Line data Source code
1 : // Copyright 2011 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "bytes"
9 : "context"
10 : "io"
11 : "math/rand/v2"
12 : "sync"
13 : "unsafe"
14 :
15 : "github.com/cockroachdb/errors"
16 : "github.com/cockroachdb/pebble/internal/base"
17 : "github.com/cockroachdb/pebble/internal/bytealloc"
18 : "github.com/cockroachdb/pebble/internal/humanize"
19 : "github.com/cockroachdb/pebble/internal/invariants"
20 : "github.com/cockroachdb/pebble/internal/keyspan"
21 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
22 : "github.com/cockroachdb/pebble/internal/manifest"
23 : "github.com/cockroachdb/pebble/internal/rangekeystack"
24 : "github.com/cockroachdb/pebble/internal/treeprinter"
25 : "github.com/cockroachdb/pebble/sstable"
26 : "github.com/cockroachdb/redact"
27 : )
28 :
29 : // iterPos describes the state of the internal iterator, in terms of whether it
30 : // is at the position returned to the user (cur), one ahead of the position
31 : // returned (next for forward iteration and prev for reverse iteration). The cur
32 : // position is split into two states, for forward and reverse iteration, since
33 : // we need to differentiate for switching directions.
34 : //
35 : // There is subtlety in what is considered the current position of the Iterator.
36 : // The internal iterator exposes a sequence of internal keys. There is not
37 : // always a single internalIterator position corresponding to the position
38 : // returned to the user. Consider the example:
39 : //
40 : // a.MERGE.9 a.MERGE.8 a.MERGE.7 a.SET.6 b.DELETE.9 b.DELETE.5 b.SET.4
41 : // \ /
42 : // \ Iterator.Key() = 'a' /
43 : //
44 : // The Iterator exposes one valid position at user key 'a' and the two exhausted
45 : // positions at the beginning and end of iteration. The underlying
46 : // internalIterator contains 7 valid positions and 2 exhausted positions.
47 : //
48 : // Iterator positioning methods must set iterPos to iterPosCur{Foward,Backward}
49 : // iff the user key at the current internalIterator position equals the
50 : // Iterator.Key returned to the user. This guarantees that a call to nextUserKey
51 : // or prevUserKey will advance to the next or previous iterator position.
52 : // iterPosCur{Forward,Backward} does not make any guarantee about the internal
53 : // iterator position among internal keys with matching user keys, and it will
54 : // vary subtly depending on the particular key kinds encountered. In the above
55 : // example, the iterator returning 'a' may set iterPosCurForward if the internal
56 : // iterator is positioned at any of a.MERGE.9, a.MERGE.8, a.MERGE.7 or a.SET.6.
57 : //
58 : // When setting iterPos to iterPosNext or iterPosPrev, the internal iterator
59 : // must be advanced to the first internalIterator position at a user key greater
60 : // (iterPosNext) or less (iterPosPrev) than the key returned to the user. An
61 : // internalIterator position that's !Valid() must also be considered greater or
62 : // less—depending on the direction of iteration—than the last valid Iterator
63 : // position.
64 : type iterPos int8
65 :
66 : const (
67 : iterPosCurForward iterPos = 0
68 : iterPosNext iterPos = 1
69 : iterPosPrev iterPos = -1
70 : iterPosCurReverse iterPos = -2
71 :
72 : // For limited iteration. When the iterator is at iterPosCurForwardPaused
73 : // - Next*() call should behave as if the internal iterator is already
74 : // at next (akin to iterPosNext).
75 : // - Prev*() call should behave as if the internal iterator is at the
76 : // current key (akin to iterPosCurForward).
77 : //
78 : // Similar semantics apply to CurReversePaused.
79 : iterPosCurForwardPaused iterPos = 2
80 : iterPosCurReversePaused iterPos = -3
81 : )
82 :
83 : // Approximate gap in bytes between samples of data read during iteration.
84 : // This is multiplied with a default ReadSamplingMultiplier of 1 << 4 to yield
85 : // 1 << 20 (1MB). The 1MB factor comes from:
86 : // https://github.com/cockroachdb/pebble/issues/29#issuecomment-494477985
87 : const readBytesPeriod uint64 = 1 << 16
88 :
89 : var errReversePrefixIteration = errors.New("pebble: unsupported reverse prefix iteration")
90 :
91 : // IteratorMetrics holds per-iterator metrics. These do not change over the
92 : // lifetime of the iterator.
93 : type IteratorMetrics struct {
94 : // The read amplification experienced by this iterator. This is the sum of
95 : // the memtables, the L0 sublevels and the non-empty Ln levels. Higher read
96 : // amplification generally results in slower reads, though allowing higher
97 : // read amplification can also result in faster writes.
98 : ReadAmp int
99 : }
100 :
101 : // IteratorStatsKind describes the two kind of iterator stats.
102 : type IteratorStatsKind int8
103 :
104 : const (
105 : // InterfaceCall represents calls to Iterator.
106 : InterfaceCall IteratorStatsKind = iota
107 : // InternalIterCall represents calls by Iterator to its internalIterator.
108 : InternalIterCall
109 : // NumStatsKind is the number of kinds, and is used for array sizing.
110 : NumStatsKind
111 : )
112 :
113 : // IteratorStats contains iteration stats.
114 : type IteratorStats struct {
115 : // ForwardSeekCount includes SeekGE, SeekPrefixGE, First.
116 : ForwardSeekCount [NumStatsKind]int
117 : // ReverseSeek includes SeekLT, Last.
118 : ReverseSeekCount [NumStatsKind]int
119 : // ForwardStepCount includes Next.
120 : ForwardStepCount [NumStatsKind]int
121 : // ReverseStepCount includes Prev.
122 : ReverseStepCount [NumStatsKind]int
123 : InternalStats InternalIteratorStats
124 : RangeKeyStats RangeKeyIteratorStats
125 : }
126 :
127 : var _ redact.SafeFormatter = &IteratorStats{}
128 :
129 : // InternalIteratorStats contains miscellaneous stats produced by internal
130 : // iterators.
131 : type InternalIteratorStats = base.InternalIteratorStats
132 :
133 : // RangeKeyIteratorStats contains miscellaneous stats about range keys
134 : // encountered by the iterator.
135 : type RangeKeyIteratorStats struct {
136 : // Count records the number of range keys encountered during
137 : // iteration. Range keys may be counted multiple times if the iterator
138 : // leaves a range key's bounds and then returns.
139 : Count int
140 : // ContainedPoints records the number of point keys encountered within the
141 : // bounds of a range key. Note that this includes point keys with suffixes
142 : // that sort both above and below the covering range key's suffix.
143 : ContainedPoints int
144 : // SkippedPoints records the count of the subset of ContainedPoints point
145 : // keys that were skipped during iteration due to range-key masking. It does
146 : // not include point keys that were never loaded because a
147 : // RangeKeyMasking.Filter excluded the entire containing block.
148 : SkippedPoints int
149 : }
150 :
151 : // Merge adds all of the argument's statistics to the receiver. It may be used
152 : // to accumulate stats across multiple iterators.
153 1 : func (s *RangeKeyIteratorStats) Merge(o RangeKeyIteratorStats) {
154 1 : s.Count += o.Count
155 1 : s.ContainedPoints += o.ContainedPoints
156 1 : s.SkippedPoints += o.SkippedPoints
157 1 : }
158 :
159 0 : func (s *RangeKeyIteratorStats) String() string {
160 0 : return redact.StringWithoutMarkers(s)
161 0 : }
162 :
163 : // SafeFormat implements the redact.SafeFormatter interface.
164 1 : func (s *RangeKeyIteratorStats) SafeFormat(p redact.SafePrinter, verb rune) {
165 1 : p.Printf("range keys: %s, contained points: %s (%s skipped)",
166 1 : humanize.Count.Uint64(uint64(s.Count)),
167 1 : humanize.Count.Uint64(uint64(s.ContainedPoints)),
168 1 : humanize.Count.Uint64(uint64(s.SkippedPoints)))
169 1 : }
170 :
171 : // LazyValue is a lazy value. See the long comment in base.LazyValue.
172 : type LazyValue = base.LazyValue
173 :
174 : // Iterator iterates over a DB's key/value pairs in key order.
175 : //
176 : // An iterator must be closed after use, but it is not necessary to read an
177 : // iterator until exhaustion.
178 : //
179 : // An iterator is not goroutine-safe, but it is safe to use multiple iterators
180 : // concurrently, with each in a dedicated goroutine.
181 : //
182 : // It is also safe to use an iterator concurrently with modifying its
183 : // underlying DB, if that DB permits modification. However, the resultant
184 : // key/value pairs are not guaranteed to be a consistent snapshot of that DB
185 : // at a particular point in time.
186 : //
187 : // If an iterator encounters an error during any operation, it is stored by
188 : // the Iterator and surfaced through the Error method. All absolute
189 : // positioning methods (eg, SeekLT, SeekGT, First, Last, etc) reset any
190 : // accumulated error before positioning. All relative positioning methods (eg,
191 : // Next, Prev) return without advancing if the iterator has an accumulated
192 : // error.
193 : type Iterator struct {
194 : // The context is stored here since (a) Iterators are expected to be
195 : // short-lived (since they pin memtables and sstables), (b) plumbing a
196 : // context into every method is very painful, (c) they do not (yet) respect
197 : // context cancellation and are only used for tracing.
198 : ctx context.Context
199 : opts IterOptions
200 : merge Merge
201 : comparer base.Comparer
202 : iter internalIterator
203 : pointIter topLevelIterator
204 : // Either readState or version is set, but not both.
205 : readState *readState
206 : version *version
207 : // rangeKey holds iteration state specific to iteration over range keys.
208 : // The range key field may be nil if the Iterator has never been configured
209 : // to iterate over range keys. Its non-nilness cannot be used to determine
210 : // if the Iterator is currently iterating over range keys: For that, consult
211 : // the IterOptions using opts.rangeKeys(). If non-nil, its rangeKeyIter
212 : // field is guaranteed to be non-nil too.
213 : rangeKey *iteratorRangeKeyState
214 : // rangeKeyMasking holds state for range-key masking of point keys.
215 : rangeKeyMasking rangeKeyMasking
216 : err error
217 : // When iterValidityState=IterValid, key represents the current key, which
218 : // is backed by keyBuf.
219 : key []byte
220 : keyBuf []byte
221 : value LazyValue
222 : // For use in LazyValue.Clone.
223 : valueBuf []byte
224 : fetcher base.LazyFetcher
225 : // For use in LazyValue.Value.
226 : lazyValueBuf []byte
227 : valueCloser io.Closer
228 : // boundsBuf holds two buffers used to store the lower and upper bounds.
229 : // Whenever the Iterator's bounds change, the new bounds are copied into
230 : // boundsBuf[boundsBufIdx]. The two bounds share a slice to reduce
231 : // allocations. opts.LowerBound and opts.UpperBound point into this slice.
232 : boundsBuf [2][]byte
233 : boundsBufIdx int
234 : // iterKV reflects the latest position of iter, except when SetBounds is
235 : // called. In that case, it is explicitly set to nil.
236 : iterKV *base.InternalKV
237 : alloc *iterAlloc
238 : getIterAlloc *getIterAlloc
239 : prefixOrFullSeekKey []byte
240 : readSampling readSampling
241 : stats IteratorStats
242 : externalReaders [][]*sstable.Reader
243 :
244 : // Following fields used when constructing an iterator stack, eg, in Clone
245 : // and SetOptions or when re-fragmenting a batch's range keys/range dels.
246 : // Non-nil if this Iterator includes a Batch.
247 : batch *Batch
248 : newIters tableNewIters
249 : newIterRangeKey keyspanimpl.TableNewSpanIter
250 : lazyCombinedIter lazyCombinedIter
251 : seqNum base.SeqNum
252 : // batchSeqNum is used by Iterators over indexed batches to detect when the
253 : // underlying batch has been mutated. The batch beneath an indexed batch may
254 : // be mutated while the Iterator is open, but new keys are not surfaced
255 : // until the next call to SetOptions.
256 : batchSeqNum base.SeqNum
257 : // batch{PointIter,RangeDelIter,RangeKeyIter} are used when the Iterator is
258 : // configured to read through an indexed batch. If a batch is set, these
259 : // iterators will be included within the iterator stack regardless of
260 : // whether the batch currently contains any keys of their kind. These
261 : // pointers are used during a call to SetOptions to refresh the Iterator's
262 : // view of its indexed batch.
263 : batchPointIter batchIter
264 : batchRangeDelIter keyspan.Iter
265 : batchRangeKeyIter keyspan.Iter
266 : // merging is a pointer to this iterator's point merging iterator. It
267 : // appears here because key visibility is handled by the merging iterator.
268 : // During SetOptions on an iterator over an indexed batch, this field is
269 : // used to update the merging iterator's batch snapshot.
270 : merging *mergingIter
271 :
272 : // Keeping the bools here after all the 8 byte aligned fields shrinks the
273 : // sizeof this struct by 24 bytes.
274 :
275 : // INVARIANT:
276 : // iterValidityState==IterAtLimit <=>
277 : // pos==iterPosCurForwardPaused || pos==iterPosCurReversePaused
278 : iterValidityState IterValidityState
279 : // Set to true by SetBounds, SetOptions. Causes the Iterator to appear
280 : // exhausted externally, while preserving the correct iterValidityState for
281 : // the iterator's internal state. Preserving the correct internal validity
282 : // is used for SeekPrefixGE(..., trySeekUsingNext), and SeekGE/SeekLT
283 : // optimizations after "no-op" calls to SetBounds and SetOptions.
284 : requiresReposition bool
285 : // The position of iter. When this is iterPos{Prev,Next} the iter has been
286 : // moved past the current key-value, which can only happen if
287 : // iterValidityState=IterValid, i.e., there is something to return to the
288 : // client for the current position.
289 : pos iterPos
290 : // Relates to the prefixOrFullSeekKey field above.
291 : hasPrefix bool
292 : // Used for deriving the value of SeekPrefixGE(..., trySeekUsingNext),
293 : // and SeekGE/SeekLT optimizations
294 : lastPositioningOp lastPositioningOpKind
295 : // Used for determining when it's safe to perform SeekGE optimizations that
296 : // reuse the iterator state to avoid the cost of a full seek if the iterator
297 : // is already positioned in the correct place. If the iterator's view of its
298 : // indexed batch was just refreshed, some optimizations cannot be applied on
299 : // the first seek after the refresh:
300 : // - SeekGE has a no-op optimization that does not seek on the internal
301 : // iterator at all if the iterator is already in the correct place.
302 : // This optimization cannot be performed if the internal iterator was
303 : // last positioned when the iterator had a different view of an
304 : // underlying batch.
305 : // - Seek[Prefix]GE set flags.TrySeekUsingNext()=true when the seek key is
306 : // greater than the previous operation's seek key, under the expectation
307 : // that the various internal iterators can use their current position to
308 : // avoid a full expensive re-seek. This applies to the batchIter as well.
309 : // However, if the view of the batch was just refreshed, the batchIter's
310 : // position is not useful because it may already be beyond new keys less
311 : // than the seek key. To prevent the use of this optimization in
312 : // batchIter, Seek[Prefix]GE set flags.BatchJustRefreshed()=true if this
313 : // bit is enabled.
314 : batchJustRefreshed bool
315 : // batchOnlyIter is set to true for Batch.NewBatchOnlyIter.
316 : batchOnlyIter bool
317 : // Used in some tests to disable the random disabling of seek optimizations.
318 : forceEnableSeekOpt bool
319 : // Set to true if NextPrefix is not currently permitted. Defaults to false
320 : // in case an iterator never had any bounds.
321 : nextPrefixNotPermittedByUpperBound bool
322 : }
323 :
324 : // cmp is a convenience shorthand for the i.comparer.Compare function.
325 2 : func (i *Iterator) cmp(a, b []byte) int {
326 2 : return i.comparer.Compare(a, b)
327 2 : }
328 :
329 : // equal is a convenience shorthand for the i.comparer.Equal function.
330 2 : func (i *Iterator) equal(a, b []byte) bool {
331 2 : return i.comparer.Equal(a, b)
332 2 : }
333 :
334 : // iteratorRangeKeyState holds an iterator's range key iteration state.
335 : type iteratorRangeKeyState struct {
336 : opts *IterOptions
337 : cmp base.Compare
338 : split base.Split
339 : // rangeKeyIter holds the range key iterator stack that iterates over the
340 : // merged spans across the entirety of the LSM.
341 : rangeKeyIter keyspan.FragmentIterator
342 : iiter keyspan.InterleavingIter
343 : // stale is set to true when the range key state recorded here (in start,
344 : // end and keys) may not be in sync with the current range key at the
345 : // interleaving iterator's current position.
346 : //
347 : // When the interelaving iterator passes over a new span, it invokes the
348 : // SpanChanged hook defined on the `rangeKeyMasking` type, which sets stale
349 : // to true if the span is non-nil.
350 : //
351 : // The parent iterator may not be positioned over the interleaving
352 : // iterator's current position (eg, i.iterPos = iterPos{Next,Prev}), so
353 : // {keys,start,end} are only updated to the new range key during a call to
354 : // Iterator.saveRangeKey.
355 : stale bool
356 : // updated is used to signal to the Iterator client whether the state of
357 : // range keys has changed since the previous iterator position through the
358 : // `RangeKeyChanged` method. It's set to true during an Iterator positioning
359 : // operation that changes the state of the current range key. Each Iterator
360 : // positioning operation sets it back to false before executing.
361 : //
362 : // TODO(jackson): The lifecycle of {stale,updated,prevPosHadRangeKey} is
363 : // intricate and confusing. Try to refactor to reduce complexity.
364 : updated bool
365 : // prevPosHadRangeKey records whether the previous Iterator position had a
366 : // range key (HasPointAndRage() = (_, true)). It's updated at the beginning
367 : // of each new Iterator positioning operation. It's required by saveRangeKey to
368 : // to set `updated` appropriately: Without this record of the previous iterator
369 : // state, it's ambiguous whether an iterator only temporarily stepped onto a
370 : // position without a range key.
371 : prevPosHadRangeKey bool
372 : // rangeKeyOnly is set to true if at the current iterator position there is
373 : // no point key, only a range key start boundary.
374 : rangeKeyOnly bool
375 : // hasRangeKey is true when the current iterator position has a covering
376 : // range key (eg, a range key with bounds [<lower>,<upper>) such that
377 : // <lower> ≤ Key() < <upper>).
378 : hasRangeKey bool
379 : // start and end are the [start, end) boundaries of the current range keys.
380 : start []byte
381 : end []byte
382 :
383 : rangeKeyBuffers
384 :
385 : // iterConfig holds fields that are used for the construction of the
386 : // iterator stack, but do not need to be directly accessed during iteration.
387 : // This struct is bundled within the iteratorRangeKeyState struct to reduce
388 : // allocations.
389 : iterConfig rangekeystack.UserIteratorConfig
390 : }
391 :
392 : type rangeKeyBuffers struct {
393 : // keys is sorted by Suffix ascending.
394 : keys []RangeKeyData
395 : // buf is used to save range-key data before moving the range-key iterator.
396 : // Start and end boundaries, suffixes and values are all copied into buf.
397 : buf bytealloc.A
398 : // internal holds buffers used by the range key internal iterators.
399 : internal rangekeystack.Buffers
400 : }
401 :
402 2 : func (b *rangeKeyBuffers) PrepareForReuse() {
403 2 : const maxKeysReuse = 100
404 2 : if len(b.keys) > maxKeysReuse {
405 0 : b.keys = nil
406 0 : }
407 : // Avoid caching the key buf if it is overly large. The constant is
408 : // fairly arbitrary.
409 2 : if cap(b.buf) >= maxKeyBufCacheSize {
410 2 : b.buf = nil
411 2 : } else {
412 2 : b.buf = b.buf[:0]
413 2 : }
414 2 : b.internal.PrepareForReuse()
415 : }
416 :
417 2 : func (i *iteratorRangeKeyState) init(cmp base.Compare, split base.Split, opts *IterOptions) {
418 2 : i.cmp = cmp
419 2 : i.split = split
420 2 : i.opts = opts
421 2 : }
422 :
423 : var iterRangeKeyStateAllocPool = sync.Pool{
424 2 : New: func() interface{} {
425 2 : return &iteratorRangeKeyState{}
426 2 : },
427 : }
428 :
429 : // isEphemeralPosition returns true iff the current iterator position is
430 : // ephemeral, and won't be visited during subsequent relative positioning
431 : // operations.
432 : //
433 : // The iterator position resulting from a SeekGE or SeekPrefixGE that lands on a
434 : // straddling range key without a coincident point key is such a position.
435 2 : func (i *Iterator) isEphemeralPosition() bool {
436 2 : return i.opts.rangeKeys() && i.rangeKey != nil && i.rangeKey.rangeKeyOnly &&
437 2 : !i.equal(i.rangeKey.start, i.key)
438 2 : }
439 :
440 : type lastPositioningOpKind int8
441 :
442 : const (
443 : unknownLastPositionOp lastPositioningOpKind = iota
444 : seekPrefixGELastPositioningOp
445 : seekGELastPositioningOp
446 : seekLTLastPositioningOp
447 : // internalNextOp is a special internal iterator positioning operation used
448 : // by CanDeterministicallySingleDelete. It exists for enforcing requirements
449 : // around calling CanDeterministicallySingleDelete at most once per external
450 : // iterator position.
451 : internalNextOp
452 : )
453 :
454 : // Limited iteration mode. Not for use with prefix iteration.
455 : //
456 : // SeekGE, SeekLT, Prev, Next have WithLimit variants, that pause the iterator
457 : // at the limit in a best-effort manner. The client should behave correctly
458 : // even if the limits are ignored. These limits are not "deep", in that they
459 : // are not passed down to the underlying collection of internalIterators. This
460 : // is because the limits are transient, and apply only until the next
461 : // iteration call. They serve mainly as a way to bound the amount of work when
462 : // two (or more) Iterators are being coordinated at a higher level.
463 : //
464 : // In limited iteration mode:
465 : // - Avoid using Iterator.Valid if the last call was to a *WithLimit() method.
466 : // The return value from the *WithLimit() method provides a more precise
467 : // disposition.
468 : // - The limit is exclusive for forward and inclusive for reverse.
469 : //
470 : //
471 : // Limited iteration mode & range keys
472 : //
473 : // Limited iteration interacts with range-key iteration. When range key
474 : // iteration is enabled, range keys are interleaved at their start boundaries.
475 : // Limited iteration must ensure that if a range key exists within the limit,
476 : // the iterator visits the range key.
477 : //
478 : // During forward limited iteration, this is trivial: An overlapping range key
479 : // must have a start boundary less than the limit, and the range key's start
480 : // boundary will be interleaved and found to be within the limit.
481 : //
482 : // During reverse limited iteration, the tail of the range key may fall within
483 : // the limit. The range key must be surfaced even if the range key's start
484 : // boundary is less than the limit, and if there are no point keys between the
485 : // current iterator position and the limit. To provide this guarantee, reverse
486 : // limited iteration ignores the limit as long as there is a range key
487 : // overlapping the iteration position.
488 :
489 : // IterValidityState captures the state of the Iterator.
490 : type IterValidityState int8
491 :
492 : const (
493 : // IterExhausted represents an Iterator that is exhausted.
494 : IterExhausted IterValidityState = iota
495 : // IterValid represents an Iterator that is valid.
496 : IterValid
497 : // IterAtLimit represents an Iterator that has a non-exhausted
498 : // internalIterator, but has reached a limit without any key for the
499 : // caller.
500 : IterAtLimit
501 : )
502 :
503 : // readSampling stores variables used to sample a read to trigger a read
504 : // compaction
505 : type readSampling struct {
506 : bytesUntilReadSampling uint64
507 : initialSamplePassed bool
508 : pendingCompactions readCompactionQueue
509 : // forceReadSampling is used for testing purposes to force a read sample on every
510 : // call to Iterator.maybeSampleRead()
511 : forceReadSampling bool
512 : }
513 :
514 2 : func (i *Iterator) findNextEntry(limit []byte) {
515 2 : i.iterValidityState = IterExhausted
516 2 : i.pos = iterPosCurForward
517 2 : if i.opts.rangeKeys() && i.rangeKey != nil {
518 2 : i.rangeKey.rangeKeyOnly = false
519 2 : }
520 :
521 : // Close the closer for the current value if one was open.
522 2 : if i.closeValueCloser() != nil {
523 0 : return
524 0 : }
525 :
526 2 : for i.iterKV != nil {
527 2 : key := i.iterKV.K
528 2 :
529 2 : // The topLevelIterator.StrictSeekPrefixGE contract requires that in
530 2 : // prefix mode [i.hasPrefix=t], every point key returned by the internal
531 2 : // iterator must have the current iteration prefix.
532 2 : if invariants.Enabled && i.hasPrefix {
533 2 : // Range keys are an exception to the contract and may return a different
534 2 : // prefix. This case is explicitly handled in the switch statement below.
535 2 : if key.Kind() != base.InternalKeyKindRangeKeySet {
536 2 : if p := i.comparer.Split.Prefix(key.UserKey); !i.equal(i.prefixOrFullSeekKey, p) {
537 0 : i.opts.logger.Fatalf("pebble: prefix violation: key %q does not have prefix %q\n", key.UserKey, i.prefixOrFullSeekKey)
538 0 : }
539 : }
540 : }
541 :
542 : // Compare with limit every time we start at a different user key.
543 : // Note that given the best-effort contract of limit, we could avoid a
544 : // comparison in the common case by doing this only after
545 : // i.nextUserKey is called for the deletes below. However that makes
546 : // the behavior non-deterministic (since the behavior will vary based
547 : // on what has been compacted), which makes it hard to test with the
548 : // metamorphic test. So we forego that performance optimization.
549 2 : if limit != nil && i.cmp(limit, i.iterKV.K.UserKey) <= 0 {
550 2 : i.iterValidityState = IterAtLimit
551 2 : i.pos = iterPosCurForwardPaused
552 2 : return
553 2 : }
554 :
555 : // If the user has configured a SkipPoint function, invoke it to see
556 : // whether we should skip over the current user key.
557 2 : if i.opts.SkipPoint != nil && key.Kind() != InternalKeyKindRangeKeySet && i.opts.SkipPoint(i.iterKV.K.UserKey) {
558 2 : // NB: We could call nextUserKey, but in some cases the SkipPoint
559 2 : // predicate function might be cheaper than nextUserKey's key copy
560 2 : // and key comparison. This should be the case for MVCC suffix
561 2 : // comparisons, for example. In the future, we could expand the
562 2 : // SkipPoint interface to give the implementor more control over
563 2 : // whether we skip over just the internal key, the user key, or even
564 2 : // the key prefix.
565 2 : i.stats.ForwardStepCount[InternalIterCall]++
566 2 : i.iterKV = i.iter.Next()
567 2 : continue
568 : }
569 :
570 2 : switch key.Kind() {
571 2 : case InternalKeyKindRangeKeySet:
572 2 : if i.hasPrefix {
573 2 : if p := i.comparer.Split.Prefix(key.UserKey); !i.equal(i.prefixOrFullSeekKey, p) {
574 2 : return
575 2 : }
576 : }
577 : // Save the current key.
578 2 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
579 2 : i.key = i.keyBuf
580 2 : i.value = LazyValue{}
581 2 : // There may also be a live point key at this userkey that we have
582 2 : // not yet read. We need to find the next entry with this user key
583 2 : // to find it. Save the range key so we don't lose it when we Next
584 2 : // the underlying iterator.
585 2 : i.saveRangeKey()
586 2 : pointKeyExists := i.nextPointCurrentUserKey()
587 2 : if i.err != nil {
588 0 : i.iterValidityState = IterExhausted
589 0 : return
590 0 : }
591 2 : i.rangeKey.rangeKeyOnly = !pointKeyExists
592 2 : i.iterValidityState = IterValid
593 2 : return
594 :
595 2 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
596 2 : // NB: treating InternalKeyKindSingleDelete as equivalent to DEL is not
597 2 : // only simpler, but is also necessary for correctness due to
598 2 : // InternalKeyKindSSTableInternalObsoleteBit.
599 2 : i.nextUserKey()
600 2 : continue
601 :
602 2 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
603 2 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
604 2 : i.key = i.keyBuf
605 2 : i.value = i.iterKV.V
606 2 : i.iterValidityState = IterValid
607 2 : i.saveRangeKey()
608 2 : return
609 :
610 2 : case InternalKeyKindMerge:
611 2 : // Resolving the merge may advance us to the next point key, which
612 2 : // may be covered by a different set of range keys. Save the range
613 2 : // key state so we don't lose it.
614 2 : i.saveRangeKey()
615 2 : if i.mergeForward(key) {
616 2 : i.iterValidityState = IterValid
617 2 : return
618 2 : }
619 :
620 : // The merge didn't yield a valid key, either because the value
621 : // merger indicated it should be deleted, or because an error was
622 : // encountered.
623 1 : i.iterValidityState = IterExhausted
624 1 : if i.err != nil {
625 1 : return
626 1 : }
627 1 : if i.pos != iterPosNext {
628 0 : i.nextUserKey()
629 0 : }
630 1 : if i.closeValueCloser() != nil {
631 0 : return
632 0 : }
633 1 : i.pos = iterPosCurForward
634 :
635 0 : default:
636 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
637 0 : i.iterValidityState = IterExhausted
638 0 : return
639 : }
640 : }
641 :
642 : // Is iterKey nil due to an error?
643 2 : if err := i.iter.Error(); err != nil {
644 1 : i.err = err
645 1 : i.iterValidityState = IterExhausted
646 1 : }
647 : }
648 :
649 2 : func (i *Iterator) nextPointCurrentUserKey() bool {
650 2 : // If the user has configured a SkipPoint function and the current user key
651 2 : // would be skipped by it, there's no need to step forward looking for a
652 2 : // point key. If we were to find one, it should be skipped anyways.
653 2 : if i.opts.SkipPoint != nil && i.opts.SkipPoint(i.key) {
654 2 : return false
655 2 : }
656 :
657 2 : i.pos = iterPosCurForward
658 2 :
659 2 : i.iterKV = i.iter.Next()
660 2 : i.stats.ForwardStepCount[InternalIterCall]++
661 2 : if i.iterKV == nil {
662 2 : if err := i.iter.Error(); err != nil {
663 0 : i.err = err
664 2 : } else {
665 2 : i.pos = iterPosNext
666 2 : }
667 2 : return false
668 : }
669 2 : if !i.equal(i.key, i.iterKV.K.UserKey) {
670 2 : i.pos = iterPosNext
671 2 : return false
672 2 : }
673 :
674 2 : key := i.iterKV.K
675 2 : switch key.Kind() {
676 0 : case InternalKeyKindRangeKeySet:
677 0 : // RangeKeySets must always be interleaved as the first internal key
678 0 : // for a user key.
679 0 : i.err = base.CorruptionErrorf("pebble: unexpected range key set mid-user key")
680 0 : return false
681 :
682 2 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
683 2 : // NB: treating InternalKeyKindSingleDelete as equivalent to DEL is not
684 2 : // only simpler, but is also necessary for correctness due to
685 2 : // InternalKeyKindSSTableInternalObsoleteBit.
686 2 : return false
687 :
688 2 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
689 2 : i.value = i.iterKV.V
690 2 : return true
691 :
692 2 : case InternalKeyKindMerge:
693 2 : return i.mergeForward(key)
694 :
695 0 : default:
696 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
697 0 : return false
698 : }
699 : }
700 :
701 : // mergeForward resolves a MERGE key, advancing the underlying iterator forward
702 : // to merge with subsequent keys with the same userkey. mergeForward returns a
703 : // boolean indicating whether or not the merge yielded a valid key. A merge may
704 : // not yield a valid key if an error occurred, in which case i.err is non-nil,
705 : // or the user's value merger specified the key to be deleted.
706 : //
707 : // mergeForward does not update iterValidityState.
708 2 : func (i *Iterator) mergeForward(key base.InternalKey) (valid bool) {
709 2 : var iterValue []byte
710 2 : iterValue, _, i.err = i.iterKV.Value(nil)
711 2 : if i.err != nil {
712 0 : return false
713 0 : }
714 2 : var valueMerger ValueMerger
715 2 : valueMerger, i.err = i.merge(key.UserKey, iterValue)
716 2 : if i.err != nil {
717 0 : return false
718 0 : }
719 :
720 2 : i.mergeNext(key, valueMerger)
721 2 : if i.err != nil {
722 1 : return false
723 1 : }
724 :
725 2 : var needDelete bool
726 2 : var value []byte
727 2 : value, needDelete, i.valueCloser, i.err = finishValueMerger(
728 2 : valueMerger, true /* includesBase */)
729 2 : i.value = base.MakeInPlaceValue(value)
730 2 : if i.err != nil {
731 0 : return false
732 0 : }
733 2 : if needDelete {
734 1 : _ = i.closeValueCloser()
735 1 : return false
736 1 : }
737 2 : return true
738 : }
739 :
740 2 : func (i *Iterator) closeValueCloser() error {
741 2 : if i.valueCloser != nil {
742 0 : i.err = i.valueCloser.Close()
743 0 : i.valueCloser = nil
744 0 : }
745 2 : return i.err
746 : }
747 :
748 2 : func (i *Iterator) nextUserKey() {
749 2 : if i.iterKV == nil {
750 2 : return
751 2 : }
752 2 : trailer := i.iterKV.K.Trailer
753 2 : done := i.iterKV.K.Trailer <= base.InternalKeyZeroSeqnumMaxTrailer
754 2 : if i.iterValidityState != IterValid {
755 2 : i.keyBuf = append(i.keyBuf[:0], i.iterKV.K.UserKey...)
756 2 : i.key = i.keyBuf
757 2 : }
758 2 : for {
759 2 : i.stats.ForwardStepCount[InternalIterCall]++
760 2 : i.iterKV = i.iter.Next()
761 2 : if i.iterKV == nil {
762 2 : if err := i.iter.Error(); err != nil {
763 1 : i.err = err
764 1 : return
765 1 : }
766 : }
767 : // NB: We're guaranteed to be on the next user key if the previous key
768 : // had a zero sequence number (`done`), or the new key has a trailer
769 : // greater or equal to the previous key's trailer. This is true because
770 : // internal keys with the same user key are sorted by InternalKeyTrailer in
771 : // strictly monotonically descending order. We expect the trailer
772 : // optimization to trigger around 50% of the time with randomly
773 : // distributed writes. We expect it to trigger very frequently when
774 : // iterating through ingested sstables, which contain keys that all have
775 : // the same sequence number.
776 2 : if done || i.iterKV == nil || i.iterKV.K.Trailer >= trailer {
777 2 : break
778 : }
779 2 : if !i.equal(i.key, i.iterKV.K.UserKey) {
780 2 : break
781 : }
782 2 : done = i.iterKV.K.Trailer <= base.InternalKeyZeroSeqnumMaxTrailer
783 2 : trailer = i.iterKV.K.Trailer
784 : }
785 : }
786 :
787 2 : func (i *Iterator) maybeSampleRead() {
788 2 : // This method is only called when a public method of Iterator is
789 2 : // returning, and below we exclude the case were the iterator is paused at
790 2 : // a limit. The effect of these choices is that keys that are deleted, but
791 2 : // are encountered during iteration, are not accounted for in the read
792 2 : // sampling and will not cause read driven compactions, even though we are
793 2 : // incurring cost in iterating over them. And this issue is not limited to
794 2 : // Iterator, which does not see the effect of range deletes, which may be
795 2 : // causing iteration work in mergingIter. It is not clear at this time
796 2 : // whether this is a deficiency worth addressing.
797 2 : if i.iterValidityState != IterValid {
798 2 : return
799 2 : }
800 2 : if i.readState == nil {
801 2 : return
802 2 : }
803 2 : if i.readSampling.forceReadSampling {
804 1 : i.sampleRead()
805 1 : return
806 1 : }
807 2 : samplingPeriod := int32(int64(readBytesPeriod) * i.readState.db.opts.Experimental.ReadSamplingMultiplier)
808 2 : if samplingPeriod <= 0 {
809 0 : return
810 0 : }
811 2 : bytesRead := uint64(len(i.key) + i.value.Len())
812 2 : for i.readSampling.bytesUntilReadSampling < bytesRead {
813 2 : i.readSampling.bytesUntilReadSampling += uint64(rand.Uint32N(2 * uint32(samplingPeriod)))
814 2 : // The block below tries to adjust for the case where this is the
815 2 : // first read in a newly-opened iterator. As bytesUntilReadSampling
816 2 : // starts off at zero, we don't want to sample the first read of
817 2 : // every newly-opened iterator, but we do want to sample some of them.
818 2 : if !i.readSampling.initialSamplePassed {
819 2 : i.readSampling.initialSamplePassed = true
820 2 : if i.readSampling.bytesUntilReadSampling > bytesRead {
821 2 : if rand.Uint64N(i.readSampling.bytesUntilReadSampling) > bytesRead {
822 2 : continue
823 : }
824 : }
825 : }
826 2 : i.sampleRead()
827 : }
828 2 : i.readSampling.bytesUntilReadSampling -= bytesRead
829 : }
830 :
831 2 : func (i *Iterator) sampleRead() {
832 2 : var topFile *manifest.FileMetadata
833 2 : topLevel, numOverlappingLevels := numLevels, 0
834 2 : mi := i.merging
835 2 : if mi == nil {
836 2 : return
837 2 : }
838 2 : if len(mi.levels) > 1 {
839 2 : mi.ForEachLevelIter(func(li *levelIter) (done bool) {
840 2 : if li.layer.IsFlushableIngests() {
841 0 : return false
842 0 : }
843 2 : l := li.layer.Level()
844 2 : if f := li.iterFile; f != nil {
845 2 : var containsKey bool
846 2 : if i.pos == iterPosNext || i.pos == iterPosCurForward ||
847 2 : i.pos == iterPosCurForwardPaused {
848 2 : containsKey = i.cmp(f.SmallestPointKey.UserKey, i.key) <= 0
849 2 : } else if i.pos == iterPosPrev || i.pos == iterPosCurReverse ||
850 2 : i.pos == iterPosCurReversePaused {
851 2 : containsKey = i.cmp(f.LargestPointKey.UserKey, i.key) >= 0
852 2 : }
853 : // Do nothing if the current key is not contained in f's
854 : // bounds. We could seek the LevelIterator at this level
855 : // to find the right file, but the performance impacts of
856 : // doing that are significant enough to negate the benefits
857 : // of read sampling in the first place. See the discussion
858 : // at:
859 : // https://github.com/cockroachdb/pebble/pull/1041#issuecomment-763226492
860 2 : if containsKey {
861 2 : numOverlappingLevels++
862 2 : if numOverlappingLevels >= 2 {
863 2 : // Terminate the loop early if at least 2 overlapping levels are found.
864 2 : return true
865 2 : }
866 2 : topLevel = l
867 2 : topFile = f
868 : }
869 : }
870 2 : return false
871 : })
872 : }
873 2 : if topFile == nil || topLevel >= numLevels {
874 2 : return
875 2 : }
876 2 : if numOverlappingLevels >= 2 {
877 2 : allowedSeeks := topFile.AllowedSeeks.Add(-1)
878 2 : if allowedSeeks == 0 {
879 1 :
880 1 : // Since the compaction queue can handle duplicates, we can keep
881 1 : // adding to the queue even once allowedSeeks hits 0.
882 1 : // In fact, we NEED to keep adding to the queue, because the queue
883 1 : // is small and evicts older and possibly useful compactions.
884 1 : topFile.AllowedSeeks.Add(topFile.InitAllowedSeeks)
885 1 :
886 1 : read := readCompaction{
887 1 : start: topFile.SmallestPointKey.UserKey,
888 1 : end: topFile.LargestPointKey.UserKey,
889 1 : level: topLevel,
890 1 : fileNum: topFile.FileNum,
891 1 : }
892 1 : i.readSampling.pendingCompactions.add(&read, i.cmp)
893 1 : }
894 : }
895 : }
896 :
897 2 : func (i *Iterator) findPrevEntry(limit []byte) {
898 2 : i.iterValidityState = IterExhausted
899 2 : i.pos = iterPosCurReverse
900 2 : if i.opts.rangeKeys() && i.rangeKey != nil {
901 2 : i.rangeKey.rangeKeyOnly = false
902 2 : }
903 :
904 : // Close the closer for the current value if one was open.
905 2 : if i.valueCloser != nil {
906 0 : i.err = i.valueCloser.Close()
907 0 : i.valueCloser = nil
908 0 : if i.err != nil {
909 0 : i.iterValidityState = IterExhausted
910 0 : return
911 0 : }
912 : }
913 :
914 2 : var valueMerger ValueMerger
915 2 : firstLoopIter := true
916 2 : rangeKeyBoundary := false
917 2 : // The code below compares with limit in multiple places. As documented in
918 2 : // findNextEntry, this is being done to make the behavior of limit
919 2 : // deterministic to allow for metamorphic testing. It is not required by
920 2 : // the best-effort contract of limit.
921 2 : for i.iterKV != nil {
922 2 : key := i.iterKV.K
923 2 :
924 2 : // NB: We cannot pause if the current key is covered by a range key.
925 2 : // Otherwise, the user might not ever learn of a range key that covers
926 2 : // the key space being iterated over in which there are no point keys.
927 2 : // Since limits are best effort, ignoring the limit in this case is
928 2 : // allowed by the contract of limit.
929 2 : if firstLoopIter && limit != nil && i.cmp(limit, i.iterKV.K.UserKey) > 0 && !i.rangeKeyWithinLimit(limit) {
930 2 : i.iterValidityState = IterAtLimit
931 2 : i.pos = iterPosCurReversePaused
932 2 : return
933 2 : }
934 2 : firstLoopIter = false
935 2 :
936 2 : if i.iterValidityState == IterValid {
937 2 : if !i.equal(key.UserKey, i.key) {
938 2 : // We've iterated to the previous user key.
939 2 : i.pos = iterPosPrev
940 2 : if valueMerger != nil {
941 2 : var needDelete bool
942 2 : var value []byte
943 2 : value, needDelete, i.valueCloser, i.err = finishValueMerger(valueMerger, true /* includesBase */)
944 2 : i.value = base.MakeInPlaceValue(value)
945 2 : if i.err == nil && needDelete {
946 1 : // The point key at this key is deleted. If we also have
947 1 : // a range key boundary at this key, we still want to
948 1 : // return. Otherwise, we need to continue looking for
949 1 : // a live key.
950 1 : i.value = LazyValue{}
951 1 : if rangeKeyBoundary {
952 0 : i.rangeKey.rangeKeyOnly = true
953 1 : } else {
954 1 : i.iterValidityState = IterExhausted
955 1 : if i.closeValueCloser() == nil {
956 1 : continue
957 : }
958 : }
959 : }
960 : }
961 2 : if i.err != nil {
962 0 : i.iterValidityState = IterExhausted
963 0 : }
964 2 : return
965 : }
966 : }
967 :
968 : // If the user has configured a SkipPoint function, invoke it to see
969 : // whether we should skip over the current user key.
970 2 : if i.opts.SkipPoint != nil && key.Kind() != InternalKeyKindRangeKeySet && i.opts.SkipPoint(key.UserKey) {
971 2 : // NB: We could call prevUserKey, but in some cases the SkipPoint
972 2 : // predicate function might be cheaper than prevUserKey's key copy
973 2 : // and key comparison. This should be the case for MVCC suffix
974 2 : // comparisons, for example. In the future, we could expand the
975 2 : // SkipPoint interface to give the implementor more control over
976 2 : // whether we skip over just the internal key, the user key, or even
977 2 : // the key prefix.
978 2 : i.stats.ReverseStepCount[InternalIterCall]++
979 2 : i.iterKV = i.iter.Prev()
980 2 : if i.iterKV == nil {
981 2 : if err := i.iter.Error(); err != nil {
982 0 : i.err = err
983 0 : i.iterValidityState = IterExhausted
984 0 : return
985 0 : }
986 : }
987 2 : if limit != nil && i.iterKV != nil && i.cmp(limit, i.iterKV.K.UserKey) > 0 && !i.rangeKeyWithinLimit(limit) {
988 2 : i.iterValidityState = IterAtLimit
989 2 : i.pos = iterPosCurReversePaused
990 2 : return
991 2 : }
992 2 : continue
993 : }
994 :
995 2 : switch key.Kind() {
996 2 : case InternalKeyKindRangeKeySet:
997 2 : // Range key start boundary markers are interleaved with the maximum
998 2 : // sequence number, so if there's a point key also at this key, we
999 2 : // must've already iterated over it.
1000 2 : // This is the final entry at this user key, so we may return
1001 2 : i.rangeKey.rangeKeyOnly = i.iterValidityState != IterValid
1002 2 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1003 2 : i.key = i.keyBuf
1004 2 : i.iterValidityState = IterValid
1005 2 : i.saveRangeKey()
1006 2 : // In all other cases, previous iteration requires advancing to
1007 2 : // iterPosPrev in order to determine if the key is live and
1008 2 : // unshadowed by another key at the same user key. In this case,
1009 2 : // because range key start boundary markers are always interleaved
1010 2 : // at the maximum sequence number, we know that there aren't any
1011 2 : // additional keys with the same user key in the backward direction.
1012 2 : //
1013 2 : // We Prev the underlying iterator once anyways for consistency, so
1014 2 : // that we can maintain the invariant during backward iteration that
1015 2 : // i.iterPos = iterPosPrev.
1016 2 : i.stats.ReverseStepCount[InternalIterCall]++
1017 2 : i.iterKV = i.iter.Prev()
1018 2 :
1019 2 : // Set rangeKeyBoundary so that on the next iteration, we know to
1020 2 : // return the key even if the MERGE point key is deleted.
1021 2 : rangeKeyBoundary = true
1022 :
1023 2 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
1024 2 : i.value = LazyValue{}
1025 2 : i.iterValidityState = IterExhausted
1026 2 : valueMerger = nil
1027 2 : i.stats.ReverseStepCount[InternalIterCall]++
1028 2 : i.iterKV = i.iter.Prev()
1029 2 : // Compare with the limit. We could optimize by only checking when
1030 2 : // we step to the previous user key, but detecting that requires a
1031 2 : // comparison too. Note that this position may already passed a
1032 2 : // number of versions of this user key, but they are all deleted, so
1033 2 : // the fact that a subsequent Prev*() call will not see them is
1034 2 : // harmless. Also note that this is the only place in the loop,
1035 2 : // other than the firstLoopIter and SkipPoint cases above, where we
1036 2 : // could step to a different user key and start processing it for
1037 2 : // returning to the caller.
1038 2 : if limit != nil && i.iterKV != nil && i.cmp(limit, i.iterKV.K.UserKey) > 0 && !i.rangeKeyWithinLimit(limit) {
1039 2 : i.iterValidityState = IterAtLimit
1040 2 : i.pos = iterPosCurReversePaused
1041 2 : return
1042 2 : }
1043 2 : continue
1044 :
1045 2 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
1046 2 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1047 2 : i.key = i.keyBuf
1048 2 : // iterValue is owned by i.iter and could change after the Prev()
1049 2 : // call, so use valueBuf instead. Note that valueBuf is only used
1050 2 : // in this one instance; everywhere else (eg. in findNextEntry),
1051 2 : // we just point i.value to the unsafe i.iter-owned value buffer.
1052 2 : i.value, i.valueBuf = i.iterKV.V.Clone(i.valueBuf[:0], &i.fetcher)
1053 2 : i.saveRangeKey()
1054 2 : i.iterValidityState = IterValid
1055 2 : i.iterKV = i.iter.Prev()
1056 2 : i.stats.ReverseStepCount[InternalIterCall]++
1057 2 : valueMerger = nil
1058 2 : continue
1059 :
1060 2 : case InternalKeyKindMerge:
1061 2 : if i.iterValidityState == IterExhausted {
1062 2 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1063 2 : i.key = i.keyBuf
1064 2 : i.saveRangeKey()
1065 2 : var iterValue []byte
1066 2 : iterValue, _, i.err = i.iterKV.Value(nil)
1067 2 : if i.err != nil {
1068 0 : return
1069 0 : }
1070 2 : valueMerger, i.err = i.merge(i.key, iterValue)
1071 2 : if i.err != nil {
1072 0 : return
1073 0 : }
1074 2 : i.iterValidityState = IterValid
1075 2 : } else if valueMerger == nil {
1076 2 : // Extract value before iterValue since we use value before iterValue
1077 2 : // and the underlying iterator is not required to provide backing
1078 2 : // memory for both simultaneously.
1079 2 : var value []byte
1080 2 : var callerOwned bool
1081 2 : value, callerOwned, i.err = i.value.Value(i.lazyValueBuf)
1082 2 : if callerOwned {
1083 0 : i.lazyValueBuf = value[:0]
1084 0 : }
1085 2 : if i.err != nil {
1086 0 : i.iterValidityState = IterExhausted
1087 0 : return
1088 0 : }
1089 2 : valueMerger, i.err = i.merge(i.key, value)
1090 2 : var iterValue []byte
1091 2 : iterValue, _, i.err = i.iterKV.Value(nil)
1092 2 : if i.err != nil {
1093 0 : i.iterValidityState = IterExhausted
1094 0 : return
1095 0 : }
1096 2 : if i.err == nil {
1097 2 : i.err = valueMerger.MergeNewer(iterValue)
1098 2 : }
1099 2 : if i.err != nil {
1100 0 : i.iterValidityState = IterExhausted
1101 0 : return
1102 0 : }
1103 2 : } else {
1104 2 : var iterValue []byte
1105 2 : iterValue, _, i.err = i.iterKV.Value(nil)
1106 2 : if i.err != nil {
1107 0 : i.iterValidityState = IterExhausted
1108 0 : return
1109 0 : }
1110 2 : i.err = valueMerger.MergeNewer(iterValue)
1111 2 : if i.err != nil {
1112 0 : i.iterValidityState = IterExhausted
1113 0 : return
1114 0 : }
1115 : }
1116 2 : i.iterKV = i.iter.Prev()
1117 2 : i.stats.ReverseStepCount[InternalIterCall]++
1118 2 : continue
1119 :
1120 0 : default:
1121 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
1122 0 : i.iterValidityState = IterExhausted
1123 0 : return
1124 : }
1125 : }
1126 : // i.iterKV == nil, so broke out of the preceding loop.
1127 :
1128 : // Is iterKey nil due to an error?
1129 2 : if i.err = i.iter.Error(); i.err != nil {
1130 1 : i.iterValidityState = IterExhausted
1131 1 : return
1132 1 : }
1133 :
1134 2 : if i.iterValidityState == IterValid {
1135 2 : i.pos = iterPosPrev
1136 2 : if valueMerger != nil {
1137 2 : var needDelete bool
1138 2 : var value []byte
1139 2 : value, needDelete, i.valueCloser, i.err = finishValueMerger(valueMerger, true /* includesBase */)
1140 2 : i.value = base.MakeInPlaceValue(value)
1141 2 : if i.err == nil && needDelete {
1142 1 : i.key = nil
1143 1 : i.value = LazyValue{}
1144 1 : i.iterValidityState = IterExhausted
1145 1 : }
1146 : }
1147 2 : if i.err != nil {
1148 0 : i.iterValidityState = IterExhausted
1149 0 : }
1150 : }
1151 : }
1152 :
1153 2 : func (i *Iterator) prevUserKey() {
1154 2 : if i.iterKV == nil {
1155 2 : return
1156 2 : }
1157 2 : if i.iterValidityState != IterValid {
1158 2 : // If we're going to compare against the prev key, we need to save the
1159 2 : // current key.
1160 2 : i.keyBuf = append(i.keyBuf[:0], i.iterKV.K.UserKey...)
1161 2 : i.key = i.keyBuf
1162 2 : }
1163 2 : for {
1164 2 : i.iterKV = i.iter.Prev()
1165 2 : i.stats.ReverseStepCount[InternalIterCall]++
1166 2 : if i.iterKV == nil {
1167 2 : if err := i.iter.Error(); err != nil {
1168 0 : i.err = err
1169 0 : i.iterValidityState = IterExhausted
1170 0 : }
1171 2 : break
1172 : }
1173 2 : if !i.equal(i.key, i.iterKV.K.UserKey) {
1174 2 : break
1175 : }
1176 : }
1177 : }
1178 :
1179 2 : func (i *Iterator) mergeNext(key InternalKey, valueMerger ValueMerger) {
1180 2 : // Save the current key.
1181 2 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1182 2 : i.key = i.keyBuf
1183 2 :
1184 2 : // Loop looking for older values for this key and merging them.
1185 2 : for {
1186 2 : i.iterKV = i.iter.Next()
1187 2 : i.stats.ForwardStepCount[InternalIterCall]++
1188 2 : if i.iterKV == nil {
1189 2 : if i.err = i.iter.Error(); i.err != nil {
1190 1 : return
1191 1 : }
1192 2 : i.pos = iterPosNext
1193 2 : return
1194 : }
1195 2 : key = i.iterKV.K
1196 2 : if !i.equal(i.key, key.UserKey) {
1197 2 : // We've advanced to the next key.
1198 2 : i.pos = iterPosNext
1199 2 : return
1200 2 : }
1201 2 : switch key.Kind() {
1202 2 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
1203 2 : // We've hit a deletion tombstone. Return everything up to this
1204 2 : // point.
1205 2 : //
1206 2 : // NB: treating InternalKeyKindSingleDelete as equivalent to DEL is not
1207 2 : // only simpler, but is also necessary for correctness due to
1208 2 : // InternalKeyKindSSTableInternalObsoleteBit.
1209 2 : return
1210 :
1211 2 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
1212 2 : // We've hit a Set value. Merge with the existing value and return.
1213 2 : var iterValue []byte
1214 2 : iterValue, _, i.err = i.iterKV.Value(nil)
1215 2 : if i.err != nil {
1216 0 : return
1217 0 : }
1218 2 : i.err = valueMerger.MergeOlder(iterValue)
1219 2 : return
1220 :
1221 2 : case InternalKeyKindMerge:
1222 2 : // We've hit another Merge value. Merge with the existing value and
1223 2 : // continue looping.
1224 2 : var iterValue []byte
1225 2 : iterValue, _, i.err = i.iterKV.Value(nil)
1226 2 : if i.err != nil {
1227 0 : return
1228 0 : }
1229 2 : i.err = valueMerger.MergeOlder(iterValue)
1230 2 : if i.err != nil {
1231 0 : return
1232 0 : }
1233 2 : continue
1234 :
1235 0 : case InternalKeyKindRangeKeySet:
1236 0 : // The RANGEKEYSET marker must sort before a MERGE at the same user key.
1237 0 : i.err = base.CorruptionErrorf("pebble: out of order range key marker")
1238 0 : return
1239 :
1240 0 : default:
1241 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
1242 0 : return
1243 : }
1244 : }
1245 : }
1246 :
1247 : // SeekGE moves the iterator to the first key/value pair whose key is greater
1248 : // than or equal to the given key. Returns true if the iterator is pointing at
1249 : // a valid entry and false otherwise.
1250 2 : func (i *Iterator) SeekGE(key []byte) bool {
1251 2 : return i.SeekGEWithLimit(key, nil) == IterValid
1252 2 : }
1253 :
1254 : // SeekGEWithLimit moves the iterator to the first key/value pair whose key is
1255 : // greater than or equal to the given key.
1256 : //
1257 : // If limit is provided, it serves as a best-effort exclusive limit. If the
1258 : // first key greater than or equal to the given search key is also greater than
1259 : // or equal to limit, the Iterator may pause and return IterAtLimit. Because
1260 : // limits are best-effort, SeekGEWithLimit may return a key beyond limit.
1261 : //
1262 : // If the Iterator is configured to iterate over range keys, SeekGEWithLimit
1263 : // guarantees it will surface any range keys with bounds overlapping the
1264 : // keyspace [key, limit).
1265 2 : func (i *Iterator) SeekGEWithLimit(key []byte, limit []byte) IterValidityState {
1266 2 : if i.rangeKey != nil {
1267 2 : // NB: Check Valid() before clearing requiresReposition.
1268 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1269 2 : // If we have a range key but did not expose it at the previous iterator
1270 2 : // position (because the iterator was not at a valid position), updated
1271 2 : // must be true. This ensures that after an iterator op sequence like:
1272 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1273 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1274 2 : // - SeekGE(...) → (IterValid, RangeBounds() = [a,b))
1275 2 : // the iterator returns RangeKeyChanged()=true.
1276 2 : //
1277 2 : // The remainder of this function will only update i.rangeKey.updated if
1278 2 : // the iterator moves into a new range key, or out of the current range
1279 2 : // key.
1280 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1281 2 : }
1282 2 : lastPositioningOp := i.lastPositioningOp
1283 2 : // Set it to unknown, since this operation may not succeed, in which case
1284 2 : // the SeekGE following this should not make any assumption about iterator
1285 2 : // position.
1286 2 : i.lastPositioningOp = unknownLastPositionOp
1287 2 : i.requiresReposition = false
1288 2 : i.err = nil // clear cached iteration error
1289 2 : i.hasPrefix = false
1290 2 : i.stats.ForwardSeekCount[InterfaceCall]++
1291 2 : if lowerBound := i.opts.GetLowerBound(); lowerBound != nil && i.cmp(key, lowerBound) < 0 {
1292 2 : key = lowerBound
1293 2 : } else if upperBound := i.opts.GetUpperBound(); upperBound != nil && i.cmp(key, upperBound) > 0 {
1294 2 : key = upperBound
1295 2 : }
1296 2 : seekInternalIter := true
1297 2 :
1298 2 : var flags base.SeekGEFlags
1299 2 : if i.batchJustRefreshed {
1300 1 : i.batchJustRefreshed = false
1301 1 : flags = flags.EnableBatchJustRefreshed()
1302 1 : }
1303 2 : if lastPositioningOp == seekGELastPositioningOp {
1304 2 : cmp := i.cmp(i.prefixOrFullSeekKey, key)
1305 2 : // If this seek is to the same or later key, and the iterator is
1306 2 : // already positioned there, this is a noop. This can be helpful for
1307 2 : // sparse key spaces that have many deleted keys, where one can avoid
1308 2 : // the overhead of iterating past them again and again.
1309 2 : if cmp <= 0 {
1310 2 : if !flags.BatchJustRefreshed() &&
1311 2 : (i.iterValidityState == IterExhausted ||
1312 2 : (i.iterValidityState == IterValid && i.cmp(key, i.key) <= 0 &&
1313 2 : (limit == nil || i.cmp(i.key, limit) < 0))) {
1314 2 : // Noop
1315 2 : if i.forceEnableSeekOpt || !testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) {
1316 2 : i.lastPositioningOp = seekGELastPositioningOp
1317 2 : return i.iterValidityState
1318 2 : }
1319 : }
1320 : // cmp == 0 is not safe to optimize since
1321 : // - i.pos could be at iterPosNext, due to a merge.
1322 : // - Even if i.pos were at iterPosCurForward, we could have a DELETE,
1323 : // SET pair for a key, and the iterator would have moved past DELETE
1324 : // but stayed at iterPosCurForward. A similar situation occurs for a
1325 : // MERGE, SET pair where the MERGE is consumed and the iterator is
1326 : // at the SET.
1327 : // We also leverage the IterAtLimit <=> i.pos invariant defined in the
1328 : // comment on iterValidityState, to exclude any cases where i.pos
1329 : // is iterPosCur{Forward,Reverse}Paused. This avoids the need to
1330 : // special-case those iterator positions and their interactions with
1331 : // TrySeekUsingNext, as the main uses for TrySeekUsingNext in CockroachDB
1332 : // do not use limited Seeks in the first place.
1333 2 : if cmp < 0 && i.iterValidityState != IterAtLimit && limit == nil {
1334 2 : flags = flags.EnableTrySeekUsingNext()
1335 2 : }
1336 2 : if testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) && !i.forceEnableSeekOpt {
1337 2 : flags = flags.DisableTrySeekUsingNext()
1338 2 : }
1339 2 : if !flags.BatchJustRefreshed() && i.pos == iterPosCurForwardPaused && i.cmp(key, i.iterKV.K.UserKey) <= 0 {
1340 2 : // Have some work to do, but don't need to seek, and we can
1341 2 : // start doing findNextEntry from i.iterKey.
1342 2 : seekInternalIter = false
1343 2 : }
1344 : }
1345 : }
1346 2 : if seekInternalIter {
1347 2 : i.iterKV = i.iter.SeekGE(key, flags)
1348 2 : i.stats.ForwardSeekCount[InternalIterCall]++
1349 2 : if err := i.iter.Error(); err != nil {
1350 1 : i.err = err
1351 1 : i.iterValidityState = IterExhausted
1352 1 : return i.iterValidityState
1353 1 : }
1354 : }
1355 2 : i.findNextEntry(limit)
1356 2 : i.maybeSampleRead()
1357 2 : if i.Error() == nil {
1358 2 : // Prepare state for a future noop optimization.
1359 2 : i.prefixOrFullSeekKey = append(i.prefixOrFullSeekKey[:0], key...)
1360 2 : i.lastPositioningOp = seekGELastPositioningOp
1361 2 : }
1362 2 : return i.iterValidityState
1363 : }
1364 :
1365 : // SeekPrefixGE moves the iterator to the first key/value pair whose key is
1366 : // greater than or equal to the given key and which has the same "prefix" as
1367 : // the given key. The prefix for a key is determined by the user-defined
1368 : // Comparer.Split function. The iterator will not observe keys not matching the
1369 : // "prefix" of the search key. Calling SeekPrefixGE puts the iterator in prefix
1370 : // iteration mode. The iterator remains in prefix iteration until a subsequent
1371 : // call to another absolute positioning method (SeekGE, SeekLT, First,
1372 : // Last). Reverse iteration (Prev) is not supported when an iterator is in
1373 : // prefix iteration mode. Returns true if the iterator is pointing at a valid
1374 : // entry and false otherwise.
1375 : //
1376 : // The semantics of SeekPrefixGE are slightly unusual and designed for
1377 : // iteration to be able to take advantage of bloom filters that have been
1378 : // created on the "prefix". If you're not using bloom filters, there is no
1379 : // reason to use SeekPrefixGE.
1380 : //
1381 : // An example Split function may separate a timestamp suffix from the prefix of
1382 : // the key.
1383 : //
1384 : // Split(<key>@<timestamp>) -> <key>
1385 : //
1386 : // Consider the keys "a@1", "a@2", "aa@3", "aa@4". The prefixes for these keys
1387 : // are "a", and "aa". Note that despite "a" and "aa" sharing a prefix by the
1388 : // usual definition, those prefixes differ by the definition of the Split
1389 : // function. To see how this works, consider the following set of calls on this
1390 : // data set:
1391 : //
1392 : // SeekPrefixGE("a@0") -> "a@1"
1393 : // Next() -> "a@2"
1394 : // Next() -> EOF
1395 : //
1396 : // If you're just looking to iterate over keys with a shared prefix, as
1397 : // defined by the configured comparer, set iterator bounds instead:
1398 : //
1399 : // iter := db.NewIter(&pebble.IterOptions{
1400 : // LowerBound: []byte("prefix"),
1401 : // UpperBound: []byte("prefiy"),
1402 : // })
1403 : // for iter.First(); iter.Valid(); iter.Next() {
1404 : // // Only keys beginning with "prefix" will be visited.
1405 : // }
1406 : //
1407 : // See ExampleIterator_SeekPrefixGE for a working example.
1408 : //
1409 : // When iterating with range keys enabled, all range keys encountered are
1410 : // truncated to the seek key's prefix's bounds. The truncation of the upper
1411 : // bound requires that the database's Comparer is configured with a
1412 : // ImmediateSuccessor method. For example, a SeekPrefixGE("a@9") call with the
1413 : // prefix "a" will truncate range key bounds to [a,ImmediateSuccessor(a)].
1414 2 : func (i *Iterator) SeekPrefixGE(key []byte) bool {
1415 2 : if i.rangeKey != nil {
1416 2 : // NB: Check Valid() before clearing requiresReposition.
1417 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1418 2 : // If we have a range key but did not expose it at the previous iterator
1419 2 : // position (because the iterator was not at a valid position), updated
1420 2 : // must be true. This ensures that after an iterator op sequence like:
1421 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1422 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1423 2 : // - SeekPrefixGE(...) → (IterValid, RangeBounds() = [a,b))
1424 2 : // the iterator returns RangeKeyChanged()=true.
1425 2 : //
1426 2 : // The remainder of this function will only update i.rangeKey.updated if
1427 2 : // the iterator moves into a new range key, or out of the current range
1428 2 : // key.
1429 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1430 2 : }
1431 2 : lastPositioningOp := i.lastPositioningOp
1432 2 : // Set it to unknown, since this operation may not succeed, in which case
1433 2 : // the SeekPrefixGE following this should not make any assumption about
1434 2 : // iterator position.
1435 2 : i.lastPositioningOp = unknownLastPositionOp
1436 2 : i.requiresReposition = false
1437 2 : i.err = nil // clear cached iteration error
1438 2 : i.stats.ForwardSeekCount[InterfaceCall]++
1439 2 : if i.comparer.ImmediateSuccessor == nil && i.opts.KeyTypes != IterKeyTypePointsOnly {
1440 0 : panic("pebble: ImmediateSuccessor must be provided for SeekPrefixGE with range keys")
1441 : }
1442 2 : prefixLen := i.comparer.Split(key)
1443 2 : keyPrefix := key[:prefixLen]
1444 2 : var flags base.SeekGEFlags
1445 2 : if i.batchJustRefreshed {
1446 1 : flags = flags.EnableBatchJustRefreshed()
1447 1 : i.batchJustRefreshed = false
1448 1 : }
1449 2 : if lastPositioningOp == seekPrefixGELastPositioningOp {
1450 2 : if !i.hasPrefix {
1451 0 : panic("lastPositioningOpsIsSeekPrefixGE is true, but hasPrefix is false")
1452 : }
1453 : // The iterator has not been repositioned after the last SeekPrefixGE.
1454 : // See if we are seeking to a larger key, since then we can optimize
1455 : // the seek by using next. Note that we could also optimize if Next
1456 : // has been called, if the iterator is not exhausted and the current
1457 : // position is <= the seek key. We are keeping this limited for now
1458 : // since such optimizations require care for correctness, and to not
1459 : // become de-optimizations (if one usually has to do all the next
1460 : // calls and then the seek). This SeekPrefixGE optimization
1461 : // specifically benefits CockroachDB.
1462 2 : cmp := i.cmp(i.prefixOrFullSeekKey, keyPrefix)
1463 2 : // cmp == 0 is not safe to optimize since
1464 2 : // - i.pos could be at iterPosNext, due to a merge.
1465 2 : // - Even if i.pos were at iterPosCurForward, we could have a DELETE,
1466 2 : // SET pair for a key, and the iterator would have moved past DELETE
1467 2 : // but stayed at iterPosCurForward. A similar situation occurs for a
1468 2 : // MERGE, SET pair where the MERGE is consumed and the iterator is
1469 2 : // at the SET.
1470 2 : // In general some versions of i.prefix could have been consumed by
1471 2 : // the iterator, so we only optimize for cmp < 0.
1472 2 : if cmp < 0 {
1473 2 : flags = flags.EnableTrySeekUsingNext()
1474 2 : }
1475 2 : if testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) && !i.forceEnableSeekOpt {
1476 2 : flags = flags.DisableTrySeekUsingNext()
1477 2 : }
1478 : }
1479 : // Make a copy of the prefix so that modifications to the key after
1480 : // SeekPrefixGE returns does not affect the stored prefix.
1481 2 : if cap(i.prefixOrFullSeekKey) < prefixLen {
1482 2 : i.prefixOrFullSeekKey = make([]byte, prefixLen)
1483 2 : } else {
1484 2 : i.prefixOrFullSeekKey = i.prefixOrFullSeekKey[:prefixLen]
1485 2 : }
1486 2 : i.hasPrefix = true
1487 2 : copy(i.prefixOrFullSeekKey, keyPrefix)
1488 2 :
1489 2 : if lowerBound := i.opts.GetLowerBound(); lowerBound != nil && i.cmp(key, lowerBound) < 0 {
1490 2 : if p := i.comparer.Split.Prefix(lowerBound); !bytes.Equal(i.prefixOrFullSeekKey, p) {
1491 2 : i.err = errors.New("pebble: SeekPrefixGE supplied with key outside of lower bound")
1492 2 : i.iterValidityState = IterExhausted
1493 2 : return false
1494 2 : }
1495 2 : key = lowerBound
1496 2 : } else if upperBound := i.opts.GetUpperBound(); upperBound != nil && i.cmp(key, upperBound) > 0 {
1497 2 : if p := i.comparer.Split.Prefix(upperBound); !bytes.Equal(i.prefixOrFullSeekKey, p) {
1498 2 : i.err = errors.New("pebble: SeekPrefixGE supplied with key outside of upper bound")
1499 2 : i.iterValidityState = IterExhausted
1500 2 : return false
1501 2 : }
1502 2 : key = upperBound
1503 : }
1504 2 : i.iterKV = i.iter.SeekPrefixGE(i.prefixOrFullSeekKey, key, flags)
1505 2 : i.stats.ForwardSeekCount[InternalIterCall]++
1506 2 : i.findNextEntry(nil)
1507 2 : i.maybeSampleRead()
1508 2 : if i.Error() == nil {
1509 2 : i.lastPositioningOp = seekPrefixGELastPositioningOp
1510 2 : }
1511 2 : return i.iterValidityState == IterValid
1512 : }
1513 :
1514 : // Deterministic disabling (in testing mode) of the seek optimizations. It uses
1515 : // the iterator pointer, since we want diversity in iterator behavior for the
1516 : // same key. Used for tests.
1517 2 : func testingDisableSeekOpt(key []byte, ptr uintptr) bool {
1518 2 : if !invariants.Enabled {
1519 0 : return false
1520 0 : }
1521 : // Fibonacci hash https://probablydance.com/2018/06/16/fibonacci-hashing-the-optimization-that-the-world-forgot-or-a-better-alternative-to-integer-modulo/
1522 2 : simpleHash := (11400714819323198485 * uint64(ptr)) >> 63
1523 2 : return key != nil && key[0]&byte(1) == 0 && simpleHash == 0
1524 : }
1525 :
1526 : // SeekLT moves the iterator to the last key/value pair whose key is less than
1527 : // the given key. Returns true if the iterator is pointing at a valid entry and
1528 : // false otherwise.
1529 2 : func (i *Iterator) SeekLT(key []byte) bool {
1530 2 : return i.SeekLTWithLimit(key, nil) == IterValid
1531 2 : }
1532 :
1533 : // SeekLTWithLimit moves the iterator to the last key/value pair whose key is
1534 : // less than the given key.
1535 : //
1536 : // If limit is provided, it serves as a best-effort inclusive limit. If the last
1537 : // key less than the given search key is also less than limit, the Iterator may
1538 : // pause and return IterAtLimit. Because limits are best-effort, SeekLTWithLimit
1539 : // may return a key beyond limit.
1540 : //
1541 : // If the Iterator is configured to iterate over range keys, SeekLTWithLimit
1542 : // guarantees it will surface any range keys with bounds overlapping the
1543 : // keyspace up to limit.
1544 2 : func (i *Iterator) SeekLTWithLimit(key []byte, limit []byte) IterValidityState {
1545 2 : if i.rangeKey != nil {
1546 2 : // NB: Check Valid() before clearing requiresReposition.
1547 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1548 2 : // If we have a range key but did not expose it at the previous iterator
1549 2 : // position (because the iterator was not at a valid position), updated
1550 2 : // must be true. This ensures that after an iterator op sequence like:
1551 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1552 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1553 2 : // - SeekLTWithLimit(...) → (IterValid, RangeBounds() = [a,b))
1554 2 : // the iterator returns RangeKeyChanged()=true.
1555 2 : //
1556 2 : // The remainder of this function will only update i.rangeKey.updated if
1557 2 : // the iterator moves into a new range key, or out of the current range
1558 2 : // key.
1559 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1560 2 : }
1561 2 : lastPositioningOp := i.lastPositioningOp
1562 2 : // Set it to unknown, since this operation may not succeed, in which case
1563 2 : // the SeekLT following this should not make any assumption about iterator
1564 2 : // position.
1565 2 : i.lastPositioningOp = unknownLastPositionOp
1566 2 : i.batchJustRefreshed = false
1567 2 : i.requiresReposition = false
1568 2 : i.err = nil // clear cached iteration error
1569 2 : i.stats.ReverseSeekCount[InterfaceCall]++
1570 2 : if upperBound := i.opts.GetUpperBound(); upperBound != nil && i.cmp(key, upperBound) > 0 {
1571 2 : key = upperBound
1572 2 : } else if lowerBound := i.opts.GetLowerBound(); lowerBound != nil && i.cmp(key, lowerBound) < 0 {
1573 2 : key = lowerBound
1574 2 : }
1575 2 : i.hasPrefix = false
1576 2 : seekInternalIter := true
1577 2 : // The following noop optimization only applies when i.batch == nil, since
1578 2 : // an iterator over a batch is iterating over mutable data, that may have
1579 2 : // changed since the last seek.
1580 2 : if lastPositioningOp == seekLTLastPositioningOp && i.batch == nil {
1581 2 : cmp := i.cmp(key, i.prefixOrFullSeekKey)
1582 2 : // If this seek is to the same or earlier key, and the iterator is
1583 2 : // already positioned there, this is a noop. This can be helpful for
1584 2 : // sparse key spaces that have many deleted keys, where one can avoid
1585 2 : // the overhead of iterating past them again and again.
1586 2 : if cmp <= 0 {
1587 2 : // NB: when pos != iterPosCurReversePaused, the invariant
1588 2 : // documented earlier implies that iterValidityState !=
1589 2 : // IterAtLimit.
1590 2 : if i.iterValidityState == IterExhausted ||
1591 2 : (i.iterValidityState == IterValid && i.cmp(i.key, key) < 0 &&
1592 2 : (limit == nil || i.cmp(limit, i.key) <= 0)) {
1593 2 : if !testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) {
1594 2 : i.lastPositioningOp = seekLTLastPositioningOp
1595 2 : return i.iterValidityState
1596 2 : }
1597 : }
1598 2 : if i.pos == iterPosCurReversePaused && i.cmp(i.iterKV.K.UserKey, key) < 0 {
1599 2 : // Have some work to do, but don't need to seek, and we can
1600 2 : // start doing findPrevEntry from i.iterKey.
1601 2 : seekInternalIter = false
1602 2 : }
1603 : }
1604 : }
1605 2 : if seekInternalIter {
1606 2 : i.iterKV = i.iter.SeekLT(key, base.SeekLTFlagsNone)
1607 2 : i.stats.ReverseSeekCount[InternalIterCall]++
1608 2 : if err := i.iter.Error(); err != nil {
1609 1 : i.err = err
1610 1 : i.iterValidityState = IterExhausted
1611 1 : return i.iterValidityState
1612 1 : }
1613 : }
1614 2 : i.findPrevEntry(limit)
1615 2 : i.maybeSampleRead()
1616 2 : if i.Error() == nil && i.batch == nil {
1617 2 : // Prepare state for a future noop optimization.
1618 2 : i.prefixOrFullSeekKey = append(i.prefixOrFullSeekKey[:0], key...)
1619 2 : i.lastPositioningOp = seekLTLastPositioningOp
1620 2 : }
1621 2 : return i.iterValidityState
1622 : }
1623 :
1624 : // First moves the iterator the first key/value pair. Returns true if the
1625 : // iterator is pointing at a valid entry and false otherwise.
1626 2 : func (i *Iterator) First() bool {
1627 2 : if i.rangeKey != nil {
1628 2 : // NB: Check Valid() before clearing requiresReposition.
1629 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1630 2 : // If we have a range key but did not expose it at the previous iterator
1631 2 : // position (because the iterator was not at a valid position), updated
1632 2 : // must be true. This ensures that after an iterator op sequence like:
1633 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1634 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1635 2 : // - First(...) → (IterValid, RangeBounds() = [a,b))
1636 2 : // the iterator returns RangeKeyChanged()=true.
1637 2 : //
1638 2 : // The remainder of this function will only update i.rangeKey.updated if
1639 2 : // the iterator moves into a new range key, or out of the current range
1640 2 : // key.
1641 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1642 2 : }
1643 2 : i.err = nil // clear cached iteration error
1644 2 : i.hasPrefix = false
1645 2 : i.batchJustRefreshed = false
1646 2 : i.lastPositioningOp = unknownLastPositionOp
1647 2 : i.requiresReposition = false
1648 2 : i.stats.ForwardSeekCount[InterfaceCall]++
1649 2 :
1650 2 : i.err = i.iterFirstWithinBounds()
1651 2 : if i.err != nil {
1652 1 : i.iterValidityState = IterExhausted
1653 1 : return false
1654 1 : }
1655 2 : i.findNextEntry(nil)
1656 2 : i.maybeSampleRead()
1657 2 : return i.iterValidityState == IterValid
1658 : }
1659 :
1660 : // Last moves the iterator the last key/value pair. Returns true if the
1661 : // iterator is pointing at a valid entry and false otherwise.
1662 2 : func (i *Iterator) Last() bool {
1663 2 : if i.rangeKey != nil {
1664 2 : // NB: Check Valid() before clearing requiresReposition.
1665 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1666 2 : // If we have a range key but did not expose it at the previous iterator
1667 2 : // position (because the iterator was not at a valid position), updated
1668 2 : // must be true. This ensures that after an iterator op sequence like:
1669 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1670 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1671 2 : // - Last(...) → (IterValid, RangeBounds() = [a,b))
1672 2 : // the iterator returns RangeKeyChanged()=true.
1673 2 : //
1674 2 : // The remainder of this function will only update i.rangeKey.updated if
1675 2 : // the iterator moves into a new range key, or out of the current range
1676 2 : // key.
1677 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1678 2 : }
1679 2 : i.err = nil // clear cached iteration error
1680 2 : i.hasPrefix = false
1681 2 : i.batchJustRefreshed = false
1682 2 : i.lastPositioningOp = unknownLastPositionOp
1683 2 : i.requiresReposition = false
1684 2 : i.stats.ReverseSeekCount[InterfaceCall]++
1685 2 :
1686 2 : if i.err = i.iterLastWithinBounds(); i.err != nil {
1687 1 : i.iterValidityState = IterExhausted
1688 1 : return false
1689 1 : }
1690 2 : i.findPrevEntry(nil)
1691 2 : i.maybeSampleRead()
1692 2 : return i.iterValidityState == IterValid
1693 : }
1694 :
1695 : // Next moves the iterator to the next key/value pair. Returns true if the
1696 : // iterator is pointing at a valid entry and false otherwise.
1697 2 : func (i *Iterator) Next() bool {
1698 2 : return i.nextWithLimit(nil) == IterValid
1699 2 : }
1700 :
1701 : // NextWithLimit moves the iterator to the next key/value pair.
1702 : //
1703 : // If limit is provided, it serves as a best-effort exclusive limit. If the next
1704 : // key is greater than or equal to limit, the Iterator may pause and return
1705 : // IterAtLimit. Because limits are best-effort, NextWithLimit may return a key
1706 : // beyond limit.
1707 : //
1708 : // If the Iterator is configured to iterate over range keys, NextWithLimit
1709 : // guarantees it will surface any range keys with bounds overlapping the
1710 : // keyspace up to limit.
1711 2 : func (i *Iterator) NextWithLimit(limit []byte) IterValidityState {
1712 2 : return i.nextWithLimit(limit)
1713 2 : }
1714 :
1715 : // NextPrefix moves the iterator to the next key/value pair with a key
1716 : // containing a different prefix than the current key. Prefixes are determined
1717 : // by Comparer.Split. Exhausts the iterator if invoked while in prefix-iteration
1718 : // mode.
1719 : //
1720 : // It is not permitted to invoke NextPrefix while at a IterAtLimit position.
1721 : // When called in this condition, NextPrefix has non-deterministic behavior.
1722 : //
1723 : // It is not permitted to invoke NextPrefix when the Iterator has an
1724 : // upper-bound that is a versioned MVCC key (see the comment for
1725 : // Comparer.Split). It returns an error in this case.
1726 2 : func (i *Iterator) NextPrefix() bool {
1727 2 : if i.nextPrefixNotPermittedByUpperBound {
1728 2 : i.lastPositioningOp = unknownLastPositionOp
1729 2 : i.requiresReposition = false
1730 2 : i.err = errors.Errorf("NextPrefix not permitted with upper bound %s",
1731 2 : i.comparer.FormatKey(i.opts.UpperBound))
1732 2 : i.iterValidityState = IterExhausted
1733 2 : return false
1734 2 : }
1735 2 : if i.hasPrefix {
1736 2 : i.iterValidityState = IterExhausted
1737 2 : return false
1738 2 : }
1739 2 : if i.Error() != nil {
1740 2 : return false
1741 2 : }
1742 2 : return i.nextPrefix() == IterValid
1743 : }
1744 :
1745 2 : func (i *Iterator) nextPrefix() IterValidityState {
1746 2 : if i.rangeKey != nil {
1747 2 : // NB: Check Valid() before clearing requiresReposition.
1748 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1749 2 : // If we have a range key but did not expose it at the previous iterator
1750 2 : // position (because the iterator was not at a valid position), updated
1751 2 : // must be true. This ensures that after an iterator op sequence like:
1752 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1753 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1754 2 : // - NextWithLimit(...) → (IterValid, RangeBounds() = [a,b))
1755 2 : // the iterator returns RangeKeyChanged()=true.
1756 2 : //
1757 2 : // The remainder of this function will only update i.rangeKey.updated if
1758 2 : // the iterator moves into a new range key, or out of the current range
1759 2 : // key.
1760 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1761 2 : }
1762 :
1763 : // Although NextPrefix documents that behavior at IterAtLimit is undefined,
1764 : // this function handles these cases as a simple prefix-agnostic Next. This
1765 : // is done for deterministic behavior in the metamorphic tests.
1766 : //
1767 : // TODO(jackson): If the metamorphic test operation generator is adjusted to
1768 : // make generation of some operations conditional on the previous
1769 : // operations, then we can remove this behavior and explicitly error.
1770 :
1771 2 : i.lastPositioningOp = unknownLastPositionOp
1772 2 : i.requiresReposition = false
1773 2 : switch i.pos {
1774 2 : case iterPosCurForward:
1775 2 : // Positioned on the current key. Advance to the next prefix.
1776 2 : i.internalNextPrefix(i.comparer.Split(i.key))
1777 2 : case iterPosCurForwardPaused:
1778 : // Positioned at a limit. Implement as a prefix-agnostic Next. See TODO
1779 : // up above. The iterator is already positioned at the next key.
1780 2 : case iterPosCurReverse:
1781 2 : // Switching directions.
1782 2 : // Unless the iterator was exhausted, reverse iteration needs to
1783 2 : // position the iterator at iterPosPrev.
1784 2 : if i.iterKV != nil {
1785 0 : i.err = errors.New("switching from reverse to forward but iter is not at prev")
1786 0 : i.iterValidityState = IterExhausted
1787 0 : return i.iterValidityState
1788 0 : }
1789 : // The Iterator is exhausted and i.iter is positioned before the first
1790 : // key. Reposition to point to the first internal key.
1791 2 : if i.err = i.iterFirstWithinBounds(); i.err != nil {
1792 0 : i.iterValidityState = IterExhausted
1793 0 : return i.iterValidityState
1794 0 : }
1795 2 : case iterPosCurReversePaused:
1796 2 : // Positioned at a limit. Implement as a prefix-agnostic Next. See TODO
1797 2 : // up above.
1798 2 : //
1799 2 : // Switching directions; The iterator must not be exhausted since it
1800 2 : // paused.
1801 2 : if i.iterKV == nil {
1802 0 : i.err = errors.New("switching paused from reverse to forward but iter is exhausted")
1803 0 : i.iterValidityState = IterExhausted
1804 0 : return i.iterValidityState
1805 0 : }
1806 2 : i.nextUserKey()
1807 2 : case iterPosPrev:
1808 2 : // The underlying iterator is pointed to the previous key (this can
1809 2 : // only happen when switching iteration directions).
1810 2 : if i.iterKV == nil {
1811 2 : // We're positioned before the first key. Need to reposition to point to
1812 2 : // the first key.
1813 2 : i.err = i.iterFirstWithinBounds()
1814 2 : if i.iterKV == nil {
1815 0 : i.iterValidityState = IterExhausted
1816 0 : return i.iterValidityState
1817 0 : }
1818 2 : if invariants.Enabled && !i.equal(i.iterKV.K.UserKey, i.key) {
1819 0 : i.opts.getLogger().Fatalf("pebble: invariant violation: First internal iterator from iterPosPrev landed on %q, not %q",
1820 0 : i.iterKV.K.UserKey, i.key)
1821 0 : }
1822 2 : } else {
1823 2 : // Move the internal iterator back onto the user key stored in
1824 2 : // i.key. iterPosPrev guarantees that it's positioned at the last
1825 2 : // key with the user key less than i.key, so we're guaranteed to
1826 2 : // land on the correct key with a single Next.
1827 2 : i.iterKV = i.iter.Next()
1828 2 : if i.iterKV == nil {
1829 1 : // This should only be possible if i.iter.Next() encountered an
1830 1 : // error.
1831 1 : if i.iter.Error() == nil {
1832 0 : i.opts.getLogger().Fatalf("pebble: invariant violation: Nexting internal iterator from iterPosPrev found nothing")
1833 0 : }
1834 : // NB: Iterator.Error() will return i.iter.Error().
1835 1 : i.iterValidityState = IterExhausted
1836 1 : return i.iterValidityState
1837 : }
1838 2 : if invariants.Enabled && !i.equal(i.iterKV.K.UserKey, i.key) {
1839 0 : i.opts.getLogger().Fatalf("pebble: invariant violation: Nexting internal iterator from iterPosPrev landed on %q, not %q",
1840 0 : i.iterKV.K.UserKey, i.key)
1841 0 : }
1842 : }
1843 : // The internal iterator is now positioned at i.key. Advance to the next
1844 : // prefix.
1845 2 : i.internalNextPrefix(i.comparer.Split(i.key))
1846 2 : case iterPosNext:
1847 2 : // Already positioned on the next key. Only call nextPrefixKey if the
1848 2 : // next key shares the same prefix.
1849 2 : if i.iterKV != nil {
1850 2 : currKeyPrefixLen := i.comparer.Split(i.key)
1851 2 : if bytes.Equal(i.comparer.Split.Prefix(i.iterKV.K.UserKey), i.key[:currKeyPrefixLen]) {
1852 2 : i.internalNextPrefix(currKeyPrefixLen)
1853 2 : }
1854 : }
1855 : }
1856 :
1857 2 : i.stats.ForwardStepCount[InterfaceCall]++
1858 2 : i.findNextEntry(nil /* limit */)
1859 2 : i.maybeSampleRead()
1860 2 : return i.iterValidityState
1861 : }
1862 :
1863 2 : func (i *Iterator) internalNextPrefix(currKeyPrefixLen int) {
1864 2 : if i.iterKV == nil {
1865 2 : return
1866 2 : }
1867 : // The Next "fast-path" is not really a fast-path when there is more than
1868 : // one version. However, even with TableFormatPebblev3, there is a small
1869 : // slowdown (~10%) for one version if we remove it and only call NextPrefix.
1870 : // When there are two versions, only calling NextPrefix is ~30% faster.
1871 2 : i.stats.ForwardStepCount[InternalIterCall]++
1872 2 : if i.iterKV = i.iter.Next(); i.iterKV == nil {
1873 2 : return
1874 2 : }
1875 2 : if !bytes.Equal(i.comparer.Split.Prefix(i.iterKV.K.UserKey), i.key[:currKeyPrefixLen]) {
1876 2 : return
1877 2 : }
1878 2 : i.stats.ForwardStepCount[InternalIterCall]++
1879 2 : i.prefixOrFullSeekKey = i.comparer.ImmediateSuccessor(i.prefixOrFullSeekKey[:0], i.key[:currKeyPrefixLen])
1880 2 : if i.iterKV.K.IsExclusiveSentinel() {
1881 0 : panic(errors.AssertionFailedf("pebble: unexpected exclusive sentinel key: %q", i.iterKV.K))
1882 : }
1883 :
1884 2 : i.iterKV = i.iter.NextPrefix(i.prefixOrFullSeekKey)
1885 2 : if invariants.Enabled && i.iterKV != nil {
1886 2 : if p := i.comparer.Split.Prefix(i.iterKV.K.UserKey); i.cmp(p, i.prefixOrFullSeekKey) < 0 {
1887 0 : panic(errors.AssertionFailedf("pebble: iter.NextPrefix did not advance beyond the current prefix: now at %q; expected to be geq %q",
1888 0 : i.iterKV.K, i.prefixOrFullSeekKey))
1889 : }
1890 : }
1891 : }
1892 :
1893 2 : func (i *Iterator) nextWithLimit(limit []byte) IterValidityState {
1894 2 : i.stats.ForwardStepCount[InterfaceCall]++
1895 2 : if i.hasPrefix {
1896 2 : if limit != nil {
1897 2 : i.err = errors.New("cannot use limit with prefix iteration")
1898 2 : i.iterValidityState = IterExhausted
1899 2 : return i.iterValidityState
1900 2 : } else if i.iterValidityState == IterExhausted {
1901 2 : // No-op, already exhausted. We avoid executing the Next because it
1902 2 : // can break invariants: Specifically, a file that fails the bloom
1903 2 : // filter test may result in its level being removed from the
1904 2 : // merging iterator. The level's removal can cause a lazy combined
1905 2 : // iterator to miss range keys and trigger a switch to combined
1906 2 : // iteration at a larger key, breaking keyspan invariants.
1907 2 : return i.iterValidityState
1908 2 : }
1909 : }
1910 2 : if i.err != nil {
1911 2 : return i.iterValidityState
1912 2 : }
1913 2 : if i.rangeKey != nil {
1914 2 : // NB: Check Valid() before clearing requiresReposition.
1915 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1916 2 : // If we have a range key but did not expose it at the previous iterator
1917 2 : // position (because the iterator was not at a valid position), updated
1918 2 : // must be true. This ensures that after an iterator op sequence like:
1919 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
1920 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1921 2 : // - NextWithLimit(...) → (IterValid, RangeBounds() = [a,b))
1922 2 : // the iterator returns RangeKeyChanged()=true.
1923 2 : //
1924 2 : // The remainder of this function will only update i.rangeKey.updated if
1925 2 : // the iterator moves into a new range key, or out of the current range
1926 2 : // key.
1927 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1928 2 : }
1929 2 : i.lastPositioningOp = unknownLastPositionOp
1930 2 : i.requiresReposition = false
1931 2 : switch i.pos {
1932 2 : case iterPosCurForward:
1933 2 : i.nextUserKey()
1934 2 : case iterPosCurForwardPaused:
1935 : // Already at the right place.
1936 2 : case iterPosCurReverse:
1937 2 : // Switching directions.
1938 2 : // Unless the iterator was exhausted, reverse iteration needs to
1939 2 : // position the iterator at iterPosPrev.
1940 2 : if i.iterKV != nil {
1941 0 : i.err = errors.New("switching from reverse to forward but iter is not at prev")
1942 0 : i.iterValidityState = IterExhausted
1943 0 : return i.iterValidityState
1944 0 : }
1945 : // We're positioned before the first key. Need to reposition to point to
1946 : // the first key.
1947 2 : if i.err = i.iterFirstWithinBounds(); i.err != nil {
1948 1 : i.iterValidityState = IterExhausted
1949 1 : return i.iterValidityState
1950 1 : }
1951 2 : case iterPosCurReversePaused:
1952 2 : // Switching directions.
1953 2 : // The iterator must not be exhausted since it paused.
1954 2 : if i.iterKV == nil {
1955 0 : i.err = errors.New("switching paused from reverse to forward but iter is exhausted")
1956 0 : i.iterValidityState = IterExhausted
1957 0 : return i.iterValidityState
1958 0 : }
1959 2 : i.nextUserKey()
1960 2 : case iterPosPrev:
1961 2 : // The underlying iterator is pointed to the previous key (this can
1962 2 : // only happen when switching iteration directions). We set
1963 2 : // i.iterValidityState to IterExhausted here to force the calls to
1964 2 : // nextUserKey to save the current key i.iter is pointing at in order
1965 2 : // to determine when the next user-key is reached.
1966 2 : i.iterValidityState = IterExhausted
1967 2 : if i.iterKV == nil {
1968 2 : // We're positioned before the first key. Need to reposition to point to
1969 2 : // the first key.
1970 2 : i.err = i.iterFirstWithinBounds()
1971 2 : } else {
1972 2 : i.nextUserKey()
1973 2 : }
1974 2 : if i.err != nil {
1975 1 : i.iterValidityState = IterExhausted
1976 1 : return i.iterValidityState
1977 1 : }
1978 2 : i.nextUserKey()
1979 2 : case iterPosNext:
1980 : // Already at the right place.
1981 : }
1982 2 : i.findNextEntry(limit)
1983 2 : i.maybeSampleRead()
1984 2 : return i.iterValidityState
1985 : }
1986 :
1987 : // Prev moves the iterator to the previous key/value pair. Returns true if the
1988 : // iterator is pointing at a valid entry and false otherwise.
1989 2 : func (i *Iterator) Prev() bool {
1990 2 : return i.PrevWithLimit(nil) == IterValid
1991 2 : }
1992 :
1993 : // PrevWithLimit moves the iterator to the previous key/value pair.
1994 : //
1995 : // If limit is provided, it serves as a best-effort inclusive limit. If the
1996 : // previous key is less than limit, the Iterator may pause and return
1997 : // IterAtLimit. Because limits are best-effort, PrevWithLimit may return a key
1998 : // beyond limit.
1999 : //
2000 : // If the Iterator is configured to iterate over range keys, PrevWithLimit
2001 : // guarantees it will surface any range keys with bounds overlapping the
2002 : // keyspace up to limit.
2003 2 : func (i *Iterator) PrevWithLimit(limit []byte) IterValidityState {
2004 2 : i.stats.ReverseStepCount[InterfaceCall]++
2005 2 : if i.err != nil {
2006 2 : return i.iterValidityState
2007 2 : }
2008 2 : if i.rangeKey != nil {
2009 2 : // NB: Check Valid() before clearing requiresReposition.
2010 2 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
2011 2 : // If we have a range key but did not expose it at the previous iterator
2012 2 : // position (because the iterator was not at a valid position), updated
2013 2 : // must be true. This ensures that after an iterator op sequence like:
2014 2 : // - Next() → (IterValid, RangeBounds() = [a,b))
2015 2 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
2016 2 : // - PrevWithLimit(...) → (IterValid, RangeBounds() = [a,b))
2017 2 : // the iterator returns RangeKeyChanged()=true.
2018 2 : //
2019 2 : // The remainder of this function will only update i.rangeKey.updated if
2020 2 : // the iterator moves into a new range key, or out of the current range
2021 2 : // key.
2022 2 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
2023 2 : }
2024 2 : i.lastPositioningOp = unknownLastPositionOp
2025 2 : i.requiresReposition = false
2026 2 : if i.hasPrefix {
2027 2 : i.err = errReversePrefixIteration
2028 2 : i.iterValidityState = IterExhausted
2029 2 : return i.iterValidityState
2030 2 : }
2031 2 : switch i.pos {
2032 2 : case iterPosCurForward:
2033 : // Switching directions, and will handle this below.
2034 2 : case iterPosCurForwardPaused:
2035 : // Switching directions, and will handle this below.
2036 2 : case iterPosCurReverse:
2037 2 : i.prevUserKey()
2038 2 : case iterPosCurReversePaused:
2039 : // Already at the right place.
2040 2 : case iterPosNext:
2041 : // The underlying iterator is pointed to the next key (this can only happen
2042 : // when switching iteration directions). We will handle this below.
2043 2 : case iterPosPrev:
2044 : // Already at the right place.
2045 : }
2046 2 : if i.pos == iterPosCurForward || i.pos == iterPosNext || i.pos == iterPosCurForwardPaused {
2047 2 : // Switching direction.
2048 2 : stepAgain := i.pos == iterPosNext
2049 2 :
2050 2 : // Synthetic range key markers are a special case. Consider SeekGE(b)
2051 2 : // which finds a range key [a, c). To ensure the user observes the range
2052 2 : // key, the Iterator pauses at Key() = b. The iterator must advance the
2053 2 : // internal iterator to see if there's also a coincident point key at
2054 2 : // 'b', leaving the iterator at iterPosNext if there's not.
2055 2 : //
2056 2 : // This is a problem: Synthetic range key markers are only interleaved
2057 2 : // during the original seek. A subsequent Prev() of i.iter will not move
2058 2 : // back onto the synthetic range key marker. In this case where the
2059 2 : // previous iterator position was a synthetic range key start boundary,
2060 2 : // we must not step a second time.
2061 2 : if i.isEphemeralPosition() {
2062 2 : stepAgain = false
2063 2 : }
2064 :
2065 : // We set i.iterValidityState to IterExhausted here to force the calls
2066 : // to prevUserKey to save the current key i.iter is pointing at in
2067 : // order to determine when the prev user-key is reached.
2068 2 : i.iterValidityState = IterExhausted
2069 2 : if i.iterKV == nil {
2070 2 : // We're positioned after the last key. Need to reposition to point to
2071 2 : // the last key.
2072 2 : i.err = i.iterLastWithinBounds()
2073 2 : } else {
2074 2 : i.prevUserKey()
2075 2 : }
2076 2 : if i.err != nil {
2077 1 : return i.iterValidityState
2078 1 : }
2079 2 : if stepAgain {
2080 2 : i.prevUserKey()
2081 2 : if i.err != nil {
2082 0 : return i.iterValidityState
2083 0 : }
2084 : }
2085 : }
2086 2 : i.findPrevEntry(limit)
2087 2 : i.maybeSampleRead()
2088 2 : return i.iterValidityState
2089 : }
2090 :
2091 : // iterFirstWithinBounds moves the internal iterator to the first key,
2092 : // respecting bounds.
2093 2 : func (i *Iterator) iterFirstWithinBounds() error {
2094 2 : i.stats.ForwardSeekCount[InternalIterCall]++
2095 2 : if lowerBound := i.opts.GetLowerBound(); lowerBound != nil {
2096 2 : i.iterKV = i.iter.SeekGE(lowerBound, base.SeekGEFlagsNone)
2097 2 : } else {
2098 2 : i.iterKV = i.iter.First()
2099 2 : }
2100 2 : if i.iterKV == nil {
2101 2 : return i.iter.Error()
2102 2 : }
2103 2 : return nil
2104 : }
2105 :
2106 : // iterLastWithinBounds moves the internal iterator to the last key, respecting
2107 : // bounds.
2108 2 : func (i *Iterator) iterLastWithinBounds() error {
2109 2 : i.stats.ReverseSeekCount[InternalIterCall]++
2110 2 : if upperBound := i.opts.GetUpperBound(); upperBound != nil {
2111 2 : i.iterKV = i.iter.SeekLT(upperBound, base.SeekLTFlagsNone)
2112 2 : } else {
2113 2 : i.iterKV = i.iter.Last()
2114 2 : }
2115 2 : if i.iterKV == nil {
2116 2 : return i.iter.Error()
2117 2 : }
2118 2 : return nil
2119 : }
2120 :
2121 : // RangeKeyData describes a range key's data, set through RangeKeySet. The key
2122 : // boundaries of the range key is provided by Iterator.RangeBounds.
2123 : type RangeKeyData struct {
2124 : Suffix []byte
2125 : Value []byte
2126 : }
2127 :
2128 : // rangeKeyWithinLimit is called during limited reverse iteration when
2129 : // positioned over a key beyond the limit. If there exists a range key that lies
2130 : // within the limit, the iterator must not pause in order to ensure the user has
2131 : // an opportunity to observe the range key within limit.
2132 : //
2133 : // It would be valid to ignore the limit whenever there's a range key covering
2134 : // the key, but that would introduce nondeterminism. To preserve determinism for
2135 : // testing, the iterator ignores the limit only if the covering range key does
2136 : // cover the keyspace within the limit.
2137 : //
2138 : // This awkwardness exists because range keys are interleaved at their inclusive
2139 : // start positions. Note that limit is inclusive.
2140 2 : func (i *Iterator) rangeKeyWithinLimit(limit []byte) bool {
2141 2 : if i.rangeKey == nil || !i.opts.rangeKeys() {
2142 2 : return false
2143 2 : }
2144 2 : s := i.rangeKey.iiter.Span()
2145 2 : // If the range key ends beyond the limit, then the range key does not cover
2146 2 : // any portion of the keyspace within the limit and it is safe to pause.
2147 2 : return s != nil && i.cmp(s.End, limit) > 0
2148 : }
2149 :
2150 : // saveRangeKey saves the current range key to the underlying iterator's current
2151 : // range key state. If the range key has not changed, saveRangeKey is a no-op.
2152 : // If there is a new range key, saveRangeKey copies all of the key, value and
2153 : // suffixes into Iterator-managed buffers.
2154 2 : func (i *Iterator) saveRangeKey() {
2155 2 : if i.rangeKey == nil || i.opts.KeyTypes == IterKeyTypePointsOnly {
2156 2 : return
2157 2 : }
2158 :
2159 2 : s := i.rangeKey.iiter.Span()
2160 2 : if s == nil {
2161 2 : i.rangeKey.hasRangeKey = false
2162 2 : i.rangeKey.updated = i.rangeKey.prevPosHadRangeKey
2163 2 : return
2164 2 : } else if !i.rangeKey.stale {
2165 2 : // The range key `s` is identical to the one currently saved. No-op.
2166 2 : return
2167 2 : }
2168 :
2169 2 : if s.KeysOrder != keyspan.BySuffixAsc {
2170 0 : panic("pebble: range key span's keys unexpectedly not in ascending suffix order")
2171 : }
2172 :
2173 : // Although `i.rangeKey.stale` is true, the span s may still be identical
2174 : // to the currently saved span. This is possible when seeking the iterator,
2175 : // which may land back on the same range key. If we previously had a range
2176 : // key and the new one has an identical start key, then it must be the same
2177 : // range key and we can avoid copying and keep `i.rangeKey.updated=false`.
2178 : //
2179 : // TODO(jackson): These key comparisons could be avoidable during relative
2180 : // positioning operations continuing in the same direction, because these
2181 : // ops will never encounter the previous position's range key while
2182 : // stale=true. However, threading whether the current op is a seek or step
2183 : // maybe isn't worth it. This key comparison is only necessary once when we
2184 : // step onto a new range key, which should be relatively rare.
2185 2 : if i.rangeKey.prevPosHadRangeKey && i.equal(i.rangeKey.start, s.Start) &&
2186 2 : i.equal(i.rangeKey.end, s.End) {
2187 2 : i.rangeKey.updated = false
2188 2 : i.rangeKey.stale = false
2189 2 : i.rangeKey.hasRangeKey = true
2190 2 : return
2191 2 : }
2192 2 : i.stats.RangeKeyStats.Count += len(s.Keys)
2193 2 : i.rangeKey.buf.Reset()
2194 2 : i.rangeKey.hasRangeKey = true
2195 2 : i.rangeKey.updated = true
2196 2 : i.rangeKey.stale = false
2197 2 : i.rangeKey.buf, i.rangeKey.start = i.rangeKey.buf.Copy(s.Start)
2198 2 : i.rangeKey.buf, i.rangeKey.end = i.rangeKey.buf.Copy(s.End)
2199 2 : i.rangeKey.keys = i.rangeKey.keys[:0]
2200 2 : for j := 0; j < len(s.Keys); j++ {
2201 2 : if invariants.Enabled {
2202 2 : if s.Keys[j].Kind() != base.InternalKeyKindRangeKeySet {
2203 0 : panic("pebble: user iteration encountered non-RangeKeySet key kind")
2204 2 : } else if j > 0 && i.comparer.CompareSuffixes(s.Keys[j].Suffix, s.Keys[j-1].Suffix) < 0 {
2205 0 : panic("pebble: user iteration encountered range keys not in suffix order")
2206 : }
2207 : }
2208 2 : var rkd RangeKeyData
2209 2 : i.rangeKey.buf, rkd.Suffix = i.rangeKey.buf.Copy(s.Keys[j].Suffix)
2210 2 : i.rangeKey.buf, rkd.Value = i.rangeKey.buf.Copy(s.Keys[j].Value)
2211 2 : i.rangeKey.keys = append(i.rangeKey.keys, rkd)
2212 : }
2213 : }
2214 :
2215 : // RangeKeyChanged indicates whether the most recent iterator positioning
2216 : // operation resulted in the iterator stepping into or out of a new range key.
2217 : // If true, previously returned range key bounds and data has been invalidated.
2218 : // If false, previously obtained range key bounds, suffix and value slices are
2219 : // still valid and may continue to be read.
2220 : //
2221 : // Invalid iterator positions are considered to not hold range keys, meaning
2222 : // that if an iterator steps from an IterExhausted or IterAtLimit position onto
2223 : // a position with a range key, RangeKeyChanged will yield true.
2224 2 : func (i *Iterator) RangeKeyChanged() bool {
2225 2 : return i.iterValidityState == IterValid && i.rangeKey != nil && i.rangeKey.updated
2226 2 : }
2227 :
2228 : // HasPointAndRange indicates whether there exists a point key, a range key or
2229 : // both at the current iterator position.
2230 2 : func (i *Iterator) HasPointAndRange() (hasPoint, hasRange bool) {
2231 2 : if i.iterValidityState != IterValid || i.requiresReposition {
2232 1 : return false, false
2233 1 : }
2234 2 : if i.opts.KeyTypes == IterKeyTypePointsOnly {
2235 2 : return true, false
2236 2 : }
2237 2 : return i.rangeKey == nil || !i.rangeKey.rangeKeyOnly, i.rangeKey != nil && i.rangeKey.hasRangeKey
2238 : }
2239 :
2240 : // RangeBounds returns the start (inclusive) and end (exclusive) bounds of the
2241 : // range key covering the current iterator position. RangeBounds returns nil
2242 : // bounds if there is no range key covering the current iterator position, or
2243 : // the iterator is not configured to surface range keys.
2244 : //
2245 : // If valid, the returned start bound is less than or equal to Key() and the
2246 : // returned end bound is greater than Key().
2247 2 : func (i *Iterator) RangeBounds() (start, end []byte) {
2248 2 : if i.rangeKey == nil || !i.opts.rangeKeys() || !i.rangeKey.hasRangeKey {
2249 0 : return nil, nil
2250 0 : }
2251 2 : return i.rangeKey.start, i.rangeKey.end
2252 : }
2253 :
2254 : // Key returns the key of the current key/value pair, or nil if done. The
2255 : // caller should not modify the contents of the returned slice, and its
2256 : // contents may change on the next call to Next.
2257 : //
2258 : // If positioned at an iterator position that only holds a range key, Key()
2259 : // always returns the start bound of the range key. Otherwise, it returns the
2260 : // point key's key.
2261 2 : func (i *Iterator) Key() []byte {
2262 2 : return i.key
2263 2 : }
2264 :
2265 : // Value returns the value of the current key/value pair, or nil if done. The
2266 : // caller should not modify the contents of the returned slice, and its
2267 : // contents may change on the next call to Next.
2268 : //
2269 : // Only valid if HasPointAndRange() returns true for hasPoint.
2270 : // Deprecated: use ValueAndErr instead.
2271 2 : func (i *Iterator) Value() []byte {
2272 2 : val, _ := i.ValueAndErr()
2273 2 : return val
2274 2 : }
2275 :
2276 : // ValueAndErr returns the value, and any error encountered in extracting the value.
2277 : // REQUIRES: i.Error()==nil and HasPointAndRange() returns true for hasPoint.
2278 : //
2279 : // The caller should not modify the contents of the returned slice, and its
2280 : // contents may change on the next call to Next.
2281 2 : func (i *Iterator) ValueAndErr() ([]byte, error) {
2282 2 : val, callerOwned, err := i.value.Value(i.lazyValueBuf)
2283 2 : if err != nil {
2284 1 : i.err = err
2285 1 : i.iterValidityState = IterExhausted
2286 1 : }
2287 2 : if callerOwned {
2288 1 : i.lazyValueBuf = val[:0]
2289 1 : }
2290 2 : return val, err
2291 : }
2292 :
2293 : // LazyValue returns the LazyValue. Only for advanced use cases.
2294 : // REQUIRES: i.Error()==nil and HasPointAndRange() returns true for hasPoint.
2295 0 : func (i *Iterator) LazyValue() LazyValue {
2296 0 : return i.value
2297 0 : }
2298 :
2299 : // RangeKeys returns the range key values and their suffixes covering the
2300 : // current iterator position. The range bounds may be retrieved separately
2301 : // through Iterator.RangeBounds().
2302 2 : func (i *Iterator) RangeKeys() []RangeKeyData {
2303 2 : if i.rangeKey == nil || !i.opts.rangeKeys() || !i.rangeKey.hasRangeKey {
2304 0 : return nil
2305 0 : }
2306 2 : return i.rangeKey.keys
2307 : }
2308 :
2309 : // Valid returns true if the iterator is positioned at a valid key/value pair
2310 : // and false otherwise.
2311 2 : func (i *Iterator) Valid() bool {
2312 2 : valid := i.iterValidityState == IterValid && !i.requiresReposition
2313 2 : if invariants.Enabled {
2314 2 : if err := i.Error(); valid && err != nil {
2315 0 : panic(errors.AssertionFailedf("pebble: iterator is valid with non-nil Error: %+v", err))
2316 : }
2317 : }
2318 2 : return valid
2319 : }
2320 :
2321 : // Error returns any accumulated error.
2322 2 : func (i *Iterator) Error() error {
2323 2 : if i.err != nil {
2324 2 : return i.err
2325 2 : }
2326 2 : if i.iter != nil {
2327 2 : return i.iter.Error()
2328 2 : }
2329 0 : return nil
2330 : }
2331 :
2332 : const maxKeyBufCacheSize = 4 << 10 // 4 KB
2333 :
2334 : // Close closes the iterator and returns any accumulated error. Exhausting
2335 : // all the key/value pairs in a table is not considered to be an error.
2336 : // It is not valid to call any method, including Close, after the iterator
2337 : // has been closed.
2338 2 : func (i *Iterator) Close() error {
2339 2 : // Close the child iterator before releasing the readState because when the
2340 2 : // readState is released sstables referenced by the readState may be deleted
2341 2 : // which will fail on Windows if the sstables are still open by the child
2342 2 : // iterator.
2343 2 : if i.iter != nil {
2344 2 : i.err = firstError(i.err, i.iter.Close())
2345 2 :
2346 2 : // Closing i.iter did not necessarily close the point and range key
2347 2 : // iterators. Calls to SetOptions may have 'disconnected' either one
2348 2 : // from i.iter if iteration key types were changed. Both point and range
2349 2 : // key iterators are preserved in case the iterator needs to switch key
2350 2 : // types again. We explicitly close both of these iterators here.
2351 2 : //
2352 2 : // NB: If the iterators were still connected to i.iter, they may be
2353 2 : // closed, but calling Close on a closed internal iterator or fragment
2354 2 : // iterator is allowed.
2355 2 : if i.pointIter != nil {
2356 2 : i.err = firstError(i.err, i.pointIter.Close())
2357 2 : }
2358 2 : if i.rangeKey != nil && i.rangeKey.rangeKeyIter != nil {
2359 2 : i.rangeKey.rangeKeyIter.Close()
2360 2 : }
2361 : }
2362 2 : err := i.err
2363 2 :
2364 2 : if i.readState != nil {
2365 2 : if i.readSampling.pendingCompactions.size > 0 {
2366 1 : // Copy pending read compactions using db.mu.Lock()
2367 1 : i.readState.db.mu.Lock()
2368 1 : i.readState.db.mu.compact.readCompactions.combine(&i.readSampling.pendingCompactions, i.cmp)
2369 1 : reschedule := i.readState.db.mu.compact.rescheduleReadCompaction
2370 1 : i.readState.db.mu.compact.rescheduleReadCompaction = false
2371 1 : concurrentCompactions := i.readState.db.mu.compact.compactingCount
2372 1 : i.readState.db.mu.Unlock()
2373 1 :
2374 1 : if reschedule && concurrentCompactions == 0 {
2375 0 : // In a read heavy workload, flushes may not happen frequently enough to
2376 0 : // schedule compactions.
2377 0 : i.readState.db.compactionSchedulers.Add(1)
2378 0 : go i.readState.db.maybeScheduleCompactionAsync()
2379 0 : }
2380 : }
2381 :
2382 2 : i.readState.unref()
2383 2 : i.readState = nil
2384 : }
2385 :
2386 2 : if i.version != nil {
2387 2 : i.version.Unref()
2388 2 : }
2389 :
2390 2 : for _, readers := range i.externalReaders {
2391 1 : for _, r := range readers {
2392 1 : err = firstError(err, r.Close())
2393 1 : }
2394 : }
2395 :
2396 : // Close the closer for the current value if one was open.
2397 2 : if i.valueCloser != nil {
2398 1 : err = firstError(err, i.valueCloser.Close())
2399 1 : i.valueCloser = nil
2400 1 : }
2401 :
2402 2 : if i.rangeKey != nil {
2403 2 :
2404 2 : i.rangeKey.rangeKeyBuffers.PrepareForReuse()
2405 2 : *i.rangeKey = iteratorRangeKeyState{
2406 2 : rangeKeyBuffers: i.rangeKey.rangeKeyBuffers,
2407 2 : }
2408 2 : iterRangeKeyStateAllocPool.Put(i.rangeKey)
2409 2 : i.rangeKey = nil
2410 2 : }
2411 2 : if alloc := i.alloc; alloc != nil {
2412 2 : var (
2413 2 : keyBuf []byte
2414 2 : boundsBuf [2][]byte
2415 2 : prefixOrFullSeekKey []byte
2416 2 : mergingIterHeapItems []*mergingIterLevel
2417 2 : )
2418 2 :
2419 2 : // Avoid caching the key buf if it is overly large. The constant is fairly
2420 2 : // arbitrary.
2421 2 : if cap(i.keyBuf) < maxKeyBufCacheSize {
2422 2 : keyBuf = i.keyBuf
2423 2 : }
2424 2 : if cap(i.prefixOrFullSeekKey) < maxKeyBufCacheSize {
2425 2 : prefixOrFullSeekKey = i.prefixOrFullSeekKey
2426 2 : }
2427 2 : for j := range i.boundsBuf {
2428 2 : if cap(i.boundsBuf[j]) < maxKeyBufCacheSize {
2429 2 : boundsBuf[j] = i.boundsBuf[j]
2430 2 : }
2431 : }
2432 2 : mergingIterHeapItems = alloc.merging.heap.items
2433 2 : *alloc = iterAlloc{
2434 2 : keyBuf: keyBuf,
2435 2 : boundsBuf: boundsBuf,
2436 2 : prefixOrFullSeekKey: prefixOrFullSeekKey,
2437 2 : merging: mergingIter{
2438 2 : heap: mergingIterHeap{items: mergingIterHeapItems},
2439 2 : },
2440 2 : }
2441 2 : iterAllocPool.Put(alloc)
2442 2 : } else if alloc := i.getIterAlloc; alloc != nil {
2443 2 : if cap(i.keyBuf) >= maxKeyBufCacheSize {
2444 0 : alloc.keyBuf = nil
2445 2 : } else {
2446 2 : alloc.keyBuf = i.keyBuf
2447 2 : }
2448 2 : *alloc = getIterAlloc{
2449 2 : keyBuf: alloc.keyBuf,
2450 2 : }
2451 2 : getIterAllocPool.Put(alloc)
2452 : }
2453 2 : return err
2454 : }
2455 :
2456 : // SetBounds sets the lower and upper bounds for the iterator. Once SetBounds
2457 : // returns, the caller is free to mutate the provided slices.
2458 : //
2459 : // The iterator will always be invalidated and must be repositioned with a call
2460 : // to SeekGE, SeekPrefixGE, SeekLT, First, or Last.
2461 2 : func (i *Iterator) SetBounds(lower, upper []byte) {
2462 2 : // Ensure that the Iterator appears exhausted, regardless of whether we
2463 2 : // actually have to invalidate the internal iterator. Optimizations that
2464 2 : // avoid exhaustion are an internal implementation detail that shouldn't
2465 2 : // leak through the interface. The caller should still call an absolute
2466 2 : // positioning method to reposition the iterator.
2467 2 : i.requiresReposition = true
2468 2 :
2469 2 : if ((i.opts.LowerBound == nil) == (lower == nil)) &&
2470 2 : ((i.opts.UpperBound == nil) == (upper == nil)) &&
2471 2 : i.equal(i.opts.LowerBound, lower) &&
2472 2 : i.equal(i.opts.UpperBound, upper) {
2473 2 : // Unchanged, noop.
2474 2 : return
2475 2 : }
2476 :
2477 : // Copy the user-provided bounds into an Iterator-owned buffer, and set them
2478 : // on i.opts.{Lower,Upper}Bound.
2479 2 : i.processBounds(lower, upper)
2480 2 :
2481 2 : i.iter.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2482 2 : // If the iterator has an open point iterator that's not currently being
2483 2 : // used, propagate the new bounds to it.
2484 2 : if i.pointIter != nil && !i.opts.pointKeys() {
2485 2 : i.pointIter.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2486 2 : }
2487 : // If the iterator has a range key iterator, propagate bounds to it. The
2488 : // top-level SetBounds on the interleaving iterator (i.iter) won't propagate
2489 : // bounds to the range key iterator stack, because the FragmentIterator
2490 : // interface doesn't define a SetBounds method. We need to directly inform
2491 : // the iterConfig stack.
2492 2 : if i.rangeKey != nil {
2493 2 : i.rangeKey.iterConfig.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2494 2 : }
2495 :
2496 : // Even though this is not a positioning operation, the alteration of the
2497 : // bounds means we cannot optimize Seeks by using Next.
2498 2 : i.invalidate()
2499 : }
2500 :
2501 : // SetContext replaces the context provided at iterator creation, or the last
2502 : // one provided by SetContext. Even though iterators are expected to be
2503 : // short-lived, there are some cases where either (a) iterators are used far
2504 : // from the code that created them, (b) iterators are reused (while being
2505 : // short-lived) for processing different requests. For such scenarios, we
2506 : // allow the caller to replace the context.
2507 0 : func (i *Iterator) SetContext(ctx context.Context) {
2508 0 : i.ctx = ctx
2509 0 : i.iter.SetContext(ctx)
2510 0 : // If the iterator has an open point iterator that's not currently being
2511 0 : // used, propagate the new context to it.
2512 0 : if i.pointIter != nil && !i.opts.pointKeys() {
2513 0 : i.pointIter.SetContext(i.ctx)
2514 0 : }
2515 : }
2516 :
2517 : // Initialization and changing of the bounds must call processBounds.
2518 : // processBounds saves the bounds and computes derived state from those
2519 : // bounds.
2520 2 : func (i *Iterator) processBounds(lower, upper []byte) {
2521 2 : // Copy the user-provided bounds into an Iterator-owned buffer. We can't
2522 2 : // overwrite the current bounds, because some internal iterators compare old
2523 2 : // and new bounds for optimizations.
2524 2 :
2525 2 : buf := i.boundsBuf[i.boundsBufIdx][:0]
2526 2 : if lower != nil {
2527 2 : buf = append(buf, lower...)
2528 2 : i.opts.LowerBound = buf
2529 2 : } else {
2530 2 : i.opts.LowerBound = nil
2531 2 : }
2532 2 : i.nextPrefixNotPermittedByUpperBound = false
2533 2 : if upper != nil {
2534 2 : buf = append(buf, upper...)
2535 2 : i.opts.UpperBound = buf[len(buf)-len(upper):]
2536 2 : if i.comparer.Split(i.opts.UpperBound) != len(i.opts.UpperBound) {
2537 2 : // Setting an upper bound that is a versioned MVCC key. This means
2538 2 : // that a key can have some MVCC versions before the upper bound and
2539 2 : // some after. This causes significant complications for NextPrefix,
2540 2 : // so we bar the user of NextPrefix.
2541 2 : i.nextPrefixNotPermittedByUpperBound = true
2542 2 : }
2543 2 : } else {
2544 2 : i.opts.UpperBound = nil
2545 2 : }
2546 2 : i.boundsBuf[i.boundsBufIdx] = buf
2547 2 : i.boundsBufIdx = 1 - i.boundsBufIdx
2548 : }
2549 :
2550 : // SetOptions sets new iterator options for the iterator. Note that the lower
2551 : // and upper bounds applied here will supersede any bounds set by previous calls
2552 : // to SetBounds.
2553 : //
2554 : // Note that the slices provided in this SetOptions must not be changed by the
2555 : // caller until the iterator is closed, or a subsequent SetBounds or SetOptions
2556 : // has returned. This is because comparisons between the existing and new bounds
2557 : // are sometimes used to optimize seeking. See the extended commentary on
2558 : // SetBounds.
2559 : //
2560 : // If the iterator was created over an indexed mutable batch, the iterator's
2561 : // view of the mutable batch is refreshed.
2562 : //
2563 : // The iterator will always be invalidated and must be repositioned with a call
2564 : // to SeekGE, SeekPrefixGE, SeekLT, First, or Last.
2565 : //
2566 : // If only lower and upper bounds need to be modified, prefer SetBounds.
2567 2 : func (i *Iterator) SetOptions(o *IterOptions) {
2568 2 : if i.externalReaders != nil {
2569 1 : if err := validateExternalIterOpts(o); err != nil {
2570 0 : panic(err)
2571 : }
2572 : }
2573 :
2574 : // Ensure that the Iterator appears exhausted, regardless of whether we
2575 : // actually have to invalidate the internal iterator. Optimizations that
2576 : // avoid exhaustion are an internal implementation detail that shouldn't
2577 : // leak through the interface. The caller should still call an absolute
2578 : // positioning method to reposition the iterator.
2579 2 : i.requiresReposition = true
2580 2 :
2581 2 : // Check if global state requires we close all internal iterators.
2582 2 : //
2583 2 : // If the Iterator is in an error state, invalidate the existing iterators
2584 2 : // so that we reconstruct an iterator state from scratch.
2585 2 : //
2586 2 : // If OnlyReadGuaranteedDurable changed, the iterator stacks are incorrect,
2587 2 : // improperly including or excluding memtables. Invalidate them so that
2588 2 : // finishInitializingIter will reconstruct them.
2589 2 : closeBoth := i.err != nil ||
2590 2 : o.OnlyReadGuaranteedDurable != i.opts.OnlyReadGuaranteedDurable
2591 2 :
2592 2 : // If either options specify block property filters for an iterator stack,
2593 2 : // reconstruct it.
2594 2 : if i.pointIter != nil && (closeBoth || len(o.PointKeyFilters) > 0 || len(i.opts.PointKeyFilters) > 0 ||
2595 2 : o.RangeKeyMasking.Filter != nil || i.opts.RangeKeyMasking.Filter != nil || o.SkipPoint != nil ||
2596 2 : i.opts.SkipPoint != nil) {
2597 2 : i.err = firstError(i.err, i.pointIter.Close())
2598 2 : i.pointIter = nil
2599 2 : }
2600 2 : if i.rangeKey != nil {
2601 2 : if closeBoth || len(o.RangeKeyFilters) > 0 || len(i.opts.RangeKeyFilters) > 0 {
2602 2 : i.rangeKey.rangeKeyIter.Close()
2603 2 : i.rangeKey = nil
2604 2 : } else {
2605 2 : // If there's still a range key iterator stack, invalidate the
2606 2 : // iterator. This ensures RangeKeyChanged() returns true if a
2607 2 : // subsequent positioning operation discovers a range key. It also
2608 2 : // prevents seek no-op optimizations.
2609 2 : i.invalidate()
2610 2 : }
2611 : }
2612 :
2613 : // If the iterator is backed by a batch that's been mutated, refresh its
2614 : // existing point and range-key iterators, and invalidate the iterator to
2615 : // prevent seek-using-next optimizations. If we don't yet have a point-key
2616 : // iterator or range-key iterator but we require one, it'll be created in
2617 : // the slow path that reconstructs the iterator in finishInitializingIter.
2618 2 : if i.batch != nil {
2619 2 : nextBatchSeqNum := (base.SeqNum(len(i.batch.data)) | base.SeqNumBatchBit)
2620 2 : if nextBatchSeqNum != i.batchSeqNum {
2621 2 : i.batchSeqNum = nextBatchSeqNum
2622 2 : if i.merging != nil {
2623 2 : i.merging.batchSnapshot = nextBatchSeqNum
2624 2 : }
2625 : // Prevent a no-op seek optimization on the next seek. We won't be
2626 : // able to reuse the top-level Iterator state, because it may be
2627 : // incorrect after the inclusion of new batch mutations.
2628 2 : i.batchJustRefreshed = true
2629 2 : if i.pointIter != nil && i.batch.countRangeDels > 0 {
2630 1 : if i.batchRangeDelIter.Count() == 0 {
2631 1 : // When we constructed this iterator, there were no
2632 1 : // rangedels in the batch. Iterator construction will
2633 1 : // have excluded the batch rangedel iterator from the
2634 1 : // point iterator stack. We need to reconstruct the
2635 1 : // point iterator to add i.batchRangeDelIter into the
2636 1 : // iterator stack.
2637 1 : i.err = firstError(i.err, i.pointIter.Close())
2638 1 : i.pointIter = nil
2639 1 : } else {
2640 1 : // There are range deletions in the batch and we already
2641 1 : // have a batch rangedel iterator. We can update the
2642 1 : // batch rangedel iterator in place.
2643 1 : //
2644 1 : // NB: There may or may not be new range deletions. We
2645 1 : // can't tell based on i.batchRangeDelIter.Count(),
2646 1 : // which is the count of fragmented range deletions, NOT
2647 1 : // the number of range deletions written to the batch
2648 1 : // [i.batch.countRangeDels].
2649 1 : i.batch.initRangeDelIter(&i.opts, &i.batchRangeDelIter, nextBatchSeqNum)
2650 1 : }
2651 : }
2652 2 : if i.rangeKey != nil && i.batch.countRangeKeys > 0 {
2653 1 : if i.batchRangeKeyIter.Count() == 0 {
2654 1 : // When we constructed this iterator, there were no range
2655 1 : // keys in the batch. Iterator construction will have
2656 1 : // excluded the batch rangekey iterator from the range key
2657 1 : // iterator stack. We need to reconstruct the range key
2658 1 : // iterator to add i.batchRangeKeyIter into the iterator
2659 1 : // stack.
2660 1 : i.rangeKey.rangeKeyIter.Close()
2661 1 : i.rangeKey = nil
2662 1 : } else {
2663 1 : // There are range keys in the batch and we already
2664 1 : // have a batch rangekey iterator. We can update the batch
2665 1 : // rangekey iterator in place.
2666 1 : //
2667 1 : // NB: There may or may not be new range keys. We can't
2668 1 : // tell based on i.batchRangeKeyIter.Count(), which is the
2669 1 : // count of fragmented range keys, NOT the number of
2670 1 : // range keys written to the batch [i.batch.countRangeKeys].
2671 1 : i.batch.initRangeKeyIter(&i.opts, &i.batchRangeKeyIter, nextBatchSeqNum)
2672 1 : i.invalidate()
2673 1 : }
2674 : }
2675 : }
2676 : }
2677 :
2678 : // Reset combinedIterState.initialized in case the iterator key types
2679 : // changed. If there's already a range key iterator stack, the combined
2680 : // iterator is already initialized. Additionally, if the iterator is not
2681 : // configured to include range keys, mark it as initialized to signal that
2682 : // lower level iterators should not trigger a switch to combined iteration.
2683 2 : i.lazyCombinedIter.combinedIterState = combinedIterState{
2684 2 : initialized: i.rangeKey != nil || !i.opts.rangeKeys(),
2685 2 : }
2686 2 :
2687 2 : boundsEqual := ((i.opts.LowerBound == nil) == (o.LowerBound == nil)) &&
2688 2 : ((i.opts.UpperBound == nil) == (o.UpperBound == nil)) &&
2689 2 : i.equal(i.opts.LowerBound, o.LowerBound) &&
2690 2 : i.equal(i.opts.UpperBound, o.UpperBound)
2691 2 :
2692 2 : if boundsEqual && o.KeyTypes == i.opts.KeyTypes &&
2693 2 : (i.pointIter != nil || !i.opts.pointKeys()) &&
2694 2 : (i.rangeKey != nil || !i.opts.rangeKeys() || i.opts.KeyTypes == IterKeyTypePointsAndRanges) &&
2695 2 : i.comparer.CompareSuffixes(o.RangeKeyMasking.Suffix, i.opts.RangeKeyMasking.Suffix) == 0 &&
2696 2 : o.UseL6Filters == i.opts.UseL6Filters {
2697 2 : // The options are identical, so we can likely use the fast path. In
2698 2 : // addition to all the above constraints, we cannot use the fast path if
2699 2 : // configured to perform lazy combined iteration but an indexed batch
2700 2 : // used by the iterator now contains range keys. Lazy combined iteration
2701 2 : // is not compatible with batch range keys because we always need to
2702 2 : // merge the batch's range keys into iteration.
2703 2 : if i.rangeKey != nil || !i.opts.rangeKeys() || i.batch == nil || i.batch.countRangeKeys == 0 {
2704 2 : // Fast path. This preserves the Seek-using-Next optimizations as
2705 2 : // long as the iterator wasn't already invalidated up above.
2706 2 : return
2707 2 : }
2708 : }
2709 : // Slow path.
2710 :
2711 : // The options changed. Save the new ones to i.opts.
2712 2 : if boundsEqual {
2713 2 : // Copying the options into i.opts will overwrite LowerBound and
2714 2 : // UpperBound fields with the user-provided slices. We need to hold on
2715 2 : // to the Pebble-owned slices, so save them and re-set them after the
2716 2 : // copy.
2717 2 : lower, upper := i.opts.LowerBound, i.opts.UpperBound
2718 2 : i.opts = *o
2719 2 : i.opts.LowerBound, i.opts.UpperBound = lower, upper
2720 2 : } else {
2721 2 : i.opts = *o
2722 2 : i.processBounds(o.LowerBound, o.UpperBound)
2723 2 : // Propagate the changed bounds to the existing point iterator.
2724 2 : // NB: We propagate i.opts.{Lower,Upper}Bound, not o.{Lower,Upper}Bound
2725 2 : // because i.opts now point to buffers owned by Pebble.
2726 2 : if i.pointIter != nil {
2727 2 : i.pointIter.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2728 2 : }
2729 2 : if i.rangeKey != nil {
2730 2 : i.rangeKey.iterConfig.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2731 2 : }
2732 : }
2733 :
2734 : // Even though this is not a positioning operation, the invalidation of the
2735 : // iterator stack means we cannot optimize Seeks by using Next.
2736 2 : i.invalidate()
2737 2 :
2738 2 : // Iterators created through NewExternalIter have a different iterator
2739 2 : // initialization process.
2740 2 : if i.externalReaders != nil {
2741 1 : finishInitializingExternal(i.ctx, i)
2742 1 : return
2743 1 : }
2744 2 : finishInitializingIter(i.ctx, i.alloc)
2745 : }
2746 :
2747 2 : func (i *Iterator) invalidate() {
2748 2 : i.lastPositioningOp = unknownLastPositionOp
2749 2 : i.hasPrefix = false
2750 2 : i.iterKV = nil
2751 2 : i.err = nil
2752 2 : // This switch statement isn't necessary for correctness since callers
2753 2 : // should call a repositioning method. We could have arbitrarily set i.pos
2754 2 : // to one of the values. But it results in more intuitive behavior in
2755 2 : // tests, which do not always reposition.
2756 2 : switch i.pos {
2757 2 : case iterPosCurForward, iterPosNext, iterPosCurForwardPaused:
2758 2 : i.pos = iterPosCurForward
2759 2 : case iterPosCurReverse, iterPosPrev, iterPosCurReversePaused:
2760 2 : i.pos = iterPosCurReverse
2761 : }
2762 2 : i.iterValidityState = IterExhausted
2763 2 : if i.rangeKey != nil {
2764 2 : i.rangeKey.iiter.Invalidate()
2765 2 : i.rangeKey.prevPosHadRangeKey = false
2766 2 : }
2767 : }
2768 :
2769 : // Metrics returns per-iterator metrics.
2770 0 : func (i *Iterator) Metrics() IteratorMetrics {
2771 0 : m := IteratorMetrics{
2772 0 : ReadAmp: 1,
2773 0 : }
2774 0 : if mi, ok := i.iter.(*mergingIter); ok {
2775 0 : m.ReadAmp = len(mi.levels)
2776 0 : }
2777 0 : return m
2778 : }
2779 :
2780 : // ResetStats resets the stats to 0.
2781 0 : func (i *Iterator) ResetStats() {
2782 0 : i.stats = IteratorStats{}
2783 0 : }
2784 :
2785 : // Stats returns the current stats.
2786 1 : func (i *Iterator) Stats() IteratorStats {
2787 1 : return i.stats
2788 1 : }
2789 :
2790 : // CloneOptions configures an iterator constructed through Iterator.Clone.
2791 : type CloneOptions struct {
2792 : // IterOptions, if non-nil, define the iterator options to configure a
2793 : // cloned iterator. If nil, the clone adopts the same IterOptions as the
2794 : // iterator being cloned.
2795 : IterOptions *IterOptions
2796 : // RefreshBatchView may be set to true when cloning an Iterator over an
2797 : // indexed batch. When false, the clone adopts the same (possibly stale)
2798 : // view of the indexed batch as the cloned Iterator. When true, the clone is
2799 : // constructed with a refreshed view of the batch, observing all of the
2800 : // batch's mutations at the time of the Clone. If the cloned iterator was
2801 : // not constructed to read over an indexed batch, RefreshVatchView has no
2802 : // effect.
2803 : RefreshBatchView bool
2804 : }
2805 :
2806 : // Clone creates a new Iterator over the same underlying data, i.e., over the
2807 : // same {batch, memtables, sstables}). The resulting iterator is not positioned.
2808 : // It starts with the same IterOptions, unless opts.IterOptions is set.
2809 : //
2810 : // When called on an Iterator over an indexed batch, the clone's visibility of
2811 : // the indexed batch is determined by CloneOptions.RefreshBatchView. If false,
2812 : // the clone inherits the iterator's current (possibly stale) view of the batch,
2813 : // and callers may call SetOptions to subsequently refresh the clone's view to
2814 : // include all batch mutations. If true, the clone is constructed with a
2815 : // complete view of the indexed batch's mutations at the time of the Clone.
2816 : //
2817 : // Callers can use Clone if they need multiple iterators that need to see
2818 : // exactly the same underlying state of the DB. This should not be used to
2819 : // extend the lifetime of the data backing the original Iterator since that
2820 : // will cause an increase in memory and disk usage (use NewSnapshot for that
2821 : // purpose).
2822 2 : func (i *Iterator) Clone(opts CloneOptions) (*Iterator, error) {
2823 2 : return i.CloneWithContext(context.Background(), opts)
2824 2 : }
2825 :
2826 : // CloneWithContext is like Clone, and additionally accepts a context for
2827 : // tracing.
2828 2 : func (i *Iterator) CloneWithContext(ctx context.Context, opts CloneOptions) (*Iterator, error) {
2829 2 : if opts.IterOptions == nil {
2830 2 : opts.IterOptions = &i.opts
2831 2 : }
2832 2 : if i.batchOnlyIter {
2833 1 : return nil, errors.Errorf("cannot Clone a batch-only Iterator")
2834 1 : }
2835 2 : readState := i.readState
2836 2 : vers := i.version
2837 2 : if readState == nil && vers == nil {
2838 1 : return nil, errors.Errorf("cannot Clone a closed Iterator")
2839 1 : }
2840 : // i is already holding a ref, so there is no race with unref here.
2841 : //
2842 : // TODO(bilal): If the underlying iterator was created on a snapshot, we could
2843 : // grab a reference to the current readState instead of reffing the original
2844 : // readState. This allows us to release references to some zombie sstables
2845 : // and memtables.
2846 2 : if readState != nil {
2847 2 : readState.ref()
2848 2 : }
2849 2 : if vers != nil {
2850 2 : vers.Ref()
2851 2 : }
2852 : // Bundle various structures under a single umbrella in order to allocate
2853 : // them together.
2854 2 : buf := iterAllocPool.Get().(*iterAlloc)
2855 2 : dbi := &buf.dbi
2856 2 : *dbi = Iterator{
2857 2 : ctx: ctx,
2858 2 : opts: *opts.IterOptions,
2859 2 : alloc: buf,
2860 2 : merge: i.merge,
2861 2 : comparer: i.comparer,
2862 2 : readState: readState,
2863 2 : version: vers,
2864 2 : keyBuf: buf.keyBuf,
2865 2 : prefixOrFullSeekKey: buf.prefixOrFullSeekKey,
2866 2 : boundsBuf: buf.boundsBuf,
2867 2 : batch: i.batch,
2868 2 : batchSeqNum: i.batchSeqNum,
2869 2 : newIters: i.newIters,
2870 2 : newIterRangeKey: i.newIterRangeKey,
2871 2 : seqNum: i.seqNum,
2872 2 : }
2873 2 : dbi.processBounds(dbi.opts.LowerBound, dbi.opts.UpperBound)
2874 2 :
2875 2 : // If the caller requested the clone have a current view of the indexed
2876 2 : // batch, set the clone's batch sequence number appropriately.
2877 2 : if i.batch != nil && opts.RefreshBatchView {
2878 1 : dbi.batchSeqNum = (base.SeqNum(len(i.batch.data)) | base.SeqNumBatchBit)
2879 1 : }
2880 :
2881 2 : return finishInitializingIter(ctx, buf), nil
2882 : }
2883 :
2884 : // Merge adds all of the argument's statistics to the receiver. It may be used
2885 : // to accumulate stats across multiple iterators.
2886 1 : func (stats *IteratorStats) Merge(o IteratorStats) {
2887 1 : for i := InterfaceCall; i < NumStatsKind; i++ {
2888 1 : stats.ForwardSeekCount[i] += o.ForwardSeekCount[i]
2889 1 : stats.ReverseSeekCount[i] += o.ReverseSeekCount[i]
2890 1 : stats.ForwardStepCount[i] += o.ForwardStepCount[i]
2891 1 : stats.ReverseStepCount[i] += o.ReverseStepCount[i]
2892 1 : }
2893 1 : stats.InternalStats.Merge(o.InternalStats)
2894 1 : stats.RangeKeyStats.Merge(o.RangeKeyStats)
2895 : }
2896 :
2897 1 : func (stats *IteratorStats) String() string {
2898 1 : return redact.StringWithoutMarkers(stats)
2899 1 : }
2900 :
2901 : // SafeFormat implements the redact.SafeFormatter interface.
2902 1 : func (stats *IteratorStats) SafeFormat(s redact.SafePrinter, verb rune) {
2903 1 : if stats.ReverseSeekCount[InterfaceCall] == 0 && stats.ReverseSeekCount[InternalIterCall] == 0 {
2904 1 : s.Printf("seeked %s times (%s internal)",
2905 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InterfaceCall])),
2906 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InternalIterCall])),
2907 1 : )
2908 1 : } else {
2909 1 : s.Printf("seeked %s times (%s fwd/%s rev, internal: %s fwd/%s rev)",
2910 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InterfaceCall]+stats.ReverseSeekCount[InterfaceCall])),
2911 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InterfaceCall])),
2912 1 : humanize.Count.Uint64(uint64(stats.ReverseSeekCount[InterfaceCall])),
2913 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InternalIterCall])),
2914 1 : humanize.Count.Uint64(uint64(stats.ReverseSeekCount[InternalIterCall])),
2915 1 : )
2916 1 : }
2917 1 : s.SafeString("; ")
2918 1 :
2919 1 : if stats.ReverseStepCount[InterfaceCall] == 0 && stats.ReverseStepCount[InternalIterCall] == 0 {
2920 1 : s.Printf("stepped %s times (%s internal)",
2921 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InterfaceCall])),
2922 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InternalIterCall])),
2923 1 : )
2924 1 : } else {
2925 1 : s.Printf("stepped %s times (%s fwd/%s rev, internal: %s fwd/%s rev)",
2926 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InterfaceCall]+stats.ReverseStepCount[InterfaceCall])),
2927 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InterfaceCall])),
2928 1 : humanize.Count.Uint64(uint64(stats.ReverseStepCount[InterfaceCall])),
2929 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InternalIterCall])),
2930 1 : humanize.Count.Uint64(uint64(stats.ReverseStepCount[InternalIterCall])),
2931 1 : )
2932 1 : }
2933 :
2934 1 : if stats.InternalStats != (InternalIteratorStats{}) {
2935 1 : s.SafeString("; ")
2936 1 : stats.InternalStats.SafeFormat(s, verb)
2937 1 : }
2938 1 : if stats.RangeKeyStats != (RangeKeyIteratorStats{}) {
2939 1 : s.SafeString(", ")
2940 1 : stats.RangeKeyStats.SafeFormat(s, verb)
2941 1 : }
2942 : }
2943 :
2944 : // CanDeterministicallySingleDelete takes a valid iterator and examines internal
2945 : // state to determine if a SingleDelete deleting Iterator.Key() would
2946 : // deterministically delete the key. CanDeterministicallySingleDelete requires
2947 : // the iterator to be oriented in the forward direction (eg, the last
2948 : // positioning operation must've been a First, a Seek[Prefix]GE, or a
2949 : // Next[Prefix][WithLimit]).
2950 : //
2951 : // This function does not change the external position of the iterator, and all
2952 : // positioning methods should behave the same as if it was never called. This
2953 : // function will only return a meaningful result the first time it's invoked at
2954 : // an iterator position. This function invalidates the iterator Value's memory,
2955 : // and the caller must not rely on the memory safety of the previous Iterator
2956 : // position.
2957 : //
2958 : // If CanDeterministicallySingleDelete returns true AND the key at the iterator
2959 : // position is not modified between the creation of the Iterator and the commit
2960 : // of a batch containing a SingleDelete over the key, then the caller can be
2961 : // assured that SingleDelete is equivalent to Delete on the local engine, but it
2962 : // may not be true on another engine that received the same writes and with
2963 : // logically equivalent state since this engine may have collapsed multiple SETs
2964 : // into one.
2965 2 : func CanDeterministicallySingleDelete(it *Iterator) (bool, error) {
2966 2 : // This function may only be called once per external iterator position. We
2967 2 : // can validate this by checking the last positioning operation.
2968 2 : if it.lastPositioningOp == internalNextOp {
2969 2 : return false, errors.New("pebble: CanDeterministicallySingleDelete called twice")
2970 2 : }
2971 2 : validity, kind := it.internalNext()
2972 2 : var shadowedBySingleDelete bool
2973 2 : for validity == internalNextValid {
2974 2 : switch kind {
2975 2 : case InternalKeyKindDelete, InternalKeyKindDeleteSized:
2976 2 : // A DEL or DELSIZED tombstone is okay. An internal key
2977 2 : // sequence like SINGLEDEL; SET; DEL; SET can be handled
2978 2 : // deterministically. If there are SETs further down, we
2979 2 : // don't care about them.
2980 2 : return true, nil
2981 2 : case InternalKeyKindSingleDelete:
2982 2 : // A SingleDelete is okay as long as when that SingleDelete was
2983 2 : // written, it was written deterministically (eg, with its own
2984 2 : // CanDeterministicallySingleDelete check). Validate that it was
2985 2 : // written deterministically. We'll allow one set to appear after
2986 2 : // the SingleDelete.
2987 2 : shadowedBySingleDelete = true
2988 2 : validity, kind = it.internalNext()
2989 2 : continue
2990 2 : case InternalKeyKindSet, InternalKeyKindSetWithDelete, InternalKeyKindMerge:
2991 2 : // If we observed a single delete, it's allowed to delete 1 key.
2992 2 : // We'll keep looping to validate that the internal keys beneath the
2993 2 : // already-written single delete are copacetic.
2994 2 : if shadowedBySingleDelete {
2995 1 : shadowedBySingleDelete = false
2996 1 : validity, kind = it.internalNext()
2997 1 : continue
2998 : }
2999 : // We encountered a shadowed SET, SETWITHDEL, MERGE. A SINGLEDEL
3000 : // that deleted the KV at the original iterator position could
3001 : // result in this key becoming visible.
3002 2 : return false, nil
3003 0 : case InternalKeyKindRangeDelete:
3004 0 : // RangeDeletes are handled by the merging iterator and should never
3005 0 : // be observed by the top-level Iterator.
3006 0 : panic(errors.AssertionFailedf("pebble: unexpected range delete"))
3007 0 : case InternalKeyKindRangeKeySet, InternalKeyKindRangeKeyUnset, InternalKeyKindRangeKeyDelete:
3008 0 : // Range keys are interleaved at the maximal sequence number and
3009 0 : // should never be observed within a user key.
3010 0 : panic(errors.AssertionFailedf("pebble: unexpected range key"))
3011 0 : default:
3012 0 : panic(errors.AssertionFailedf("pebble: unexpected key kind: %s", errors.Safe(kind)))
3013 : }
3014 : }
3015 2 : if validity == internalNextError {
3016 2 : return false, it.Error()
3017 2 : }
3018 2 : return true, nil
3019 : }
3020 :
3021 : // internalNextValidity enumerates the potential outcomes of a call to
3022 : // internalNext.
3023 : type internalNextValidity int8
3024 :
3025 : const (
3026 : // internalNextError is returned by internalNext when an error occurred and
3027 : // the caller is responsible for checking iter.Error().
3028 : internalNextError internalNextValidity = iota
3029 : // internalNextExhausted is returned by internalNext when the next internal
3030 : // key is an internal key with a different user key than Iterator.Key().
3031 : internalNextExhausted
3032 : // internalNextValid is returned by internalNext when the internal next
3033 : // found a shadowed internal key with a user key equal to Iterator.Key().
3034 : internalNextValid
3035 : )
3036 :
3037 : // internalNext advances internal Iterator state forward to expose the
3038 : // InternalKeyKind of the next internal key with a user key equal to Key().
3039 : //
3040 : // internalNext is a highly specialized operation and is unlikely to be
3041 : // generally useful. See Iterator.Next for how to reposition the iterator to the
3042 : // next key. internalNext requires the Iterator to be at a valid position in the
3043 : // forward direction (the last positioning operation must've been a First, a
3044 : // Seek[Prefix]GE, or a Next[Prefix][WithLimit] and Valid() must return true).
3045 : //
3046 : // internalNext, unlike all other Iterator methods, exposes internal LSM state.
3047 : // internalNext advances the Iterator's internal iterator to the next shadowed
3048 : // key with a user key equal to Key(). When a key is overwritten or deleted, its
3049 : // removal from the LSM occurs lazily as a part of compactions. internalNext
3050 : // allows the caller to see whether an obsolete internal key exists with the
3051 : // current Key(), and what it's key kind is. Note that the existence of an
3052 : // internal key is nondeterministic and dependent on internal LSM state. These
3053 : // semantics are unlikely to be applicable to almost all use cases.
3054 : //
3055 : // If internalNext finds a key that shares the same user key as Key(), it
3056 : // returns internalNextValid and the internal key's kind. If internalNext
3057 : // encounters an error, it returns internalNextError and the caller is expected
3058 : // to call Iterator.Error() to retrieve it. In all other circumstances,
3059 : // internalNext returns internalNextExhausted, indicating that there are no more
3060 : // additional internal keys with the user key Key().
3061 : //
3062 : // internalNext does not change the external position of the iterator, and a
3063 : // Next operation should behave the same as if internalNext was never called.
3064 : // internalNext does invalidate the iterator Value's memory, and the caller must
3065 : // not rely on the memory safety of the previous Iterator position.
3066 2 : func (i *Iterator) internalNext() (internalNextValidity, base.InternalKeyKind) {
3067 2 : i.stats.ForwardStepCount[InterfaceCall]++
3068 2 : if i.err != nil {
3069 2 : return internalNextError, base.InternalKeyKindInvalid
3070 2 : } else if i.iterValidityState != IterValid {
3071 2 : return internalNextExhausted, base.InternalKeyKindInvalid
3072 2 : }
3073 2 : i.lastPositioningOp = internalNextOp
3074 2 :
3075 2 : switch i.pos {
3076 2 : case iterPosCurForward:
3077 2 : i.iterKV = i.iter.Next()
3078 2 : if i.iterKV == nil {
3079 2 : // We check i.iter.Error() here and return an internalNextError enum
3080 2 : // variant so that the caller does not need to check i.iter.Error()
3081 2 : // in the common case that the next internal key has a new user key.
3082 2 : if i.err = i.iter.Error(); i.err != nil {
3083 0 : return internalNextError, base.InternalKeyKindInvalid
3084 0 : }
3085 2 : i.pos = iterPosNext
3086 2 : return internalNextExhausted, base.InternalKeyKindInvalid
3087 2 : } else if i.comparer.Equal(i.iterKV.K.UserKey, i.key) {
3088 2 : return internalNextValid, i.iterKV.Kind()
3089 2 : }
3090 2 : i.pos = iterPosNext
3091 2 : return internalNextExhausted, base.InternalKeyKindInvalid
3092 2 : case iterPosCurReverse, iterPosCurReversePaused, iterPosPrev:
3093 2 : i.err = errors.New("switching from reverse to forward via internalNext is prohibited")
3094 2 : i.iterValidityState = IterExhausted
3095 2 : return internalNextError, base.InternalKeyKindInvalid
3096 2 : case iterPosNext, iterPosCurForwardPaused:
3097 2 : // The previous method already moved onto the next user key. This is
3098 2 : // only possible if
3099 2 : // - the last positioning method was a call to internalNext, and we
3100 2 : // advanced to a new user key.
3101 2 : // - the previous non-internalNext iterator operation encountered a
3102 2 : // range key or merge, forcing an internal Next that found a new
3103 2 : // user key that's not equal to i.Iterator.Key().
3104 2 : return internalNextExhausted, base.InternalKeyKindInvalid
3105 0 : default:
3106 0 : panic("unreachable")
3107 : }
3108 : }
3109 :
3110 : var _ base.IteratorDebug = (*Iterator)(nil)
3111 :
3112 : // DebugTree implements the base.IteratorDebug interface.
3113 0 : func (i *Iterator) DebugTree(tp treeprinter.Node) {
3114 0 : n := tp.Childf("%T(%p)", i, i)
3115 0 : if i.iter != nil {
3116 0 : i.iter.DebugTree(n)
3117 0 : }
3118 0 : if i.pointIter != nil {
3119 0 : i.pointIter.DebugTree(n)
3120 0 : }
3121 : }
|
