Count++
2493 2 :
2494 2 : d.updateReadStateLocked(d.opts.DebugCheck)
2495 2 : // updateReadStateLocked could have generated obsolete tables, schedule a
2496 2 : // cleanup job if necessary.
2497 2 : d.deleteObsoleteFiles(jobID)
2498 2 : d.updateTableStatsLocked(ve.NewFiles)
2499 2 : // The ingestion may have pushed a level over the threshold for compaction,
2500 2 : // so check to see if one is necessary and schedule it.
2501 2 : d.maybeScheduleCompaction()
2502 2 : var toValidate []manifest.NewFileEntry
2503 2 : dedup := make(map[base.DiskFileNum]struct{})
2504 2 : for _, entry := range ve.NewFiles {
2505 2 : if _, ok := dedup[entry.Meta.FileBacking.DiskFileNum]; !ok {
2506 2 : toValidate = append(toValidate, entry)
2507 2 : dedup[entry.Meta.FileBacking.DiskFileNum] = struct{}{}
2508 2 : }
2509 : }
2510 2 : d.maybeValidateSSTablesLocked(toValidate)
2511 2 : return ve, nil
2512 : }
2513 :
2514 : // maybeValidateSSTablesLocked adds the slice of newFileEntrys to the pending
2515 : // queue of files to be validated, when the feature is enabled.
2516 : //
2517 : // Note that if two entries with the same backing file are added twice, then the
2518 : // block checksums for the backing file will be validated twice.
2519 : //
2520 : // DB.mu must be locked when calling.
2521 2 : func (d *DB) maybeValidateSSTablesLocked(newFiles []newFileEntry) {
2522 2 : // Only add to the validation queue when the feature is enabled.
2523 2 : if !d.opts.Experimental.ValidateOnIngest {
2524 2 : return
2525 2 : }
2526 :
2527 2 : d.mu.tableValidation.pending = append(d.mu.tableValidation.pending, newFiles...)
2528 2 : if d.shouldValidateSSTablesLocked() {
2529 2 : go d.validateSSTables()
2530 2 : }
2531 : }
2532 :
2533 : // shouldValidateSSTablesLocked returns true if SSTable validation should run.
2534 : // DB.mu must be locked when calling.
2535 2 : func (d *DB) shouldValidateSSTablesLocked() bool {
2536 2 : return !d.mu.tableValidation.validating &&
2537 2 : d.closed.Load() == nil &&
2538 2 : d.opts.Experimental.ValidateOnIngest &&
2539 2 : len(d.mu.tableValidation.pending) > 0
2540 2 : }
2541 :
2542 : // validateSSTables runs a round of validation on the tables in the pending
2543 : // queue.
2544 2 : func (d *DB) validateSSTables() {
2545 2 : d.mu.Lock()
2546 2 : if !d.shouldValidateSSTablesLocked() {
2547 2 : d.mu.Unlock()
2548 2 : return
2549 2 : }
2550 :
2551 2 : pending := d.mu.tableValidation.pending
2552 2 : d.mu.tableValidation.pending = nil
2553 2 : d.mu.tableValidation.validating = true
2554 2 : jobID := d.newJobIDLocked()
2555 2 : rs := d.loadReadState()
2556 2 :
2557 2 : // Drop DB.mu before performing IO.
2558 2 : d.mu.Unlock()
2559 2 :
2560 2 : // Validate all tables in the pending queue. This could lead to a situation
2561 2 : // where we are starving IO from other tasks due to having to page through
2562 2 : // all the blocks in all the sstables in the queue.
2563 2 : // TODO(travers): Add some form of pacing to avoid IO starvation.
2564 2 :
2565 2 : // If we fail to validate any files due to reasons other than uncovered
2566 2 : // corruption, accumulate them and re-queue them for another attempt.
2567 2 : var retry []manifest.NewFileEntry
2568 2 :
2569 2 : for _, f := range pending {
2570 2 : // The file may have been moved or deleted since it was ingested, in
2571 2 : // which case we skip.
2572 2 : if !rs.current.Contains(f.Level, f.Meta) {
2573 2 : // Assume the file was moved to a lower level. It is rare enough
2574 2 : // that a table is moved or deleted between the time it was ingested
2575 2 : // and the time the validation routine runs that the overall cost of
2576 2 : // this inner loop is tolerably low, when amortized over all
2577 2 : // ingested tables.
2578 2 : found := false
2579 2 : for i := f.Level + 1; i < numLevels; i++ {
2580 2 : if rs.current.Contains(i, f.Meta) {
2581 1 : found = true
2582 1 : break
2583 : }
2584 : }
2585 2 : if !found {
2586 2 : continue
2587 : }
2588 : }
2589 :
2590 2 : var err error
2591 2 : if f.Meta.Virtual {
2592 2 : err = d.fileCache.withVirtualReader(
2593 2 : f.Meta.VirtualMeta(), func(v sstable.VirtualReader) error {
2594 2 : return v.ValidateBlockChecksumsOnBacking()
2595 2 : })
2596 2 : } else {
2597 2 : err = d.fileCache.withReader(
2598 2 : f.Meta.PhysicalMeta(), func(r *sstable.Reader) error {
2599 2 : return r.ValidateBlockChecksums()
2600 2 : })
2601 : }
2602 :
2603 2 : if err != nil {
2604 1 : if IsCorruptionError(err) {
2605 1 : // TODO(travers): Hook into the corruption reporting pipeline, once
2606 1 : // available. See pebble#1192.
2607 1 : d.opts.Logger.Fatalf("pebble: encountered corruption during ingestion: %s", err)
2608 1 : } else {
2609 1 : // If there was some other, possibly transient, error that
2610 1 : // caused table validation to fail inform the EventListener and
2611 1 : // move on. We remember the table so that we can retry it in a
2612 1 : // subsequent table validation job.
2613 1 : //
2614 1 : // TODO(jackson): If the error is not transient, this will retry
2615 1 : // validation indefinitely. While not great, it's the same
2616 1 : // behavior as erroring flushes and compactions. We should
2617 1 : // address this as a part of #270.
2618 1 : d.opts.EventListener.BackgroundError(err)
2619 1 : retry = append(retry, f)
2620 1 : continue
2621 : }
2622 : }
2623 :
2624 2 : d.opts.EventListener.TableValidated(TableValidatedInfo{
2625 2 : JobID: int(jobID),
2626 2 : Meta: f.Meta,
2627 2 : })
2628 : }
2629 2 : rs.unref()
2630 2 : d.mu.Lock()
2631 2 : defer d.mu.Unlock()
2632 2 : d.mu.tableValidation.pending = append(d.mu.tableValidation.pending, retry...)
2633 2 : d.mu.tableValidation.validating = false
2634 2 : d.mu.tableValidation.cond.Broadcast()
2635 2 : if d.shouldValidateSSTablesLocked() {
2636 2 : go d.validateSSTables()
2637 2 : }
2638 : }
|
