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/bytesprofile"
19 : "github.com/cockroachdb/pebble/internal/humanize"
20 : "github.com/cockroachdb/pebble/internal/inflight"
21 : "github.com/cockroachdb/pebble/internal/invariants"
22 : "github.com/cockroachdb/pebble/internal/keyspan"
23 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
24 : "github.com/cockroachdb/pebble/internal/manifest"
25 : "github.com/cockroachdb/pebble/internal/rangekeystack"
26 : "github.com/cockroachdb/pebble/internal/treeprinter"
27 : "github.com/cockroachdb/pebble/sstable/blob"
28 : "github.com/cockroachdb/redact"
29 : )
30 :
31 : // iterPos describes the state of the internal iterator, in terms of whether it
32 : // is at the position returned to the user (cur), one ahead of the position
33 : // returned (next for forward iteration and prev for reverse iteration). The cur
34 : // position is split into two states, for forward and reverse iteration, since
35 : // we need to differentiate for switching directions.
36 : //
37 : // There is subtlety in what is considered the current position of the Iterator.
38 : // The internal iterator exposes a sequence of internal keys. There is not
39 : // always a single internalIterator position corresponding to the position
40 : // returned to the user. Consider the example:
41 : //
42 : // a.MERGE.9 a.MERGE.8 a.MERGE.7 a.SET.6 b.DELETE.9 b.DELETE.5 b.SET.4
43 : // \ /
44 : // \ Iterator.Key() = 'a' /
45 : //
46 : // The Iterator exposes one valid position at user key 'a' and the two exhausted
47 : // positions at the beginning and end of iteration. The underlying
48 : // internalIterator contains 7 valid positions and 2 exhausted positions.
49 : //
50 : // Iterator positioning methods must set iterPos to iterPosCur{Foward,Backward}
51 : // iff the user key at the current internalIterator position equals the
52 : // Iterator.Key returned to the user. This guarantees that a call to nextUserKey
53 : // or prevUserKey will advance to the next or previous iterator position.
54 : // iterPosCur{Forward,Backward} does not make any guarantee about the internal
55 : // iterator position among internal keys with matching user keys, and it will
56 : // vary subtly depending on the particular key kinds encountered. In the above
57 : // example, the iterator returning 'a' may set iterPosCurForward if the internal
58 : // iterator is positioned at any of a.MERGE.9, a.MERGE.8, a.MERGE.7 or a.SET.6.
59 : //
60 : // When setting iterPos to iterPosNext or iterPosPrev, the internal iterator
61 : // must be advanced to the first internalIterator position at a user key greater
62 : // (iterPosNext) or less (iterPosPrev) than the key returned to the user. An
63 : // internalIterator position that's !Valid() must also be considered greater or
64 : // less—depending on the direction of iteration—than the last valid Iterator
65 : // position.
66 : type iterPos int8
67 :
68 : const (
69 : iterPosCurForward iterPos = 0
70 : iterPosNext iterPos = 1
71 : iterPosPrev iterPos = -1
72 : iterPosCurReverse iterPos = -2
73 :
74 : // For limited iteration. When the iterator is at iterPosCurForwardPaused
75 : // - Next*() call should behave as if the internal iterator is already
76 : // at next (akin to iterPosNext).
77 : // - Prev*() call should behave as if the internal iterator is at the
78 : // current key (akin to iterPosCurForward).
79 : //
80 : // Similar semantics apply to CurReversePaused.
81 : iterPosCurForwardPaused iterPos = 2
82 : iterPosCurReversePaused iterPos = -3
83 : )
84 :
85 : // Approximate gap in bytes between samples of data read during iteration.
86 : // This is multiplied with a default ReadSamplingMultiplier of 1 << 4 to yield
87 : // 1 << 20 (1MB). The 1MB factor comes from:
88 : // https://github.com/cockroachdb/pebble/issues/29#issuecomment-494477985
89 : const readBytesPeriod uint64 = 1 << 16
90 :
91 : var errReversePrefixIteration = errors.New("pebble: unsupported reverse prefix iteration")
92 :
93 : // IteratorMetrics holds per-iterator metrics. These do not change over the
94 : // lifetime of the iterator.
95 : type IteratorMetrics struct {
96 : // The read amplification experienced by this iterator. This is the sum of
97 : // the memtables, the L0 sublevels and the non-empty Ln levels. Higher read
98 : // amplification generally results in slower reads, though allowing higher
99 : // read amplification can also result in faster writes.
100 : ReadAmp int
101 : }
102 :
103 : // IteratorStatsKind describes the two kind of iterator stats.
104 : type IteratorStatsKind int8
105 :
106 : const (
107 : // InterfaceCall represents calls to Iterator.
108 : InterfaceCall IteratorStatsKind = iota
109 : // InternalIterCall represents calls by Iterator to its internalIterator.
110 : InternalIterCall
111 : // NumStatsKind is the number of kinds, and is used for array sizing.
112 : NumStatsKind
113 : )
114 :
115 : // IteratorStats contains iteration stats.
116 : type IteratorStats struct {
117 : // ForwardSeekCount includes SeekGE, SeekPrefixGE, First.
118 : ForwardSeekCount [NumStatsKind]int
119 : // ReverseSeek includes SeekLT, Last.
120 : ReverseSeekCount [NumStatsKind]int
121 : // ForwardStepCount includes Next.
122 : ForwardStepCount [NumStatsKind]int
123 : // ReverseStepCount includes Prev.
124 : ReverseStepCount [NumStatsKind]int
125 : InternalStats InternalIteratorStats
126 : RangeKeyStats RangeKeyIteratorStats
127 : }
128 :
129 : var _ redact.SafeFormatter = &IteratorStats{}
130 :
131 : // InternalIteratorStats contains miscellaneous stats produced by internal
132 : // iterators.
133 : type InternalIteratorStats = base.InternalIteratorStats
134 :
135 : // BlockReadStats contains iterators stats about block reads.
136 : type BlockReadStats = base.BlockReadStats
137 :
138 : // RangeKeyIteratorStats contains miscellaneous stats about range keys
139 : // encountered by the iterator.
140 : type RangeKeyIteratorStats struct {
141 : // Count records the number of range keys encountered during
142 : // iteration. Range keys may be counted multiple times if the iterator
143 : // leaves a range key's bounds and then returns.
144 : Count int
145 : // ContainedPoints records the number of point keys encountered within the
146 : // bounds of a range key. Note that this includes point keys with suffixes
147 : // that sort both above and below the covering range key's suffix.
148 : ContainedPoints int
149 : // SkippedPoints records the count of the subset of ContainedPoints point
150 : // keys that were skipped during iteration due to range-key masking. It does
151 : // not include point keys that were never loaded because a
152 : // RangeKeyMasking.Filter excluded the entire containing block.
153 : SkippedPoints int
154 : }
155 :
156 : // Merge adds all of the argument's statistics to the receiver. It may be used
157 : // to accumulate stats across multiple iterators.
158 1 : func (s *RangeKeyIteratorStats) Merge(o RangeKeyIteratorStats) {
159 1 : s.Count += o.Count
160 1 : s.ContainedPoints += o.ContainedPoints
161 1 : s.SkippedPoints += o.SkippedPoints
162 1 : }
163 :
164 0 : func (s *RangeKeyIteratorStats) String() string {
165 0 : return redact.StringWithoutMarkers(s)
166 0 : }
167 :
168 : // SafeFormat implements the redact.SafeFormatter interface.
169 1 : func (s *RangeKeyIteratorStats) SafeFormat(p redact.SafePrinter, verb rune) {
170 1 : p.Printf("range keys: %s, contained points: %s (%s skipped)",
171 1 : humanize.Count.Uint64(uint64(s.Count)),
172 1 : humanize.Count.Uint64(uint64(s.ContainedPoints)),
173 1 : humanize.Count.Uint64(uint64(s.SkippedPoints)))
174 1 : }
175 :
176 : // LazyValue is a lazy value. See the long comment in base.LazyValue.
177 : type LazyValue = base.LazyValue
178 :
179 : // Iterator iterates over a DB's key/value pairs in key order.
180 : //
181 : // An iterator must be closed after use, but it is not necessary to read an
182 : // iterator until exhaustion.
183 : //
184 : // An iterator is not goroutine-safe, but it is safe to use multiple iterators
185 : // concurrently, with each in a dedicated goroutine.
186 : //
187 : // It is also safe to use an iterator concurrently with modifying its
188 : // underlying DB, if that DB permits modification. However, the resultant
189 : // key/value pairs are not guaranteed to be a consistent snapshot of that DB
190 : // at a particular point in time.
191 : //
192 : // If an iterator encounters an error during any operation, it is stored by
193 : // the Iterator and surfaced through the Error method. All absolute
194 : // positioning methods (eg, SeekLT, SeekGT, First, Last, etc) reset any
195 : // accumulated error before positioning. All relative positioning methods (eg,
196 : // Next, Prev) return without advancing if the iterator has an accumulated
197 : // error.
198 : type Iterator struct {
199 : // blobValueFetcher is the ValueFetcher to use when retrieving values stored
200 : // externally in blob files.
201 : blobValueFetcher blob.ValueFetcher
202 :
203 : // All fields below this field are cleared during Iterator.Close before
204 : // returning the Iterator to the pool. Any fields above this field must also
205 : // be cleared, but may be cleared as a part of the body of Iterator.Close
206 : // (eg, if these types have their own Close method that zeroes their
207 : // fields).
208 : clearForReuseBoundary struct{}
209 :
210 : // The context is stored here since (a) Iterators are expected to be
211 : // short-lived (since they pin memtables and sstables), (b) plumbing a
212 : // context into every method is very painful, (c) they do not (yet) respect
213 : // context cancellation and are only used for tracing.
214 : ctx context.Context
215 : // TODO(radu): group fields that are always set from `db` fields (tracker,
216 : // merge, comparer, fc) into a single struct we can point to.
217 : tracker *inflight.Tracker
218 : trackerHandle inflight.Handle
219 : opts IterOptions
220 : merge Merge
221 : comparer *base.Comparer
222 : iter internalIterator
223 : pointIter topLevelIterator
224 : // Either readState or version is set, but not both.
225 : readState *readState
226 : version *manifest.Version
227 : // rangeKey holds iteration state specific to iteration over range keys.
228 : // The range key field may be nil if the Iterator has never been configured
229 : // to iterate over range keys. Its non-nilness cannot be used to determine
230 : // if the Iterator is currently iterating over range keys: For that, consult
231 : // the IterOptions using opts.rangeKeys(). If non-nil, its rangeKeyIter
232 : // field is guaranteed to be non-nil too.
233 : rangeKey *iteratorRangeKeyState
234 : // rangeKeyMasking holds state for range-key masking of point keys.
235 : rangeKeyMasking rangeKeyMasking
236 : err error
237 : // When iterValidityState=IterValid, key represents the current key, which
238 : // is backed by keyBuf.
239 : key []byte
240 : keyBuf []byte
241 : value base.InternalValue
242 : // For use in LazyValue.Clone.
243 : valueBuf []byte
244 : fetcher base.LazyFetcher
245 : // For use in LazyValue.Value.
246 : lazyValueBuf []byte
247 : valueCloser io.Closer
248 : // boundsBuf holds two buffers used to store the lower and upper bounds.
249 : // Whenever the Iterator's bounds change, the new bounds are copied into
250 : // boundsBuf[boundsBufIdx]. The two bounds share a slice to reduce
251 : // allocations. opts.LowerBound and opts.UpperBound point into this slice.
252 : boundsBuf [2][]byte
253 : boundsBufIdx int
254 : // iterKV reflects the latest position of iter, except when SetBounds is
255 : // called. In that case, it is explicitly set to nil.
256 : iterKV *base.InternalKV
257 : alloc *iterAlloc
258 : prefixOrFullSeekKey []byte
259 : readSampling readSampling
260 : stats IteratorStats
261 : externalIter *externalIterState
262 : // Following fields used when constructing an iterator stack, eg, in Clone
263 : // and SetOptions or when re-fragmenting a batch's range keys/range dels.
264 : fc *fileCacheHandle
265 : newIters tableNewIters
266 : newIterRangeKey keyspanimpl.TableNewSpanIter
267 : valueRetrievalProfile *bytesprofile.Profile
268 : lazyCombinedIter lazyCombinedIter
269 : seqNum base.SeqNum
270 : // batch is non-nil if this Iterator includes an indexed batch. Batch
271 : // contains all the state pertaining to iterating over the indexed batch.
272 : // The iteratorBatchState struct is bundled within the iterAlloc struct to
273 : // reduce allocations.
274 : batch *iteratorBatchState
275 : // merging is a pointer to this iterator's point merging iterator. It
276 : // appears here because key visibility is handled by the merging iterator.
277 : // During SetOptions on an iterator over an indexed batch, this field is
278 : // used to update the merging iterator's batch snapshot.
279 : merging *mergingIter
280 :
281 : // Keeping the bools here after all the 8 byte aligned fields shrinks the
282 : // sizeof this struct by 24 bytes.
283 :
284 : // INVARIANT:
285 : // iterValidityState==IterAtLimit <=>
286 : // pos==iterPosCurForwardPaused || pos==iterPosCurReversePaused
287 : iterValidityState IterValidityState
288 : // Set to true by SetBounds, SetOptions. Causes the Iterator to appear
289 : // exhausted externally, while preserving the correct iterValidityState for
290 : // the iterator's internal state. Preserving the correct internal validity
291 : // is used for SeekPrefixGE(..., trySeekUsingNext), and SeekGE/SeekLT
292 : // optimizations after "no-op" calls to SetBounds and SetOptions.
293 : requiresReposition bool
294 : // The position of iter. When this is iterPos{Prev,Next} the iter has been
295 : // moved past the current key-value, which can only happen if
296 : // iterValidityState=IterValid, i.e., there is something to return to the
297 : // client for the current position.
298 : pos iterPos
299 : // Relates to the prefixOrFullSeekKey field above.
300 : hasPrefix bool
301 : // Used for deriving the value of SeekPrefixGE(..., trySeekUsingNext),
302 : // and SeekGE/SeekLT optimizations
303 : lastPositioningOp lastPositioningOpKind
304 : // Used for determining when it's safe to perform SeekGE optimizations that
305 : // reuse the iterator state to avoid the cost of a full seek if the iterator
306 : // is already positioned in the correct place. If the iterator's view of its
307 : // indexed batch was just refreshed, some optimizations cannot be applied on
308 : // the first seek after the refresh:
309 : // - SeekGE has a no-op optimization that does not seek on the internal
310 : // iterator at all if the iterator is already in the correct place.
311 : // This optimization cannot be performed if the internal iterator was
312 : // last positioned when the iterator had a different view of an
313 : // underlying batch.
314 : // - Seek[Prefix]GE set flags.TrySeekUsingNext()=true when the seek key is
315 : // greater than the previous operation's seek key, under the expectation
316 : // that the various internal iterators can use their current position to
317 : // avoid a full expensive re-seek. This applies to the batchIter as well.
318 : // However, if the view of the batch was just refreshed, the batchIter's
319 : // position is not useful because it may already be beyond new keys less
320 : // than the seek key. To prevent the use of this optimization in
321 : // batchIter, Seek[Prefix]GE set flags.BatchJustRefreshed()=true if this
322 : // bit is enabled.
323 : batchJustRefreshed bool
324 : // batchOnlyIter is set to true for Batch.NewBatchOnlyIter.
325 : batchOnlyIter bool
326 : // Used in some tests to disable the random disabling of seek optimizations.
327 : forceEnableSeekOpt bool
328 : // Set to true if NextPrefix is not currently permitted. Defaults to false
329 : // in case an iterator never had any bounds.
330 : nextPrefixNotPermittedByUpperBound bool
331 : }
332 :
333 : const clearOff = unsafe.Offsetof(Iterator{}.clearForReuseBoundary)
334 : const clearLen = unsafe.Sizeof(Iterator{}) - clearOff
335 :
336 1 : func (i *Iterator) clearForReuse() {
337 1 : *(*[clearLen]byte)(unsafe.Add(unsafe.Pointer(i), clearOff)) = [clearLen]byte{}
338 1 : }
339 :
340 : // cmp is a convenience shorthand for the i.comparer.Compare function.
341 1 : func (i *Iterator) cmp(a, b []byte) int {
342 1 : return i.comparer.Compare(a, b)
343 1 : }
344 :
345 : // equal is a convenience shorthand for the i.comparer.Equal function.
346 1 : func (i *Iterator) equal(a, b []byte) bool {
347 1 : return i.comparer.Equal(a, b)
348 1 : }
349 :
350 : // iteratorBatchState holds state pertaining to iterating over an indexed batch.
351 : // When an iterator is configured to read through an indexed batch, the iterator
352 : // maintains a pointer to this struct. This struct is embedded within the
353 : // iterAlloc struct to reduce allocations.
354 : type iteratorBatchState struct {
355 : batch *Batch
356 : // batchSeqNum is used by Iterators over indexed batches to detect when the
357 : // underlying batch has been mutated. The batch beneath an indexed batch may
358 : // be mutated while the Iterator is open, but new keys are not surfaced
359 : // until the next call to SetOptions.
360 : batchSeqNum base.SeqNum
361 : // batch{PointIter,RangeDelIter,RangeKeyIter} are used when the Iterator is
362 : // configured to read through an indexed batch. If a batch is set, these
363 : // iterators will be included within the iterator stack regardless of
364 : // whether the batch currently contains any keys of their kind. These
365 : // pointers are used during a call to SetOptions to refresh the Iterator's
366 : // view of its indexed batch.
367 : pointIter batchIter
368 : rangeDelIter keyspan.Iter
369 : rangeKeyIter keyspan.Iter
370 : }
371 :
372 : // iteratorRangeKeyState holds an iterator's range key iteration state.
373 : type iteratorRangeKeyState struct {
374 : // rangeKeyIter holds the range key iterator stack that iterates over the
375 : // merged spans across the entirety of the LSM.
376 : rangeKeyIter keyspan.FragmentIterator
377 : iiter keyspan.InterleavingIter
378 : // stale is set to true when the range key state recorded here (in start,
379 : // end and keys) may not be in sync with the current range key at the
380 : // interleaving iterator's current position.
381 : //
382 : // When the interelaving iterator passes over a new span, it invokes the
383 : // SpanChanged hook defined on the `rangeKeyMasking` type, which sets stale
384 : // to true if the span is non-nil.
385 : //
386 : // The parent iterator may not be positioned over the interleaving
387 : // iterator's current position (eg, i.iterPos = iterPos{Next,Prev}), so
388 : // {keys,start,end} are only updated to the new range key during a call to
389 : // Iterator.saveRangeKey.
390 : stale bool
391 : // updated is used to signal to the Iterator client whether the state of
392 : // range keys has changed since the previous iterator position through the
393 : // `RangeKeyChanged` method. It's set to true during an Iterator positioning
394 : // operation that changes the state of the current range key. Each Iterator
395 : // positioning operation sets it back to false before executing.
396 : //
397 : // TODO(jackson): The lifecycle of {stale,updated,prevPosHadRangeKey} is
398 : // intricate and confusing. Try to refactor to reduce complexity.
399 : updated bool
400 : // prevPosHadRangeKey records whether the previous Iterator position had a
401 : // range key (HasPointAndRage() = (_, true)). It's updated at the beginning
402 : // of each new Iterator positioning operation. It's required by saveRangeKey to
403 : // to set `updated` appropriately: Without this record of the previous iterator
404 : // state, it's ambiguous whether an iterator only temporarily stepped onto a
405 : // position without a range key.
406 : prevPosHadRangeKey bool
407 : // rangeKeyOnly is set to true if at the current iterator position there is
408 : // no point key, only a range key start boundary.
409 : rangeKeyOnly bool
410 : // hasRangeKey is true when the current iterator position has a covering
411 : // range key (eg, a range key with bounds [<lower>,<upper>) such that
412 : // <lower> ≤ Key() < <upper>).
413 : hasRangeKey bool
414 : // start and end are the [start, end) boundaries of the current range keys.
415 : start []byte
416 : end []byte
417 :
418 : rangeKeyBuffers
419 :
420 : // iterConfig holds fields that are used for the construction of the
421 : // iterator stack, but do not need to be directly accessed during iteration.
422 : // This struct is bundled within the iteratorRangeKeyState struct to reduce
423 : // allocations.
424 : iterConfig rangekeystack.UserIteratorConfig
425 : }
426 :
427 : type rangeKeyBuffers struct {
428 : // keys is sorted by Suffix ascending.
429 : keys []RangeKeyData
430 : // buf is used to save range-key data before moving the range-key iterator.
431 : // Start and end boundaries, suffixes and values are all copied into buf.
432 : buf bytealloc.A
433 : // internal holds buffers used by the range key internal iterators.
434 : internal rangekeystack.Buffers
435 : }
436 :
437 1 : func (b *rangeKeyBuffers) PrepareForReuse() {
438 1 : const maxKeysReuse = 100
439 1 : if len(b.keys) > maxKeysReuse {
440 0 : b.keys = nil
441 0 : }
442 : // Avoid caching the key buf if it is overly large. The constant is
443 : // fairly arbitrary.
444 1 : if cap(b.buf) >= maxKeyBufCacheSize {
445 1 : b.buf = nil
446 1 : } else {
447 1 : b.buf = b.buf[:0]
448 1 : }
449 1 : b.internal.PrepareForReuse()
450 : }
451 :
452 : var iterRangeKeyStateAllocPool = sync.Pool{
453 1 : New: func() interface{} {
454 1 : return &iteratorRangeKeyState{}
455 1 : },
456 : }
457 :
458 : // isEphemeralPosition returns true iff the current iterator position is
459 : // ephemeral, and won't be visited during subsequent relative positioning
460 : // operations.
461 : //
462 : // The iterator position resulting from a SeekGE or SeekPrefixGE that lands on a
463 : // straddling range key without a coincident point key is such a position.
464 1 : func (i *Iterator) isEphemeralPosition() bool {
465 1 : return i.opts.rangeKeys() && i.rangeKey != nil && i.rangeKey.rangeKeyOnly &&
466 1 : !i.equal(i.rangeKey.start, i.key)
467 1 : }
468 :
469 : type lastPositioningOpKind int8
470 :
471 : const (
472 : unknownLastPositionOp lastPositioningOpKind = iota
473 : seekPrefixGELastPositioningOp
474 : seekGELastPositioningOp
475 : seekLTLastPositioningOp
476 : // internalNextOp is a special internal iterator positioning operation used
477 : // by CanDeterministicallySingleDelete. It exists for enforcing requirements
478 : // around calling CanDeterministicallySingleDelete at most once per external
479 : // iterator position.
480 : internalNextOp
481 : )
482 :
483 : // Limited iteration mode. Not for use with prefix iteration.
484 : //
485 : // SeekGE, SeekLT, Prev, Next have WithLimit variants, that pause the iterator
486 : // at the limit in a best-effort manner. The client should behave correctly
487 : // even if the limits are ignored. These limits are not "deep", in that they
488 : // are not passed down to the underlying collection of internalIterators. This
489 : // is because the limits are transient, and apply only until the next
490 : // iteration call. They serve mainly as a way to bound the amount of work when
491 : // two (or more) Iterators are being coordinated at a higher level.
492 : //
493 : // In limited iteration mode:
494 : // - Avoid using Iterator.Valid if the last call was to a *WithLimit() method.
495 : // The return value from the *WithLimit() method provides a more precise
496 : // disposition.
497 : // - The limit is exclusive for forward and inclusive for reverse.
498 : //
499 : //
500 : // Limited iteration mode & range keys
501 : //
502 : // Limited iteration interacts with range-key iteration. When range key
503 : // iteration is enabled, range keys are interleaved at their start boundaries.
504 : // Limited iteration must ensure that if a range key exists within the limit,
505 : // the iterator visits the range key.
506 : //
507 : // During forward limited iteration, this is trivial: An overlapping range key
508 : // must have a start boundary less than the limit, and the range key's start
509 : // boundary will be interleaved and found to be within the limit.
510 : //
511 : // During reverse limited iteration, the tail of the range key may fall within
512 : // the limit. The range key must be surfaced even if the range key's start
513 : // boundary is less than the limit, and if there are no point keys between the
514 : // current iterator position and the limit. To provide this guarantee, reverse
515 : // limited iteration ignores the limit as long as there is a range key
516 : // overlapping the iteration position.
517 :
518 : // IterValidityState captures the state of the Iterator.
519 : type IterValidityState int8
520 :
521 : const (
522 : // IterExhausted represents an Iterator that is exhausted.
523 : IterExhausted IterValidityState = iota
524 : // IterValid represents an Iterator that is valid.
525 : IterValid
526 : // IterAtLimit represents an Iterator that has a non-exhausted
527 : // internalIterator, but has reached a limit without any key for the
528 : // caller.
529 : IterAtLimit
530 : )
531 :
532 : // readSampling stores variables used to sample a read to trigger a read
533 : // compaction
534 : type readSampling struct {
535 : bytesUntilReadSampling uint64
536 : initialSamplePassed bool
537 : pendingCompactions readCompactionQueue
538 : // forceReadSampling is used for testing purposes to force a read sample on every
539 : // call to Iterator.maybeSampleRead()
540 : forceReadSampling bool
541 : }
542 :
543 1 : func (i *Iterator) findNextEntry(limit []byte) {
544 1 : i.iterValidityState = IterExhausted
545 1 : i.pos = iterPosCurForward
546 1 : if i.opts.rangeKeys() && i.rangeKey != nil {
547 1 : i.rangeKey.rangeKeyOnly = false
548 1 : }
549 :
550 : // Close the closer for the current value if one was open.
551 1 : if i.closeValueCloser() != nil {
552 0 : return
553 0 : }
554 :
555 1 : for i.iterKV != nil {
556 1 : key := i.iterKV.K
557 1 :
558 1 : // The topLevelIterator.StrictSeekPrefixGE contract requires that in
559 1 : // prefix mode [i.hasPrefix=t], every point key returned by the internal
560 1 : // iterator must have the current iteration prefix.
561 1 : if invariants.Enabled && i.hasPrefix {
562 1 : // Range keys are an exception to the contract and may return a different
563 1 : // prefix. This case is explicitly handled in the switch statement below.
564 1 : if key.Kind() != base.InternalKeyKindRangeKeySet {
565 1 : if p := i.comparer.Split.Prefix(key.UserKey); !i.equal(i.prefixOrFullSeekKey, p) {
566 0 : i.opts.logger.Fatalf("pebble: prefix violation: key %q does not have prefix %q\n", key.UserKey, i.prefixOrFullSeekKey)
567 0 : }
568 : }
569 : }
570 :
571 : // Compare with limit every time we start at a different user key.
572 : // Note that given the best-effort contract of limit, we could avoid a
573 : // comparison in the common case by doing this only after
574 : // i.nextUserKey is called for the deletes below. However that makes
575 : // the behavior non-deterministic (since the behavior will vary based
576 : // on what has been compacted), which makes it hard to test with the
577 : // metamorphic test. So we forego that performance optimization.
578 1 : if limit != nil && i.cmp(limit, i.iterKV.K.UserKey) <= 0 {
579 1 : i.iterValidityState = IterAtLimit
580 1 : i.pos = iterPosCurForwardPaused
581 1 : return
582 1 : }
583 :
584 : // If the user has configured a SkipPoint function, invoke it to see
585 : // whether we should skip over the current user key.
586 1 : if i.opts.SkipPoint != nil && key.Kind() != InternalKeyKindRangeKeySet && i.opts.SkipPoint(i.iterKV.K.UserKey) {
587 1 : // NB: We could call nextUserKey, but in some cases the SkipPoint
588 1 : // predicate function might be cheaper than nextUserKey's key copy
589 1 : // and key comparison. This should be the case for MVCC suffix
590 1 : // comparisons, for example. In the future, we could expand the
591 1 : // SkipPoint interface to give the implementor more control over
592 1 : // whether we skip over just the internal key, the user key, or even
593 1 : // the key prefix.
594 1 : i.stats.ForwardStepCount[InternalIterCall]++
595 1 : i.iterKV = i.iter.Next()
596 1 : continue
597 : }
598 :
599 1 : switch key.Kind() {
600 1 : case InternalKeyKindRangeKeySet:
601 1 : if i.hasPrefix {
602 1 : if p := i.comparer.Split.Prefix(key.UserKey); !i.equal(i.prefixOrFullSeekKey, p) {
603 1 : return
604 1 : }
605 : }
606 : // Save the current key.
607 1 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
608 1 : i.key = i.keyBuf
609 1 : i.value = base.InternalValue{}
610 1 : // There may also be a live point key at this userkey that we have
611 1 : // not yet read. We need to find the next entry with this user key
612 1 : // to find it. Save the range key so we don't lose it when we Next
613 1 : // the underlying iterator.
614 1 : i.saveRangeKey()
615 1 : pointKeyExists := i.nextPointCurrentUserKey()
616 1 : if i.err != nil {
617 0 : i.iterValidityState = IterExhausted
618 0 : return
619 0 : }
620 1 : i.rangeKey.rangeKeyOnly = !pointKeyExists
621 1 : i.iterValidityState = IterValid
622 1 : return
623 :
624 1 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
625 1 : // NB: treating InternalKeyKindSingleDelete as equivalent to DEL is not
626 1 : // only simpler, but is also necessary for correctness due to
627 1 : // InternalKeyKindSSTableInternalObsoleteBit.
628 1 : i.nextUserKey()
629 1 : continue
630 :
631 1 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
632 1 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
633 1 : i.key = i.keyBuf
634 1 : i.value = i.iterKV.V
635 1 : i.iterValidityState = IterValid
636 1 : i.saveRangeKey()
637 1 : return
638 :
639 1 : case InternalKeyKindMerge:
640 1 : // Resolving the merge may advance us to the next point key, which
641 1 : // may be covered by a different set of range keys. Save the range
642 1 : // key state so we don't lose it.
643 1 : i.saveRangeKey()
644 1 : if i.mergeForward(key) {
645 1 : i.iterValidityState = IterValid
646 1 : return
647 1 : }
648 :
649 : // The merge didn't yield a valid key, either because the value
650 : // merger indicated it should be deleted, or because an error was
651 : // encountered.
652 1 : i.iterValidityState = IterExhausted
653 1 : if i.err != nil {
654 1 : return
655 1 : }
656 1 : if i.pos != iterPosNext {
657 0 : i.nextUserKey()
658 0 : }
659 1 : if i.closeValueCloser() != nil {
660 0 : return
661 0 : }
662 1 : i.pos = iterPosCurForward
663 :
664 0 : default:
665 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
666 0 : i.iterValidityState = IterExhausted
667 0 : return
668 : }
669 : }
670 :
671 : // Is iterKey nil due to an error?
672 1 : if err := i.iter.Error(); err != nil {
673 1 : i.err = err
674 1 : i.iterValidityState = IterExhausted
675 1 : }
676 : }
677 :
678 1 : func (i *Iterator) nextPointCurrentUserKey() bool {
679 1 : // If the user has configured a SkipPoint function and the current user key
680 1 : // would be skipped by it, there's no need to step forward looking for a
681 1 : // point key. If we were to find one, it should be skipped anyways.
682 1 : if i.opts.SkipPoint != nil && i.opts.SkipPoint(i.key) {
683 1 : return false
684 1 : }
685 :
686 1 : i.pos = iterPosCurForward
687 1 :
688 1 : i.iterKV = i.iter.Next()
689 1 : i.stats.ForwardStepCount[InternalIterCall]++
690 1 : if i.iterKV == nil {
691 1 : if err := i.iter.Error(); err != nil {
692 0 : i.err = err
693 1 : } else {
694 1 : i.pos = iterPosNext
695 1 : }
696 1 : return false
697 : }
698 1 : if !i.equal(i.key, i.iterKV.K.UserKey) {
699 1 : i.pos = iterPosNext
700 1 : return false
701 1 : }
702 :
703 1 : key := i.iterKV.K
704 1 : switch key.Kind() {
705 0 : case InternalKeyKindRangeKeySet:
706 0 : // RangeKeySets must always be interleaved as the first internal key
707 0 : // for a user key.
708 0 : i.err = base.CorruptionErrorf("pebble: unexpected range key set mid-user key")
709 0 : return false
710 :
711 1 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
712 1 : // NB: treating InternalKeyKindSingleDelete as equivalent to DEL is not
713 1 : // only simpler, but is also necessary for correctness due to
714 1 : // InternalKeyKindSSTableInternalObsoleteBit.
715 1 : return false
716 :
717 1 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
718 1 : i.value = i.iterKV.V
719 1 : return true
720 :
721 1 : case InternalKeyKindMerge:
722 1 : return i.mergeForward(key)
723 :
724 0 : default:
725 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
726 0 : return false
727 : }
728 : }
729 :
730 : // mergeForward resolves a MERGE key, advancing the underlying iterator forward
731 : // to merge with subsequent keys with the same userkey. mergeForward returns a
732 : // boolean indicating whether or not the merge yielded a valid key. A merge may
733 : // not yield a valid key if an error occurred, in which case i.err is non-nil,
734 : // or the user's value merger specified the key to be deleted.
735 : //
736 : // mergeForward does not update iterValidityState.
737 1 : func (i *Iterator) mergeForward(key base.InternalKey) (valid bool) {
738 1 : var iterValue []byte
739 1 : iterValue, _, i.err = i.iterKV.Value(nil)
740 1 : if i.err != nil {
741 0 : return false
742 0 : }
743 1 : var valueMerger ValueMerger
744 1 : valueMerger, i.err = i.merge(key.UserKey, iterValue)
745 1 : if i.err != nil {
746 0 : return false
747 0 : }
748 :
749 1 : i.mergeNext(key, valueMerger)
750 1 : if i.err != nil {
751 1 : return false
752 1 : }
753 :
754 1 : var needDelete bool
755 1 : var value []byte
756 1 : value, needDelete, i.valueCloser, i.err = finishValueMerger(
757 1 : valueMerger, true /* includesBase */)
758 1 : i.value = base.MakeInPlaceValue(value)
759 1 : if i.err != nil {
760 0 : return false
761 0 : }
762 1 : if needDelete {
763 1 : _ = i.closeValueCloser()
764 1 : return false
765 1 : }
766 1 : return true
767 : }
768 :
769 1 : func (i *Iterator) closeValueCloser() error {
770 1 : if i.valueCloser != nil {
771 0 : i.err = i.valueCloser.Close()
772 0 : i.valueCloser = nil
773 0 : }
774 1 : return i.err
775 : }
776 :
777 1 : func (i *Iterator) nextUserKey() {
778 1 : if i.iterKV == nil {
779 1 : return
780 1 : }
781 1 : trailer := i.iterKV.K.Trailer
782 1 : done := i.iterKV.K.Trailer <= base.InternalKeyZeroSeqnumMaxTrailer
783 1 : if i.iterValidityState != IterValid {
784 1 : i.keyBuf = append(i.keyBuf[:0], i.iterKV.K.UserKey...)
785 1 : i.key = i.keyBuf
786 1 : }
787 1 : for {
788 1 : i.stats.ForwardStepCount[InternalIterCall]++
789 1 : i.iterKV = i.iter.Next()
790 1 : if i.iterKV == nil {
791 1 : if err := i.iter.Error(); err != nil {
792 1 : i.err = err
793 1 : return
794 1 : }
795 : }
796 : // NB: We're guaranteed to be on the next user key if the previous key
797 : // had a zero sequence number (`done`), or the new key has a trailer
798 : // greater or equal to the previous key's trailer. This is true because
799 : // internal keys with the same user key are sorted by InternalKeyTrailer in
800 : // strictly monotonically descending order. We expect the trailer
801 : // optimization to trigger around 50% of the time with randomly
802 : // distributed writes. We expect it to trigger very frequently when
803 : // iterating through ingested sstables, which contain keys that all have
804 : // the same sequence number.
805 1 : if done || i.iterKV == nil || i.iterKV.K.Trailer >= trailer {
806 1 : break
807 : }
808 1 : if !i.equal(i.key, i.iterKV.K.UserKey) {
809 1 : break
810 : }
811 1 : done = i.iterKV.K.Trailer <= base.InternalKeyZeroSeqnumMaxTrailer
812 1 : trailer = i.iterKV.K.Trailer
813 : }
814 : }
815 :
816 1 : func (i *Iterator) maybeSampleRead() {
817 1 : // This method is only called when a public method of Iterator is
818 1 : // returning, and below we exclude the case were the iterator is paused at
819 1 : // a limit. The effect of these choices is that keys that are deleted, but
820 1 : // are encountered during iteration, are not accounted for in the read
821 1 : // sampling and will not cause read driven compactions, even though we are
822 1 : // incurring cost in iterating over them. And this issue is not limited to
823 1 : // Iterator, which does not see the effect of range deletes, which may be
824 1 : // causing iteration work in mergingIter. It is not clear at this time
825 1 : // whether this is a deficiency worth addressing.
826 1 : if i.iterValidityState != IterValid {
827 1 : return
828 1 : }
829 1 : if i.readState == nil {
830 1 : return
831 1 : }
832 1 : if i.readSampling.forceReadSampling {
833 1 : i.sampleRead()
834 1 : return
835 1 : }
836 1 : samplingPeriod := int32(int64(readBytesPeriod) * i.readState.db.opts.Experimental.ReadSamplingMultiplier)
837 1 : if samplingPeriod <= 0 {
838 0 : return
839 0 : }
840 1 : bytesRead := uint64(len(i.key) + i.value.Len())
841 1 : for i.readSampling.bytesUntilReadSampling < bytesRead {
842 1 : i.readSampling.bytesUntilReadSampling += uint64(rand.Uint32N(2 * uint32(samplingPeriod)))
843 1 : // The block below tries to adjust for the case where this is the
844 1 : // first read in a newly-opened iterator. As bytesUntilReadSampling
845 1 : // starts off at zero, we don't want to sample the first read of
846 1 : // every newly-opened iterator, but we do want to sample some of them.
847 1 : if !i.readSampling.initialSamplePassed {
848 1 : i.readSampling.initialSamplePassed = true
849 1 : if i.readSampling.bytesUntilReadSampling > bytesRead {
850 1 : if rand.Uint64N(i.readSampling.bytesUntilReadSampling) > bytesRead {
851 1 : continue
852 : }
853 : }
854 : }
855 1 : i.sampleRead()
856 : }
857 1 : i.readSampling.bytesUntilReadSampling -= bytesRead
858 : }
859 :
860 1 : func (i *Iterator) sampleRead() {
861 1 : var topFile *manifest.TableMetadata
862 1 : topLevel, numOverlappingLevels := numLevels, 0
863 1 : mi := i.merging
864 1 : if mi == nil {
865 1 : return
866 1 : }
867 1 : if len(mi.levels) > 1 {
868 1 : mi.ForEachLevelIter(func(li *levelIter) (done bool) {
869 1 : if li.layer.IsFlushableIngests() {
870 0 : return false
871 0 : }
872 1 : l := li.layer.Level()
873 1 : if f := li.iterFile; f != nil {
874 1 : var containsKey bool
875 1 : if i.pos == iterPosNext || i.pos == iterPosCurForward ||
876 1 : i.pos == iterPosCurForwardPaused {
877 1 : containsKey = i.cmp(f.PointKeyBounds.SmallestUserKey(), i.key) <= 0
878 1 : } else if i.pos == iterPosPrev || i.pos == iterPosCurReverse ||
879 1 : i.pos == iterPosCurReversePaused {
880 1 : containsKey = i.cmp(f.PointKeyBounds.LargestUserKey(), i.key) >= 0
881 1 : }
882 : // Do nothing if the current key is not contained in f's
883 : // bounds. We could seek the LevelIterator at this level
884 : // to find the right file, but the performance impacts of
885 : // doing that are significant enough to negate the benefits
886 : // of read sampling in the first place. See the discussion
887 : // at:
888 : // https://github.com/cockroachdb/pebble/pull/1041#issuecomment-763226492
889 1 : if containsKey {
890 1 : numOverlappingLevels++
891 1 : if numOverlappingLevels >= 2 {
892 1 : // Terminate the loop early if at least 2 overlapping levels are found.
893 1 : return true
894 1 : }
895 1 : topLevel = l
896 1 : topFile = f
897 : }
898 : }
899 1 : return false
900 : })
901 : }
902 1 : if topFile == nil || topLevel >= numLevels {
903 1 : return
904 1 : }
905 1 : if numOverlappingLevels >= 2 {
906 1 : allowedSeeks := topFile.AllowedSeeks.Add(-1)
907 1 : if allowedSeeks == 0 {
908 1 :
909 1 : // Since the compaction queue can handle duplicates, we can keep
910 1 : // adding to the queue even once allowedSeeks hits 0.
911 1 : // In fact, we NEED to keep adding to the queue, because the queue
912 1 : // is small and evicts older and possibly useful compactions.
913 1 : topFile.AllowedSeeks.Add(topFile.InitAllowedSeeks)
914 1 :
915 1 : read := readCompaction{
916 1 : start: topFile.PointKeyBounds.SmallestUserKey(),
917 1 : end: topFile.PointKeyBounds.LargestUserKey(),
918 1 : level: topLevel,
919 1 : tableNum: topFile.TableNum,
920 1 : }
921 1 : i.readSampling.pendingCompactions.add(&read, i.cmp)
922 1 : }
923 : }
924 : }
925 :
926 1 : func (i *Iterator) findPrevEntry(limit []byte) {
927 1 : i.iterValidityState = IterExhausted
928 1 : i.pos = iterPosCurReverse
929 1 : if i.opts.rangeKeys() && i.rangeKey != nil {
930 1 : i.rangeKey.rangeKeyOnly = false
931 1 : }
932 :
933 : // Close the closer for the current value if one was open.
934 1 : if i.valueCloser != nil {
935 0 : i.err = i.valueCloser.Close()
936 0 : i.valueCloser = nil
937 0 : if i.err != nil {
938 0 : i.iterValidityState = IterExhausted
939 0 : return
940 0 : }
941 : }
942 :
943 1 : var valueMerger ValueMerger
944 1 : firstLoopIter := true
945 1 : rangeKeyBoundary := false
946 1 : // The code below compares with limit in multiple places. As documented in
947 1 : // findNextEntry, this is being done to make the behavior of limit
948 1 : // deterministic to allow for metamorphic testing. It is not required by
949 1 : // the best-effort contract of limit.
950 1 : for i.iterKV != nil {
951 1 : key := i.iterKV.K
952 1 :
953 1 : // NB: We cannot pause if the current key is covered by a range key.
954 1 : // Otherwise, the user might not ever learn of a range key that covers
955 1 : // the key space being iterated over in which there are no point keys.
956 1 : // Since limits are best effort, ignoring the limit in this case is
957 1 : // allowed by the contract of limit.
958 1 : if firstLoopIter && limit != nil && i.cmp(limit, i.iterKV.K.UserKey) > 0 && !i.rangeKeyWithinLimit(limit) {
959 1 : i.iterValidityState = IterAtLimit
960 1 : i.pos = iterPosCurReversePaused
961 1 : return
962 1 : }
963 1 : firstLoopIter = false
964 1 :
965 1 : if i.iterValidityState == IterValid {
966 1 : if !i.equal(key.UserKey, i.key) {
967 1 : // We've iterated to the previous user key.
968 1 : i.pos = iterPosPrev
969 1 : if valueMerger != nil {
970 1 : var needDelete bool
971 1 : var value []byte
972 1 : value, needDelete, i.valueCloser, i.err = finishValueMerger(valueMerger, true /* includesBase */)
973 1 : i.value = base.MakeInPlaceValue(value)
974 1 : if i.err == nil && needDelete {
975 1 : // The point key at this key is deleted. If we also have
976 1 : // a range key boundary at this key, we still want to
977 1 : // return. Otherwise, we need to continue looking for
978 1 : // a live key.
979 1 : i.value = base.InternalValue{}
980 1 : if rangeKeyBoundary {
981 0 : i.rangeKey.rangeKeyOnly = true
982 1 : } else {
983 1 : i.iterValidityState = IterExhausted
984 1 : if i.closeValueCloser() == nil {
985 1 : continue
986 : }
987 : }
988 : }
989 : }
990 1 : if i.err != nil {
991 0 : i.iterValidityState = IterExhausted
992 0 : }
993 1 : return
994 : }
995 : }
996 :
997 : // If the user has configured a SkipPoint function, invoke it to see
998 : // whether we should skip over the current user key.
999 1 : if i.opts.SkipPoint != nil && key.Kind() != InternalKeyKindRangeKeySet && i.opts.SkipPoint(key.UserKey) {
1000 1 : // NB: We could call prevUserKey, but in some cases the SkipPoint
1001 1 : // predicate function might be cheaper than prevUserKey's key copy
1002 1 : // and key comparison. This should be the case for MVCC suffix
1003 1 : // comparisons, for example. In the future, we could expand the
1004 1 : // SkipPoint interface to give the implementor more control over
1005 1 : // whether we skip over just the internal key, the user key, or even
1006 1 : // the key prefix.
1007 1 : i.stats.ReverseStepCount[InternalIterCall]++
1008 1 : i.iterKV = i.iter.Prev()
1009 1 : if i.iterKV == nil {
1010 1 : if err := i.iter.Error(); err != nil {
1011 0 : i.err = err
1012 0 : i.iterValidityState = IterExhausted
1013 0 : return
1014 0 : }
1015 : }
1016 1 : if limit != nil && i.iterKV != nil && i.cmp(limit, i.iterKV.K.UserKey) > 0 && !i.rangeKeyWithinLimit(limit) {
1017 1 : i.iterValidityState = IterAtLimit
1018 1 : i.pos = iterPosCurReversePaused
1019 1 : return
1020 1 : }
1021 1 : continue
1022 : }
1023 :
1024 1 : switch key.Kind() {
1025 1 : case InternalKeyKindRangeKeySet:
1026 1 : // Range key start boundary markers are interleaved with the maximum
1027 1 : // sequence number, so if there's a point key also at this key, we
1028 1 : // must've already iterated over it.
1029 1 : // This is the final entry at this user key, so we may return
1030 1 : i.rangeKey.rangeKeyOnly = i.iterValidityState != IterValid
1031 1 : if i.rangeKey.rangeKeyOnly {
1032 1 : // The point iterator is now invalid, so clear the point value.
1033 1 : i.value = base.InternalValue{}
1034 1 : }
1035 1 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1036 1 : i.key = i.keyBuf
1037 1 : i.iterValidityState = IterValid
1038 1 : i.saveRangeKey()
1039 1 : // In all other cases, previous iteration requires advancing to
1040 1 : // iterPosPrev in order to determine if the key is live and
1041 1 : // unshadowed by another key at the same user key. In this case,
1042 1 : // because range key start boundary markers are always interleaved
1043 1 : // at the maximum sequence number, we know that there aren't any
1044 1 : // additional keys with the same user key in the backward direction.
1045 1 : //
1046 1 : // We Prev the underlying iterator once anyways for consistency, so
1047 1 : // that we can maintain the invariant during backward iteration that
1048 1 : // i.iterPos = iterPosPrev.
1049 1 : i.stats.ReverseStepCount[InternalIterCall]++
1050 1 : i.iterKV = i.iter.Prev()
1051 1 :
1052 1 : // Set rangeKeyBoundary so that on the next iteration, we know to
1053 1 : // return the key even if the MERGE point key is deleted.
1054 1 : rangeKeyBoundary = true
1055 :
1056 1 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
1057 1 : i.value = base.InternalValue{}
1058 1 : i.iterValidityState = IterExhausted
1059 1 : valueMerger = nil
1060 1 : i.stats.ReverseStepCount[InternalIterCall]++
1061 1 : i.iterKV = i.iter.Prev()
1062 1 : // Compare with the limit. We could optimize by only checking when
1063 1 : // we step to the previous user key, but detecting that requires a
1064 1 : // comparison too. Note that this position may already passed a
1065 1 : // number of versions of this user key, but they are all deleted, so
1066 1 : // the fact that a subsequent Prev*() call will not see them is
1067 1 : // harmless. Also note that this is the only place in the loop,
1068 1 : // other than the firstLoopIter and SkipPoint cases above, where we
1069 1 : // could step to a different user key and start processing it for
1070 1 : // returning to the caller.
1071 1 : if limit != nil && i.iterKV != nil && i.cmp(limit, i.iterKV.K.UserKey) > 0 && !i.rangeKeyWithinLimit(limit) {
1072 1 : i.iterValidityState = IterAtLimit
1073 1 : i.pos = iterPosCurReversePaused
1074 1 : return
1075 1 : }
1076 1 : continue
1077 :
1078 1 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
1079 1 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1080 1 : i.key = i.keyBuf
1081 1 : // iterValue is owned by i.iter and could change after the Prev()
1082 1 : // call, so use valueBuf instead. Note that valueBuf is only used
1083 1 : // in this one instance; everywhere else (eg. in findNextEntry),
1084 1 : // we just point i.value to the unsafe i.iter-owned value buffer.
1085 1 : i.value, i.valueBuf = i.iterKV.V.Clone(i.valueBuf[:0], &i.fetcher)
1086 1 : i.saveRangeKey()
1087 1 : i.iterValidityState = IterValid
1088 1 : i.iterKV = i.iter.Prev()
1089 1 : i.stats.ReverseStepCount[InternalIterCall]++
1090 1 : valueMerger = nil
1091 1 : continue
1092 :
1093 1 : case InternalKeyKindMerge:
1094 1 : if i.iterValidityState == IterExhausted {
1095 1 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1096 1 : i.key = i.keyBuf
1097 1 : i.saveRangeKey()
1098 1 : var iterValue []byte
1099 1 : iterValue, _, i.err = i.iterKV.Value(nil)
1100 1 : if i.err != nil {
1101 0 : return
1102 0 : }
1103 1 : valueMerger, i.err = i.merge(i.key, iterValue)
1104 1 : if i.err != nil {
1105 0 : return
1106 0 : }
1107 1 : i.iterValidityState = IterValid
1108 1 : } else if valueMerger == nil {
1109 1 : // Extract value before iterValue since we use value before iterValue
1110 1 : // and the underlying iterator is not required to provide backing
1111 1 : // memory for both simultaneously.
1112 1 : var value []byte
1113 1 : var callerOwned bool
1114 1 : value, callerOwned, i.err = i.value.Value(i.lazyValueBuf)
1115 1 : if callerOwned {
1116 0 : i.lazyValueBuf = value[:0]
1117 0 : }
1118 1 : if i.err != nil {
1119 0 : i.iterValidityState = IterExhausted
1120 0 : return
1121 0 : }
1122 1 : valueMerger, i.err = i.merge(i.key, value)
1123 1 : var iterValue []byte
1124 1 : iterValue, _, i.err = i.iterKV.Value(nil)
1125 1 : if i.err != nil {
1126 0 : i.iterValidityState = IterExhausted
1127 0 : return
1128 0 : }
1129 1 : if i.err == nil {
1130 1 : i.err = valueMerger.MergeNewer(iterValue)
1131 1 : }
1132 1 : if i.err != nil {
1133 0 : i.iterValidityState = IterExhausted
1134 0 : return
1135 0 : }
1136 1 : } else {
1137 1 : var iterValue []byte
1138 1 : iterValue, _, i.err = i.iterKV.Value(nil)
1139 1 : if i.err != nil {
1140 0 : i.iterValidityState = IterExhausted
1141 0 : return
1142 0 : }
1143 1 : i.err = valueMerger.MergeNewer(iterValue)
1144 1 : if i.err != nil {
1145 0 : i.iterValidityState = IterExhausted
1146 0 : return
1147 0 : }
1148 : }
1149 1 : i.iterKV = i.iter.Prev()
1150 1 : i.stats.ReverseStepCount[InternalIterCall]++
1151 1 : continue
1152 :
1153 0 : default:
1154 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
1155 0 : i.iterValidityState = IterExhausted
1156 0 : return
1157 : }
1158 : }
1159 : // i.iterKV == nil, so broke out of the preceding loop.
1160 :
1161 : // Is iterKey nil due to an error?
1162 1 : if i.err = i.iter.Error(); i.err != nil {
1163 1 : i.iterValidityState = IterExhausted
1164 1 : return
1165 1 : }
1166 :
1167 1 : if i.iterValidityState == IterValid {
1168 1 : i.pos = iterPosPrev
1169 1 : if valueMerger != nil {
1170 1 : var needDelete bool
1171 1 : var value []byte
1172 1 : value, needDelete, i.valueCloser, i.err = finishValueMerger(valueMerger, true /* includesBase */)
1173 1 : i.value = base.MakeInPlaceValue(value)
1174 1 : if i.err == nil && needDelete {
1175 1 : i.key = nil
1176 1 : i.value = base.InternalValue{}
1177 1 : i.iterValidityState = IterExhausted
1178 1 : }
1179 : }
1180 1 : if i.err != nil {
1181 0 : i.iterValidityState = IterExhausted
1182 0 : }
1183 : }
1184 : }
1185 :
1186 1 : func (i *Iterator) prevUserKey() {
1187 1 : if i.iterKV == nil {
1188 1 : return
1189 1 : }
1190 1 : if i.iterValidityState != IterValid {
1191 1 : // If we're going to compare against the prev key, we need to save the
1192 1 : // current key.
1193 1 : i.keyBuf = append(i.keyBuf[:0], i.iterKV.K.UserKey...)
1194 1 : i.key = i.keyBuf
1195 1 : }
1196 1 : for {
1197 1 : i.iterKV = i.iter.Prev()
1198 1 : i.stats.ReverseStepCount[InternalIterCall]++
1199 1 : if i.iterKV == nil {
1200 1 : if err := i.iter.Error(); err != nil {
1201 0 : i.err = err
1202 0 : i.iterValidityState = IterExhausted
1203 0 : }
1204 1 : break
1205 : }
1206 1 : if !i.equal(i.key, i.iterKV.K.UserKey) {
1207 1 : break
1208 : }
1209 : }
1210 : }
1211 :
1212 1 : func (i *Iterator) mergeNext(key InternalKey, valueMerger ValueMerger) {
1213 1 : // Save the current key.
1214 1 : i.keyBuf = append(i.keyBuf[:0], key.UserKey...)
1215 1 : i.key = i.keyBuf
1216 1 :
1217 1 : // Loop looking for older values for this key and merging them.
1218 1 : for {
1219 1 : i.iterKV = i.iter.Next()
1220 1 : i.stats.ForwardStepCount[InternalIterCall]++
1221 1 : if i.iterKV == nil {
1222 1 : if i.err = i.iter.Error(); i.err != nil {
1223 1 : return
1224 1 : }
1225 1 : i.pos = iterPosNext
1226 1 : return
1227 : }
1228 1 : key = i.iterKV.K
1229 1 : if !i.equal(i.key, key.UserKey) {
1230 1 : // We've advanced to the next key.
1231 1 : i.pos = iterPosNext
1232 1 : return
1233 1 : }
1234 1 : switch key.Kind() {
1235 1 : case InternalKeyKindDelete, InternalKeyKindSingleDelete, InternalKeyKindDeleteSized:
1236 1 : // We've hit a deletion tombstone. Return everything up to this
1237 1 : // point.
1238 1 : //
1239 1 : // NB: treating InternalKeyKindSingleDelete as equivalent to DEL is not
1240 1 : // only simpler, but is also necessary for correctness due to
1241 1 : // InternalKeyKindSSTableInternalObsoleteBit.
1242 1 : return
1243 :
1244 1 : case InternalKeyKindSet, InternalKeyKindSetWithDelete:
1245 1 : // We've hit a Set value. Merge with the existing value and return.
1246 1 : var iterValue []byte
1247 1 : iterValue, _, i.err = i.iterKV.Value(nil)
1248 1 : if i.err != nil {
1249 0 : return
1250 0 : }
1251 1 : i.err = valueMerger.MergeOlder(iterValue)
1252 1 : return
1253 :
1254 1 : case InternalKeyKindMerge:
1255 1 : // We've hit another Merge value. Merge with the existing value and
1256 1 : // continue looping.
1257 1 : var iterValue []byte
1258 1 : iterValue, _, i.err = i.iterKV.Value(nil)
1259 1 : if i.err != nil {
1260 0 : return
1261 0 : }
1262 1 : i.err = valueMerger.MergeOlder(iterValue)
1263 1 : if i.err != nil {
1264 0 : return
1265 0 : }
1266 1 : continue
1267 :
1268 0 : case InternalKeyKindRangeKeySet:
1269 0 : // The RANGEKEYSET marker must sort before a MERGE at the same user key.
1270 0 : i.err = base.CorruptionErrorf("pebble: out of order range key marker")
1271 0 : return
1272 :
1273 0 : default:
1274 0 : i.err = base.CorruptionErrorf("pebble: invalid internal key kind: %d", errors.Safe(key.Kind()))
1275 0 : return
1276 : }
1277 : }
1278 : }
1279 :
1280 : // SeekGE moves the iterator to the first key/value pair whose key is greater
1281 : // than or equal to the given key. Returns true if the iterator is pointing at
1282 : // a valid entry and false otherwise.
1283 1 : func (i *Iterator) SeekGE(key []byte) bool {
1284 1 : return i.SeekGEWithLimit(key, nil) == IterValid
1285 1 : }
1286 :
1287 : // SeekGEWithLimit moves the iterator to the first key/value pair whose key is
1288 : // greater than or equal to the given key.
1289 : //
1290 : // If limit is provided, it serves as a best-effort exclusive limit. If the
1291 : // first key greater than or equal to the given search key is also greater than
1292 : // or equal to limit, the Iterator may pause and return IterAtLimit. Because
1293 : // limits are best-effort, SeekGEWithLimit may return a key beyond limit.
1294 : //
1295 : // If the Iterator is configured to iterate over range keys, SeekGEWithLimit
1296 : // guarantees it will surface any range keys with bounds overlapping the
1297 : // keyspace [key, limit).
1298 1 : func (i *Iterator) SeekGEWithLimit(key []byte, limit []byte) IterValidityState {
1299 1 : if i.rangeKey != nil {
1300 1 : // NB: Check Valid() before clearing requiresReposition.
1301 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1302 1 : // If we have a range key but did not expose it at the previous iterator
1303 1 : // position (because the iterator was not at a valid position), updated
1304 1 : // must be true. This ensures that after an iterator op sequence like:
1305 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1306 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1307 1 : // - SeekGE(...) → (IterValid, RangeBounds() = [a,b))
1308 1 : // the iterator returns RangeKeyChanged()=true.
1309 1 : //
1310 1 : // The remainder of this function will only update i.rangeKey.updated if
1311 1 : // the iterator moves into a new range key, or out of the current range
1312 1 : // key.
1313 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1314 1 : }
1315 1 : lastPositioningOp := i.lastPositioningOp
1316 1 : // Set it to unknown, since this operation may not succeed, in which case
1317 1 : // the SeekGE following this should not make any assumption about iterator
1318 1 : // position.
1319 1 : i.lastPositioningOp = unknownLastPositionOp
1320 1 : i.requiresReposition = false
1321 1 : i.err = nil // clear cached iteration error
1322 1 : i.hasPrefix = false
1323 1 : i.stats.ForwardSeekCount[InterfaceCall]++
1324 1 : if lowerBound := i.opts.GetLowerBound(); lowerBound != nil && i.cmp(key, lowerBound) < 0 {
1325 1 : key = lowerBound
1326 1 : } else if upperBound := i.opts.GetUpperBound(); upperBound != nil && i.cmp(key, upperBound) > 0 {
1327 1 : key = upperBound
1328 1 : }
1329 1 : seekInternalIter := true
1330 1 :
1331 1 : var flags base.SeekGEFlags
1332 1 : if i.batchJustRefreshed {
1333 1 : i.batchJustRefreshed = false
1334 1 : flags = flags.EnableBatchJustRefreshed()
1335 1 : }
1336 1 : if lastPositioningOp == seekGELastPositioningOp {
1337 1 : cmp := i.cmp(i.prefixOrFullSeekKey, key)
1338 1 : // If this seek is to the same or later key, and the iterator is
1339 1 : // already positioned there, this is a noop. This can be helpful for
1340 1 : // sparse key spaces that have many deleted keys, where one can avoid
1341 1 : // the overhead of iterating past them again and again.
1342 1 : if cmp <= 0 {
1343 1 : if !flags.BatchJustRefreshed() &&
1344 1 : (i.iterValidityState == IterExhausted ||
1345 1 : (i.iterValidityState == IterValid && i.cmp(key, i.key) <= 0 &&
1346 1 : (limit == nil || i.cmp(i.key, limit) < 0))) {
1347 1 : // Noop
1348 1 : if i.forceEnableSeekOpt || !testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) {
1349 1 : i.lastPositioningOp = seekGELastPositioningOp
1350 1 : return i.iterValidityState
1351 1 : }
1352 : }
1353 : // cmp == 0 is not safe to optimize since
1354 : // - i.pos could be at iterPosNext, due to a merge.
1355 : // - Even if i.pos were at iterPosCurForward, we could have a DELETE,
1356 : // SET pair for a key, and the iterator would have moved past DELETE
1357 : // but stayed at iterPosCurForward. A similar situation occurs for a
1358 : // MERGE, SET pair where the MERGE is consumed and the iterator is
1359 : // at the SET.
1360 : // We also leverage the IterAtLimit <=> i.pos invariant defined in the
1361 : // comment on iterValidityState, to exclude any cases where i.pos
1362 : // is iterPosCur{Forward,Reverse}Paused. This avoids the need to
1363 : // special-case those iterator positions and their interactions with
1364 : // TrySeekUsingNext, as the main uses for TrySeekUsingNext in CockroachDB
1365 : // do not use limited Seeks in the first place.
1366 1 : if cmp < 0 && i.iterValidityState != IterAtLimit && limit == nil {
1367 1 : flags = flags.EnableTrySeekUsingNext()
1368 1 : }
1369 1 : if testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) && !i.forceEnableSeekOpt {
1370 1 : flags = flags.DisableTrySeekUsingNext()
1371 1 : }
1372 1 : if !flags.BatchJustRefreshed() && i.pos == iterPosCurForwardPaused && i.cmp(key, i.iterKV.K.UserKey) <= 0 {
1373 1 : // Have some work to do, but don't need to seek, and we can
1374 1 : // start doing findNextEntry from i.iterKey.
1375 1 : seekInternalIter = false
1376 1 : }
1377 : }
1378 : }
1379 1 : if seekInternalIter {
1380 1 : i.iterKV = i.iter.SeekGE(key, flags)
1381 1 : i.stats.ForwardSeekCount[InternalIterCall]++
1382 1 : if err := i.iter.Error(); err != nil {
1383 1 : i.err = err
1384 1 : i.iterValidityState = IterExhausted
1385 1 : return i.iterValidityState
1386 1 : }
1387 : }
1388 1 : i.findNextEntry(limit)
1389 1 : i.maybeSampleRead()
1390 1 : if i.Error() == nil {
1391 1 : // Prepare state for a future noop optimization.
1392 1 : i.prefixOrFullSeekKey = append(i.prefixOrFullSeekKey[:0], key...)
1393 1 : i.lastPositioningOp = seekGELastPositioningOp
1394 1 : }
1395 1 : return i.iterValidityState
1396 : }
1397 :
1398 : // SeekPrefixGE moves the iterator to the first key/value pair whose key is
1399 : // greater than or equal to the given key and which has the same "prefix" as
1400 : // the given key. The prefix for a key is determined by the user-defined
1401 : // Comparer.Split function. The iterator will not observe keys not matching the
1402 : // "prefix" of the search key. Calling SeekPrefixGE puts the iterator in prefix
1403 : // iteration mode. The iterator remains in prefix iteration until a subsequent
1404 : // call to another absolute positioning method (SeekGE, SeekLT, First,
1405 : // Last). Reverse iteration (Prev) is not supported when an iterator is in
1406 : // prefix iteration mode. Returns true if the iterator is pointing at a valid
1407 : // entry and false otherwise.
1408 : //
1409 : // The semantics of SeekPrefixGE are slightly unusual and designed for
1410 : // iteration to be able to take advantage of bloom filters that have been
1411 : // created on the "prefix". If you're not using bloom filters, there is no
1412 : // reason to use SeekPrefixGE.
1413 : //
1414 : // An example Split function may separate a timestamp suffix from the prefix of
1415 : // the key.
1416 : //
1417 : // Split(<key>@<timestamp>) -> <key>
1418 : //
1419 : // Consider the keys "a@1", "a@2", "aa@3", "aa@4". The prefixes for these keys
1420 : // are "a", and "aa". Note that despite "a" and "aa" sharing a prefix by the
1421 : // usual definition, those prefixes differ by the definition of the Split
1422 : // function. To see how this works, consider the following set of calls on this
1423 : // data set:
1424 : //
1425 : // SeekPrefixGE("a@0") -> "a@1"
1426 : // Next() -> "a@2"
1427 : // Next() -> EOF
1428 : //
1429 : // If you're just looking to iterate over keys with a shared prefix, as
1430 : // defined by the configured comparer, set iterator bounds instead:
1431 : //
1432 : // iter := db.NewIter(&pebble.IterOptions{
1433 : // LowerBound: []byte("prefix"),
1434 : // UpperBound: []byte("prefiy"),
1435 : // })
1436 : // for iter.First(); iter.Valid(); iter.Next() {
1437 : // // Only keys beginning with "prefix" will be visited.
1438 : // }
1439 : //
1440 : // See ExampleIterator_SeekPrefixGE for a working example.
1441 : //
1442 : // When iterating with range keys enabled, all range keys encountered are
1443 : // truncated to the seek key's prefix's bounds. The truncation of the upper
1444 : // bound requires that the database's Comparer is configured with a
1445 : // ImmediateSuccessor method. For example, a SeekPrefixGE("a@9") call with the
1446 : // prefix "a" will truncate range key bounds to [a,ImmediateSuccessor(a)].
1447 1 : func (i *Iterator) SeekPrefixGE(key []byte) bool {
1448 1 : if i.rangeKey != nil {
1449 1 : // NB: Check Valid() before clearing requiresReposition.
1450 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1451 1 : // If we have a range key but did not expose it at the previous iterator
1452 1 : // position (because the iterator was not at a valid position), updated
1453 1 : // must be true. This ensures that after an iterator op sequence like:
1454 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1455 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1456 1 : // - SeekPrefixGE(...) → (IterValid, RangeBounds() = [a,b))
1457 1 : // the iterator returns RangeKeyChanged()=true.
1458 1 : //
1459 1 : // The remainder of this function will only update i.rangeKey.updated if
1460 1 : // the iterator moves into a new range key, or out of the current range
1461 1 : // key.
1462 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1463 1 : }
1464 1 : lastPositioningOp := i.lastPositioningOp
1465 1 : // Set it to unknown, since this operation may not succeed, in which case
1466 1 : // the SeekPrefixGE following this should not make any assumption about
1467 1 : // iterator position.
1468 1 : i.lastPositioningOp = unknownLastPositionOp
1469 1 : i.requiresReposition = false
1470 1 : i.err = nil // clear cached iteration error
1471 1 : i.stats.ForwardSeekCount[InterfaceCall]++
1472 1 : if i.comparer.ImmediateSuccessor == nil && i.opts.KeyTypes != IterKeyTypePointsOnly {
1473 0 : panic("pebble: ImmediateSuccessor must be provided for SeekPrefixGE with range keys")
1474 : }
1475 1 : prefixLen := i.comparer.Split(key)
1476 1 : keyPrefix := key[:prefixLen]
1477 1 : var flags base.SeekGEFlags
1478 1 : if i.batchJustRefreshed {
1479 1 : flags = flags.EnableBatchJustRefreshed()
1480 1 : i.batchJustRefreshed = false
1481 1 : }
1482 1 : if lastPositioningOp == seekPrefixGELastPositioningOp {
1483 1 : if !i.hasPrefix {
1484 0 : panic("lastPositioningOpsIsSeekPrefixGE is true, but hasPrefix is false")
1485 : }
1486 : // The iterator has not been repositioned after the last SeekPrefixGE.
1487 : // See if we are seeking to a larger key, since then we can optimize
1488 : // the seek by using next. Note that we could also optimize if Next
1489 : // has been called, if the iterator is not exhausted and the current
1490 : // position is <= the seek key. We are keeping this limited for now
1491 : // since such optimizations require care for correctness, and to not
1492 : // become de-optimizations (if one usually has to do all the next
1493 : // calls and then the seek). This SeekPrefixGE optimization
1494 : // specifically benefits CockroachDB.
1495 1 : cmp := i.cmp(i.prefixOrFullSeekKey, keyPrefix)
1496 1 : // cmp == 0 is not safe to optimize since
1497 1 : // - i.pos could be at iterPosNext, due to a merge.
1498 1 : // - Even if i.pos were at iterPosCurForward, we could have a DELETE,
1499 1 : // SET pair for a key, and the iterator would have moved past DELETE
1500 1 : // but stayed at iterPosCurForward. A similar situation occurs for a
1501 1 : // MERGE, SET pair where the MERGE is consumed and the iterator is
1502 1 : // at the SET.
1503 1 : // In general some versions of i.prefix could have been consumed by
1504 1 : // the iterator, so we only optimize for cmp < 0.
1505 1 : if cmp < 0 {
1506 1 : flags = flags.EnableTrySeekUsingNext()
1507 1 : }
1508 1 : if testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) && !i.forceEnableSeekOpt {
1509 1 : flags = flags.DisableTrySeekUsingNext()
1510 1 : }
1511 : }
1512 : // Make a copy of the prefix so that modifications to the key after
1513 : // SeekPrefixGE returns does not affect the stored prefix.
1514 1 : if cap(i.prefixOrFullSeekKey) < prefixLen {
1515 1 : i.prefixOrFullSeekKey = make([]byte, prefixLen)
1516 1 : } else {
1517 1 : i.prefixOrFullSeekKey = i.prefixOrFullSeekKey[:prefixLen]
1518 1 : }
1519 1 : i.hasPrefix = true
1520 1 : copy(i.prefixOrFullSeekKey, keyPrefix)
1521 1 :
1522 1 : if lowerBound := i.opts.GetLowerBound(); lowerBound != nil && i.cmp(key, lowerBound) < 0 {
1523 1 : if p := i.comparer.Split.Prefix(lowerBound); !bytes.Equal(i.prefixOrFullSeekKey, p) {
1524 1 : i.err = errors.New("pebble: SeekPrefixGE supplied with key outside of lower bound")
1525 1 : i.iterValidityState = IterExhausted
1526 1 : return false
1527 1 : }
1528 1 : key = lowerBound
1529 1 : } else if upperBound := i.opts.GetUpperBound(); upperBound != nil && i.cmp(key, upperBound) > 0 {
1530 1 : if p := i.comparer.Split.Prefix(upperBound); !bytes.Equal(i.prefixOrFullSeekKey, p) {
1531 1 : i.err = errors.New("pebble: SeekPrefixGE supplied with key outside of upper bound")
1532 1 : i.iterValidityState = IterExhausted
1533 1 : return false
1534 1 : }
1535 1 : key = upperBound
1536 : }
1537 1 : i.iterKV = i.iter.SeekPrefixGE(i.prefixOrFullSeekKey, key, flags)
1538 1 : i.stats.ForwardSeekCount[InternalIterCall]++
1539 1 : i.findNextEntry(nil)
1540 1 : i.maybeSampleRead()
1541 1 : if i.Error() == nil {
1542 1 : i.lastPositioningOp = seekPrefixGELastPositioningOp
1543 1 : }
1544 1 : return i.iterValidityState == IterValid
1545 : }
1546 :
1547 : // Deterministic disabling (in testing mode) of the seek optimizations. It uses
1548 : // the iterator pointer, since we want diversity in iterator behavior for the
1549 : // same key. Used for tests.
1550 1 : func testingDisableSeekOpt(key []byte, ptr uintptr) bool {
1551 1 : if !invariants.Enabled {
1552 0 : return false
1553 0 : }
1554 : // Fibonacci hash https://probablydance.com/2018/06/16/fibonacci-hashing-the-optimization-that-the-world-forgot-or-a-better-alternative-to-integer-modulo/
1555 1 : simpleHash := (11400714819323198485 * uint64(ptr)) >> 63
1556 1 : return key != nil && key[0]&byte(1) == 0 && simpleHash == 0
1557 : }
1558 :
1559 : // SeekLT moves the iterator to the last key/value pair whose key is less than
1560 : // the given key. Returns true if the iterator is pointing at a valid entry and
1561 : // false otherwise.
1562 1 : func (i *Iterator) SeekLT(key []byte) bool {
1563 1 : return i.SeekLTWithLimit(key, nil) == IterValid
1564 1 : }
1565 :
1566 : // SeekLTWithLimit moves the iterator to the last key/value pair whose key is
1567 : // less than the given key.
1568 : //
1569 : // If limit is provided, it serves as a best-effort inclusive limit. If the last
1570 : // key less than the given search key is also less than limit, the Iterator may
1571 : // pause and return IterAtLimit. Because limits are best-effort, SeekLTWithLimit
1572 : // may return a key beyond limit.
1573 : //
1574 : // If the Iterator is configured to iterate over range keys, SeekLTWithLimit
1575 : // guarantees it will surface any range keys with bounds overlapping the
1576 : // keyspace up to limit.
1577 1 : func (i *Iterator) SeekLTWithLimit(key []byte, limit []byte) IterValidityState {
1578 1 : if i.rangeKey != nil {
1579 1 : // NB: Check Valid() before clearing requiresReposition.
1580 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1581 1 : // If we have a range key but did not expose it at the previous iterator
1582 1 : // position (because the iterator was not at a valid position), updated
1583 1 : // must be true. This ensures that after an iterator op sequence like:
1584 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1585 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1586 1 : // - SeekLTWithLimit(...) → (IterValid, RangeBounds() = [a,b))
1587 1 : // the iterator returns RangeKeyChanged()=true.
1588 1 : //
1589 1 : // The remainder of this function will only update i.rangeKey.updated if
1590 1 : // the iterator moves into a new range key, or out of the current range
1591 1 : // key.
1592 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1593 1 : }
1594 1 : lastPositioningOp := i.lastPositioningOp
1595 1 : // Set it to unknown, since this operation may not succeed, in which case
1596 1 : // the SeekLT following this should not make any assumption about iterator
1597 1 : // position.
1598 1 : i.lastPositioningOp = unknownLastPositionOp
1599 1 : i.batchJustRefreshed = false
1600 1 : i.requiresReposition = false
1601 1 : i.err = nil // clear cached iteration error
1602 1 : i.stats.ReverseSeekCount[InterfaceCall]++
1603 1 : if upperBound := i.opts.GetUpperBound(); upperBound != nil && i.cmp(key, upperBound) > 0 {
1604 1 : key = upperBound
1605 1 : } else if lowerBound := i.opts.GetLowerBound(); lowerBound != nil && i.cmp(key, lowerBound) < 0 {
1606 1 : key = lowerBound
1607 1 : }
1608 1 : i.hasPrefix = false
1609 1 : seekInternalIter := true
1610 1 : // The following noop optimization only applies when i.batch == nil, since
1611 1 : // an iterator over a batch is iterating over mutable data, that may have
1612 1 : // changed since the last seek.
1613 1 : if lastPositioningOp == seekLTLastPositioningOp && i.batch == nil {
1614 1 : cmp := i.cmp(key, i.prefixOrFullSeekKey)
1615 1 : // If this seek is to the same or earlier key, and the iterator is
1616 1 : // already positioned there, this is a noop. This can be helpful for
1617 1 : // sparse key spaces that have many deleted keys, where one can avoid
1618 1 : // the overhead of iterating past them again and again.
1619 1 : if cmp <= 0 {
1620 1 : // NB: when pos != iterPosCurReversePaused, the invariant
1621 1 : // documented earlier implies that iterValidityState !=
1622 1 : // IterAtLimit.
1623 1 : if i.iterValidityState == IterExhausted ||
1624 1 : (i.iterValidityState == IterValid && i.cmp(i.key, key) < 0 &&
1625 1 : (limit == nil || i.cmp(limit, i.key) <= 0)) {
1626 1 : if !testingDisableSeekOpt(key, uintptr(unsafe.Pointer(i))) {
1627 1 : i.lastPositioningOp = seekLTLastPositioningOp
1628 1 : return i.iterValidityState
1629 1 : }
1630 : }
1631 1 : if i.pos == iterPosCurReversePaused && i.cmp(i.iterKV.K.UserKey, key) < 0 {
1632 1 : // Have some work to do, but don't need to seek, and we can
1633 1 : // start doing findPrevEntry from i.iterKey.
1634 1 : seekInternalIter = false
1635 1 : }
1636 : }
1637 : }
1638 1 : if seekInternalIter {
1639 1 : i.iterKV = i.iter.SeekLT(key, base.SeekLTFlagsNone)
1640 1 : i.stats.ReverseSeekCount[InternalIterCall]++
1641 1 : if err := i.iter.Error(); err != nil {
1642 1 : i.err = err
1643 1 : i.iterValidityState = IterExhausted
1644 1 : return i.iterValidityState
1645 1 : }
1646 : }
1647 1 : i.findPrevEntry(limit)
1648 1 : i.maybeSampleRead()
1649 1 : if i.Error() == nil && i.batch == nil {
1650 1 : // Prepare state for a future noop optimization.
1651 1 : i.prefixOrFullSeekKey = append(i.prefixOrFullSeekKey[:0], key...)
1652 1 : i.lastPositioningOp = seekLTLastPositioningOp
1653 1 : }
1654 1 : return i.iterValidityState
1655 : }
1656 :
1657 : // First moves the iterator the first key/value pair. Returns true if the
1658 : // iterator is pointing at a valid entry and false otherwise.
1659 1 : func (i *Iterator) First() bool {
1660 1 : if i.rangeKey != nil {
1661 1 : // NB: Check Valid() before clearing requiresReposition.
1662 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1663 1 : // If we have a range key but did not expose it at the previous iterator
1664 1 : // position (because the iterator was not at a valid position), updated
1665 1 : // must be true. This ensures that after an iterator op sequence like:
1666 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1667 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1668 1 : // - First(...) → (IterValid, RangeBounds() = [a,b))
1669 1 : // the iterator returns RangeKeyChanged()=true.
1670 1 : //
1671 1 : // The remainder of this function will only update i.rangeKey.updated if
1672 1 : // the iterator moves into a new range key, or out of the current range
1673 1 : // key.
1674 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1675 1 : }
1676 1 : i.err = nil // clear cached iteration error
1677 1 : i.hasPrefix = false
1678 1 : i.batchJustRefreshed = false
1679 1 : i.lastPositioningOp = unknownLastPositionOp
1680 1 : i.requiresReposition = false
1681 1 : i.stats.ForwardSeekCount[InterfaceCall]++
1682 1 :
1683 1 : i.err = i.iterFirstWithinBounds()
1684 1 : if i.err != nil {
1685 1 : i.iterValidityState = IterExhausted
1686 1 : return false
1687 1 : }
1688 1 : i.findNextEntry(nil)
1689 1 : i.maybeSampleRead()
1690 1 : return i.iterValidityState == IterValid
1691 : }
1692 :
1693 : // Last moves the iterator the last key/value pair. Returns true if the
1694 : // iterator is pointing at a valid entry and false otherwise.
1695 1 : func (i *Iterator) Last() bool {
1696 1 : if i.rangeKey != nil {
1697 1 : // NB: Check Valid() before clearing requiresReposition.
1698 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1699 1 : // If we have a range key but did not expose it at the previous iterator
1700 1 : // position (because the iterator was not at a valid position), updated
1701 1 : // must be true. This ensures that after an iterator op sequence like:
1702 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1703 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1704 1 : // - Last(...) → (IterValid, RangeBounds() = [a,b))
1705 1 : // the iterator returns RangeKeyChanged()=true.
1706 1 : //
1707 1 : // The remainder of this function will only update i.rangeKey.updated if
1708 1 : // the iterator moves into a new range key, or out of the current range
1709 1 : // key.
1710 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1711 1 : }
1712 1 : i.err = nil // clear cached iteration error
1713 1 : i.hasPrefix = false
1714 1 : i.batchJustRefreshed = false
1715 1 : i.lastPositioningOp = unknownLastPositionOp
1716 1 : i.requiresReposition = false
1717 1 : i.stats.ReverseSeekCount[InterfaceCall]++
1718 1 :
1719 1 : if i.err = i.iterLastWithinBounds(); i.err != nil {
1720 1 : i.iterValidityState = IterExhausted
1721 1 : return false
1722 1 : }
1723 1 : i.findPrevEntry(nil)
1724 1 : i.maybeSampleRead()
1725 1 : return i.iterValidityState == IterValid
1726 : }
1727 :
1728 : // Next moves the iterator to the next key/value pair. Returns true if the
1729 : // iterator is pointing at a valid entry and false otherwise.
1730 1 : func (i *Iterator) Next() bool {
1731 1 : return i.nextWithLimit(nil) == IterValid
1732 1 : }
1733 :
1734 : // NextWithLimit moves the iterator to the next key/value pair.
1735 : //
1736 : // If limit is provided, it serves as a best-effort exclusive limit. If the next
1737 : // key is greater than or equal to limit, the Iterator may pause and return
1738 : // IterAtLimit. Because limits are best-effort, NextWithLimit may return a key
1739 : // beyond limit.
1740 : //
1741 : // If the Iterator is configured to iterate over range keys, NextWithLimit
1742 : // guarantees it will surface any range keys with bounds overlapping the
1743 : // keyspace up to limit.
1744 1 : func (i *Iterator) NextWithLimit(limit []byte) IterValidityState {
1745 1 : return i.nextWithLimit(limit)
1746 1 : }
1747 :
1748 : // NextPrefix moves the iterator to the next key/value pair with a key
1749 : // containing a different prefix than the current key. Prefixes are determined
1750 : // by Comparer.Split. Exhausts the iterator if invoked while in prefix-iteration
1751 : // mode.
1752 : //
1753 : // It is not permitted to invoke NextPrefix while at a IterAtLimit position.
1754 : // When called in this condition, NextPrefix has non-deterministic behavior.
1755 : //
1756 : // It is not permitted to invoke NextPrefix when the Iterator has an
1757 : // upper-bound that is a versioned MVCC key (see the comment for
1758 : // Comparer.Split). It returns an error in this case.
1759 1 : func (i *Iterator) NextPrefix() bool {
1760 1 : if i.nextPrefixNotPermittedByUpperBound {
1761 1 : i.lastPositioningOp = unknownLastPositionOp
1762 1 : i.requiresReposition = false
1763 1 : i.err = errors.Errorf("NextPrefix not permitted with upper bound %s",
1764 1 : i.comparer.FormatKey(i.opts.UpperBound))
1765 1 : i.iterValidityState = IterExhausted
1766 1 : return false
1767 1 : }
1768 1 : if i.hasPrefix {
1769 1 : i.iterValidityState = IterExhausted
1770 1 : return false
1771 1 : }
1772 1 : if i.Error() != nil {
1773 1 : return false
1774 1 : }
1775 1 : return i.nextPrefix() == IterValid
1776 : }
1777 :
1778 1 : func (i *Iterator) nextPrefix() IterValidityState {
1779 1 : if i.rangeKey != nil {
1780 1 : // NB: Check Valid() before clearing requiresReposition.
1781 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1782 1 : // If we have a range key but did not expose it at the previous iterator
1783 1 : // position (because the iterator was not at a valid position), updated
1784 1 : // must be true. This ensures that after an iterator op sequence like:
1785 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1786 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1787 1 : // - NextWithLimit(...) → (IterValid, RangeBounds() = [a,b))
1788 1 : // the iterator returns RangeKeyChanged()=true.
1789 1 : //
1790 1 : // The remainder of this function will only update i.rangeKey.updated if
1791 1 : // the iterator moves into a new range key, or out of the current range
1792 1 : // key.
1793 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1794 1 : }
1795 :
1796 : // Although NextPrefix documents that behavior at IterAtLimit is undefined,
1797 : // this function handles these cases as a simple prefix-agnostic Next. This
1798 : // is done for deterministic behavior in the metamorphic tests.
1799 : //
1800 : // TODO(jackson): If the metamorphic test operation generator is adjusted to
1801 : // make generation of some operations conditional on the previous
1802 : // operations, then we can remove this behavior and explicitly error.
1803 :
1804 1 : i.lastPositioningOp = unknownLastPositionOp
1805 1 : i.requiresReposition = false
1806 1 : switch i.pos {
1807 1 : case iterPosCurForward:
1808 1 : // Positioned on the current key. Advance to the next prefix.
1809 1 : i.internalNextPrefix(i.comparer.Split(i.key))
1810 0 : case iterPosCurForwardPaused:
1811 : // Positioned at a limit. Implement as a prefix-agnostic Next. See TODO
1812 : // up above. The iterator is already positioned at the next key.
1813 1 : case iterPosCurReverse:
1814 1 : // Switching directions.
1815 1 : // Unless the iterator was exhausted, reverse iteration needs to
1816 1 : // position the iterator at iterPosPrev.
1817 1 : if i.iterKV != nil {
1818 0 : i.err = errors.New("switching from reverse to forward but iter is not at prev")
1819 0 : i.iterValidityState = IterExhausted
1820 0 : return i.iterValidityState
1821 0 : }
1822 : // The Iterator is exhausted and i.iter is positioned before the first
1823 : // key. Reposition to point to the first internal key.
1824 1 : if i.err = i.iterFirstWithinBounds(); i.err != nil {
1825 0 : i.iterValidityState = IterExhausted
1826 0 : return i.iterValidityState
1827 0 : }
1828 1 : case iterPosCurReversePaused:
1829 1 : // Positioned at a limit. Implement as a prefix-agnostic Next. See TODO
1830 1 : // up above.
1831 1 : //
1832 1 : // Switching directions; The iterator must not be exhausted since it
1833 1 : // paused.
1834 1 : if i.iterKV == nil {
1835 0 : i.err = errors.New("switching paused from reverse to forward but iter is exhausted")
1836 0 : i.iterValidityState = IterExhausted
1837 0 : return i.iterValidityState
1838 0 : }
1839 1 : i.nextUserKey()
1840 1 : case iterPosPrev:
1841 1 : // The underlying iterator is pointed to the previous key (this can
1842 1 : // only happen when switching iteration directions).
1843 1 : if i.iterKV == nil {
1844 1 : // We're positioned before the first key. Need to reposition to point to
1845 1 : // the first key.
1846 1 : i.err = i.iterFirstWithinBounds()
1847 1 : if i.iterKV == nil {
1848 0 : i.iterValidityState = IterExhausted
1849 0 : return i.iterValidityState
1850 0 : }
1851 1 : if invariants.Enabled && !i.equal(i.iterKV.K.UserKey, i.key) {
1852 0 : i.opts.getLogger().Fatalf("pebble: invariant violation: First internal iterator from iterPosPrev landed on %q, not %q",
1853 0 : i.iterKV.K.UserKey, i.key)
1854 0 : }
1855 1 : } else {
1856 1 : // Move the internal iterator back onto the user key stored in
1857 1 : // i.key. iterPosPrev guarantees that it's positioned at the last
1858 1 : // key with the user key less than i.key, so we're guaranteed to
1859 1 : // land on the correct key with a single Next.
1860 1 : i.iterKV = i.iter.Next()
1861 1 : if i.iterKV == nil {
1862 1 : // This should only be possible if i.iter.Next() encountered an
1863 1 : // error.
1864 1 : if i.iter.Error() == nil {
1865 0 : i.opts.getLogger().Fatalf("pebble: invariant violation: Nexting internal iterator from iterPosPrev found nothing")
1866 0 : }
1867 : // NB: Iterator.Error() will return i.iter.Error().
1868 1 : i.iterValidityState = IterExhausted
1869 1 : return i.iterValidityState
1870 : }
1871 1 : if invariants.Enabled && !i.equal(i.iterKV.K.UserKey, i.key) {
1872 0 : i.opts.getLogger().Fatalf("pebble: invariant violation: Nexting internal iterator from iterPosPrev landed on %q, not %q",
1873 0 : i.iterKV.K.UserKey, i.key)
1874 0 : }
1875 : }
1876 : // The internal iterator is now positioned at i.key. Advance to the next
1877 : // prefix.
1878 1 : i.internalNextPrefix(i.comparer.Split(i.key))
1879 1 : case iterPosNext:
1880 1 : // Already positioned on the next key. Only call nextPrefixKey if the
1881 1 : // next key shares the same prefix.
1882 1 : if i.iterKV != nil {
1883 1 : currKeyPrefixLen := i.comparer.Split(i.key)
1884 1 : if bytes.Equal(i.comparer.Split.Prefix(i.iterKV.K.UserKey), i.key[:currKeyPrefixLen]) {
1885 1 : i.internalNextPrefix(currKeyPrefixLen)
1886 1 : }
1887 : }
1888 : }
1889 :
1890 1 : i.stats.ForwardStepCount[InterfaceCall]++
1891 1 : i.findNextEntry(nil /* limit */)
1892 1 : i.maybeSampleRead()
1893 1 : return i.iterValidityState
1894 : }
1895 :
1896 1 : func (i *Iterator) internalNextPrefix(currKeyPrefixLen int) {
1897 1 : if i.iterKV == nil {
1898 1 : return
1899 1 : }
1900 : // The Next "fast-path" is not really a fast-path when there is more than
1901 : // one version. However, even with TableFormatPebblev3, there is a small
1902 : // slowdown (~10%) for one version if we remove it and only call NextPrefix.
1903 : // When there are two versions, only calling NextPrefix is ~30% faster.
1904 1 : i.stats.ForwardStepCount[InternalIterCall]++
1905 1 : if i.iterKV = i.iter.Next(); i.iterKV == nil {
1906 1 : return
1907 1 : }
1908 1 : if !bytes.Equal(i.comparer.Split.Prefix(i.iterKV.K.UserKey), i.key[:currKeyPrefixLen]) {
1909 1 : return
1910 1 : }
1911 1 : i.stats.ForwardStepCount[InternalIterCall]++
1912 1 : i.prefixOrFullSeekKey = i.comparer.ImmediateSuccessor(i.prefixOrFullSeekKey[:0], i.key[:currKeyPrefixLen])
1913 1 : if i.iterKV.K.IsExclusiveSentinel() {
1914 0 : panic(errors.AssertionFailedf("pebble: unexpected exclusive sentinel key: %q", i.iterKV.K))
1915 : }
1916 :
1917 1 : i.iterKV = i.iter.NextPrefix(i.prefixOrFullSeekKey)
1918 1 : if invariants.Enabled && i.iterKV != nil {
1919 1 : if p := i.comparer.Split.Prefix(i.iterKV.K.UserKey); i.cmp(p, i.prefixOrFullSeekKey) < 0 {
1920 0 : panic(errors.AssertionFailedf("pebble: iter.NextPrefix did not advance beyond the current prefix: now at %q; expected to be geq %q",
1921 0 : i.iterKV.K, i.prefixOrFullSeekKey))
1922 : }
1923 : }
1924 : }
1925 :
1926 1 : func (i *Iterator) nextWithLimit(limit []byte) IterValidityState {
1927 1 : i.stats.ForwardStepCount[InterfaceCall]++
1928 1 : if i.hasPrefix {
1929 1 : if limit != nil {
1930 1 : i.err = errors.New("cannot use limit with prefix iteration")
1931 1 : i.iterValidityState = IterExhausted
1932 1 : return i.iterValidityState
1933 1 : } else if i.iterValidityState == IterExhausted {
1934 1 : // No-op, already exhausted. We avoid executing the Next because it
1935 1 : // can break invariants: Specifically, a file that fails the bloom
1936 1 : // filter test may result in its level being removed from the
1937 1 : // merging iterator. The level's removal can cause a lazy combined
1938 1 : // iterator to miss range keys and trigger a switch to combined
1939 1 : // iteration at a larger key, breaking keyspan invariants.
1940 1 : return i.iterValidityState
1941 1 : }
1942 : }
1943 1 : if i.err != nil {
1944 1 : return i.iterValidityState
1945 1 : }
1946 1 : if i.rangeKey != nil {
1947 1 : // NB: Check Valid() before clearing requiresReposition.
1948 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
1949 1 : // If we have a range key but did not expose it at the previous iterator
1950 1 : // position (because the iterator was not at a valid position), updated
1951 1 : // must be true. This ensures that after an iterator op sequence like:
1952 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
1953 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
1954 1 : // - NextWithLimit(...) → (IterValid, RangeBounds() = [a,b))
1955 1 : // the iterator returns RangeKeyChanged()=true.
1956 1 : //
1957 1 : // The remainder of this function will only update i.rangeKey.updated if
1958 1 : // the iterator moves into a new range key, or out of the current range
1959 1 : // key.
1960 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
1961 1 : }
1962 1 : i.lastPositioningOp = unknownLastPositionOp
1963 1 : i.requiresReposition = false
1964 1 : switch i.pos {
1965 1 : case iterPosCurForward:
1966 1 : i.nextUserKey()
1967 1 : case iterPosCurForwardPaused:
1968 : // Already at the right place.
1969 1 : case iterPosCurReverse:
1970 1 : // Switching directions.
1971 1 : // Unless the iterator was exhausted, reverse iteration needs to
1972 1 : // position the iterator at iterPosPrev.
1973 1 : if i.iterKV != nil {
1974 0 : i.err = errors.New("switching from reverse to forward but iter is not at prev")
1975 0 : i.iterValidityState = IterExhausted
1976 0 : return i.iterValidityState
1977 0 : }
1978 : // We're positioned before the first key. Need to reposition to point to
1979 : // the first key.
1980 1 : if i.err = i.iterFirstWithinBounds(); i.err != nil {
1981 1 : i.iterValidityState = IterExhausted
1982 1 : return i.iterValidityState
1983 1 : }
1984 1 : case iterPosCurReversePaused:
1985 1 : // Switching directions.
1986 1 : // The iterator must not be exhausted since it paused.
1987 1 : if i.iterKV == nil {
1988 0 : i.err = errors.New("switching paused from reverse to forward but iter is exhausted")
1989 0 : i.iterValidityState = IterExhausted
1990 0 : return i.iterValidityState
1991 0 : }
1992 1 : i.nextUserKey()
1993 1 : case iterPosPrev:
1994 1 : // The underlying iterator is pointed to the previous key (this can
1995 1 : // only happen when switching iteration directions). We set
1996 1 : // i.iterValidityState to IterExhausted here to force the calls to
1997 1 : // nextUserKey to save the current key i.iter is pointing at in order
1998 1 : // to determine when the next user-key is reached.
1999 1 : i.iterValidityState = IterExhausted
2000 1 : if i.iterKV == nil {
2001 1 : // We're positioned before the first key. Need to reposition to point to
2002 1 : // the first key.
2003 1 : i.err = i.iterFirstWithinBounds()
2004 1 : } else {
2005 1 : i.nextUserKey()
2006 1 : }
2007 1 : if i.err != nil {
2008 1 : i.iterValidityState = IterExhausted
2009 1 : return i.iterValidityState
2010 1 : }
2011 1 : i.nextUserKey()
2012 1 : case iterPosNext:
2013 : // Already at the right place.
2014 : }
2015 1 : i.findNextEntry(limit)
2016 1 : i.maybeSampleRead()
2017 1 : return i.iterValidityState
2018 : }
2019 :
2020 : // Prev moves the iterator to the previous key/value pair. Returns true if the
2021 : // iterator is pointing at a valid entry and false otherwise.
2022 1 : func (i *Iterator) Prev() bool {
2023 1 : return i.PrevWithLimit(nil) == IterValid
2024 1 : }
2025 :
2026 : // PrevWithLimit moves the iterator to the previous key/value pair.
2027 : //
2028 : // If limit is provided, it serves as a best-effort inclusive limit. If the
2029 : // previous key is less than limit, the Iterator may pause and return
2030 : // IterAtLimit. Because limits are best-effort, PrevWithLimit may return a key
2031 : // beyond limit.
2032 : //
2033 : // If the Iterator is configured to iterate over range keys, PrevWithLimit
2034 : // guarantees it will surface any range keys with bounds overlapping the
2035 : // keyspace up to limit.
2036 1 : func (i *Iterator) PrevWithLimit(limit []byte) IterValidityState {
2037 1 : i.stats.ReverseStepCount[InterfaceCall]++
2038 1 : if i.err != nil {
2039 1 : return i.iterValidityState
2040 1 : }
2041 1 : if i.rangeKey != nil {
2042 1 : // NB: Check Valid() before clearing requiresReposition.
2043 1 : i.rangeKey.prevPosHadRangeKey = i.rangeKey.hasRangeKey && i.Valid()
2044 1 : // If we have a range key but did not expose it at the previous iterator
2045 1 : // position (because the iterator was not at a valid position), updated
2046 1 : // must be true. This ensures that after an iterator op sequence like:
2047 1 : // - Next() → (IterValid, RangeBounds() = [a,b))
2048 1 : // - NextWithLimit(...) → (IterAtLimit, RangeBounds() = -)
2049 1 : // - PrevWithLimit(...) → (IterValid, RangeBounds() = [a,b))
2050 1 : // the iterator returns RangeKeyChanged()=true.
2051 1 : //
2052 1 : // The remainder of this function will only update i.rangeKey.updated if
2053 1 : // the iterator moves into a new range key, or out of the current range
2054 1 : // key.
2055 1 : i.rangeKey.updated = i.rangeKey.hasRangeKey && !i.Valid() && i.opts.rangeKeys()
2056 1 : }
2057 1 : i.lastPositioningOp = unknownLastPositionOp
2058 1 : i.requiresReposition = false
2059 1 : if i.hasPrefix {
2060 1 : i.err = errReversePrefixIteration
2061 1 : i.iterValidityState = IterExhausted
2062 1 : return i.iterValidityState
2063 1 : }
2064 1 : switch i.pos {
2065 1 : case iterPosCurForward:
2066 : // Switching directions, and will handle this below.
2067 1 : case iterPosCurForwardPaused:
2068 : // Switching directions, and will handle this below.
2069 1 : case iterPosCurReverse:
2070 1 : i.prevUserKey()
2071 1 : case iterPosCurReversePaused:
2072 : // Already at the right place.
2073 1 : case iterPosNext:
2074 : // The underlying iterator is pointed to the next key (this can only happen
2075 : // when switching iteration directions). We will handle this below.
2076 1 : case iterPosPrev:
2077 : // Already at the right place.
2078 : }
2079 1 : if i.pos == iterPosCurForward || i.pos == iterPosNext || i.pos == iterPosCurForwardPaused {
2080 1 : // Switching direction.
2081 1 : stepAgain := i.pos == iterPosNext
2082 1 :
2083 1 : // Synthetic range key markers are a special case. Consider SeekGE(b)
2084 1 : // which finds a range key [a, c). To ensure the user observes the range
2085 1 : // key, the Iterator pauses at Key() = b. The iterator must advance the
2086 1 : // internal iterator to see if there's also a coincident point key at
2087 1 : // 'b', leaving the iterator at iterPosNext if there's not.
2088 1 : //
2089 1 : // This is a problem: Synthetic range key markers are only interleaved
2090 1 : // during the original seek. A subsequent Prev() of i.iter will not move
2091 1 : // back onto the synthetic range key marker. In this case where the
2092 1 : // previous iterator position was a synthetic range key start boundary,
2093 1 : // we must not step a second time.
2094 1 : if i.isEphemeralPosition() {
2095 1 : stepAgain = false
2096 1 : }
2097 :
2098 : // We set i.iterValidityState to IterExhausted here to force the calls
2099 : // to prevUserKey to save the current key i.iter is pointing at in
2100 : // order to determine when the prev user-key is reached.
2101 1 : i.iterValidityState = IterExhausted
2102 1 : if i.iterKV == nil {
2103 1 : // We're positioned after the last key. Need to reposition to point to
2104 1 : // the last key.
2105 1 : i.err = i.iterLastWithinBounds()
2106 1 : } else {
2107 1 : i.prevUserKey()
2108 1 : }
2109 1 : if i.err != nil {
2110 1 : return i.iterValidityState
2111 1 : }
2112 1 : if stepAgain {
2113 1 : i.prevUserKey()
2114 1 : if i.err != nil {
2115 0 : return i.iterValidityState
2116 0 : }
2117 : }
2118 : }
2119 1 : i.findPrevEntry(limit)
2120 1 : i.maybeSampleRead()
2121 1 : return i.iterValidityState
2122 : }
2123 :
2124 : // iterFirstWithinBounds moves the internal iterator to the first key,
2125 : // respecting bounds.
2126 1 : func (i *Iterator) iterFirstWithinBounds() error {
2127 1 : i.stats.ForwardSeekCount[InternalIterCall]++
2128 1 : if lowerBound := i.opts.GetLowerBound(); lowerBound != nil {
2129 1 : i.iterKV = i.iter.SeekGE(lowerBound, base.SeekGEFlagsNone)
2130 1 : } else {
2131 1 : i.iterKV = i.iter.First()
2132 1 : }
2133 1 : if i.iterKV == nil {
2134 1 : return i.iter.Error()
2135 1 : }
2136 1 : return nil
2137 : }
2138 :
2139 : // iterLastWithinBounds moves the internal iterator to the last key, respecting
2140 : // bounds.
2141 1 : func (i *Iterator) iterLastWithinBounds() error {
2142 1 : i.stats.ReverseSeekCount[InternalIterCall]++
2143 1 : if upperBound := i.opts.GetUpperBound(); upperBound != nil {
2144 1 : i.iterKV = i.iter.SeekLT(upperBound, base.SeekLTFlagsNone)
2145 1 : } else {
2146 1 : i.iterKV = i.iter.Last()
2147 1 : }
2148 1 : if i.iterKV == nil {
2149 1 : return i.iter.Error()
2150 1 : }
2151 1 : return nil
2152 : }
2153 :
2154 : // RangeKeyData describes a range key's data, set through RangeKeySet. The key
2155 : // boundaries of the range key is provided by Iterator.RangeBounds.
2156 : type RangeKeyData struct {
2157 : Suffix []byte
2158 : Value []byte
2159 : }
2160 :
2161 : // rangeKeyWithinLimit is called during limited reverse iteration when
2162 : // positioned over a key beyond the limit. If there exists a range key that lies
2163 : // within the limit, the iterator must not pause in order to ensure the user has
2164 : // an opportunity to observe the range key within limit.
2165 : //
2166 : // It would be valid to ignore the limit whenever there's a range key covering
2167 : // the key, but that would introduce nondeterminism. To preserve determinism for
2168 : // testing, the iterator ignores the limit only if the covering range key does
2169 : // cover the keyspace within the limit.
2170 : //
2171 : // This awkwardness exists because range keys are interleaved at their inclusive
2172 : // start positions. Note that limit is inclusive.
2173 1 : func (i *Iterator) rangeKeyWithinLimit(limit []byte) bool {
2174 1 : if i.rangeKey == nil || !i.opts.rangeKeys() {
2175 1 : return false
2176 1 : }
2177 1 : s := i.rangeKey.iiter.Span()
2178 1 : // If the range key ends beyond the limit, then the range key does not cover
2179 1 : // any portion of the keyspace within the limit and it is safe to pause.
2180 1 : return s != nil && i.cmp(s.End, limit) > 0
2181 : }
2182 :
2183 : // saveRangeKey saves the current range key to the underlying iterator's current
2184 : // range key state. If the range key has not changed, saveRangeKey is a no-op.
2185 : // If there is a new range key, saveRangeKey copies all of the key, value and
2186 : // suffixes into Iterator-managed buffers.
2187 1 : func (i *Iterator) saveRangeKey() {
2188 1 : if i.rangeKey == nil || i.opts.KeyTypes == IterKeyTypePointsOnly {
2189 1 : return
2190 1 : }
2191 :
2192 1 : s := i.rangeKey.iiter.Span()
2193 1 : if s == nil {
2194 1 : i.rangeKey.hasRangeKey = false
2195 1 : i.rangeKey.updated = i.rangeKey.prevPosHadRangeKey
2196 1 : return
2197 1 : } else if !i.rangeKey.stale {
2198 1 : // The range key `s` is identical to the one currently saved. No-op.
2199 1 : return
2200 1 : }
2201 :
2202 1 : if s.KeysOrder != keyspan.BySuffixAsc {
2203 0 : panic("pebble: range key span's keys unexpectedly not in ascending suffix order")
2204 : }
2205 :
2206 : // Although `i.rangeKey.stale` is true, the span s may still be identical
2207 : // to the currently saved span. This is possible when seeking the iterator,
2208 : // which may land back on the same range key. If we previously had a range
2209 : // key and the new one has an identical start key, then it must be the same
2210 : // range key and we can avoid copying and keep `i.rangeKey.updated=false`.
2211 : //
2212 : // TODO(jackson): These key comparisons could be avoidable during relative
2213 : // positioning operations continuing in the same direction, because these
2214 : // ops will never encounter the previous position's range key while
2215 : // stale=true. However, threading whether the current op is a seek or step
2216 : // maybe isn't worth it. This key comparison is only necessary once when we
2217 : // step onto a new range key, which should be relatively rare.
2218 1 : if i.rangeKey.prevPosHadRangeKey && i.equal(i.rangeKey.start, s.Start) &&
2219 1 : i.equal(i.rangeKey.end, s.End) {
2220 1 : i.rangeKey.updated = false
2221 1 : i.rangeKey.stale = false
2222 1 : i.rangeKey.hasRangeKey = true
2223 1 : return
2224 1 : }
2225 1 : i.stats.RangeKeyStats.Count += len(s.Keys)
2226 1 : i.rangeKey.buf.Reset()
2227 1 : i.rangeKey.hasRangeKey = true
2228 1 : i.rangeKey.updated = true
2229 1 : i.rangeKey.stale = false
2230 1 : i.rangeKey.buf, i.rangeKey.start = i.rangeKey.buf.Copy(s.Start)
2231 1 : i.rangeKey.buf, i.rangeKey.end = i.rangeKey.buf.Copy(s.End)
2232 1 : i.rangeKey.keys = i.rangeKey.keys[:0]
2233 1 : for j := 0; j < len(s.Keys); j++ {
2234 1 : if invariants.Enabled {
2235 1 : if s.Keys[j].Kind() != base.InternalKeyKindRangeKeySet {
2236 0 : panic("pebble: user iteration encountered non-RangeKeySet key kind")
2237 1 : } else if j > 0 && i.comparer.CompareRangeSuffixes(s.Keys[j].Suffix, s.Keys[j-1].Suffix) < 0 {
2238 0 : panic("pebble: user iteration encountered range keys not in suffix order")
2239 : }
2240 : }
2241 1 : var rkd RangeKeyData
2242 1 : i.rangeKey.buf, rkd.Suffix = i.rangeKey.buf.Copy(s.Keys[j].Suffix)
2243 1 : i.rangeKey.buf, rkd.Value = i.rangeKey.buf.Copy(s.Keys[j].Value)
2244 1 : i.rangeKey.keys = append(i.rangeKey.keys, rkd)
2245 : }
2246 : }
2247 :
2248 : // RangeKeyChanged indicates whether the most recent iterator positioning
2249 : // operation resulted in the iterator stepping into or out of a new range key.
2250 : // If true, previously returned range key bounds and data has been invalidated.
2251 : // If false, previously obtained range key bounds, suffix and value slices are
2252 : // still valid and may continue to be read.
2253 : //
2254 : // Invalid iterator positions are considered to not hold range keys, meaning
2255 : // that if an iterator steps from an IterExhausted or IterAtLimit position onto
2256 : // a position with a range key, RangeKeyChanged will yield true.
2257 1 : func (i *Iterator) RangeKeyChanged() bool {
2258 1 : return i.iterValidityState == IterValid && i.rangeKey != nil && i.rangeKey.updated
2259 1 : }
2260 :
2261 : // HasPointAndRange indicates whether there exists a point key, a range key or
2262 : // both at the current iterator position.
2263 1 : func (i *Iterator) HasPointAndRange() (hasPoint, hasRange bool) {
2264 1 : if i.iterValidityState != IterValid || i.requiresReposition {
2265 1 : return false, false
2266 1 : }
2267 1 : if i.opts.KeyTypes == IterKeyTypePointsOnly {
2268 1 : return true, false
2269 1 : }
2270 1 : return i.rangeKey == nil || !i.rangeKey.rangeKeyOnly, i.rangeKey != nil && i.rangeKey.hasRangeKey
2271 : }
2272 :
2273 : // RangeBounds returns the start (inclusive) and end (exclusive) bounds of the
2274 : // range key covering the current iterator position. RangeBounds returns nil
2275 : // bounds if there is no range key covering the current iterator position, or
2276 : // the iterator is not configured to surface range keys.
2277 : //
2278 : // If valid, the returned start bound is less than or equal to Key() and the
2279 : // returned end bound is greater than Key().
2280 1 : func (i *Iterator) RangeBounds() (start, end []byte) {
2281 1 : if i.rangeKey == nil || !i.opts.rangeKeys() || !i.rangeKey.hasRangeKey {
2282 0 : return nil, nil
2283 0 : }
2284 1 : return i.rangeKey.start, i.rangeKey.end
2285 : }
2286 :
2287 : // Key returns the key of the current key/value pair, or nil if done. The
2288 : // caller should not modify the contents of the returned slice, and its
2289 : // contents may change on the next call to Next.
2290 : //
2291 : // If positioned at an iterator position that only holds a range key, Key()
2292 : // always returns the start bound of the range key. Otherwise, it returns the
2293 : // point key's key.
2294 1 : func (i *Iterator) Key() []byte {
2295 1 : return i.key
2296 1 : }
2297 :
2298 : // Value returns the value of the current key/value pair, or nil if done. The
2299 : // caller should not modify the contents of the returned slice, and its
2300 : // contents may change on the next call to Next.
2301 : //
2302 : // Only valid if HasPointAndRange() returns true for hasPoint.
2303 : // Deprecated: use ValueAndErr instead.
2304 1 : func (i *Iterator) Value() []byte {
2305 1 : val, _ := i.ValueAndErr()
2306 1 : return val
2307 1 : }
2308 :
2309 : // ValueAndErr returns the value, and any error encountered in extracting the value.
2310 : // REQUIRES: i.Error()==nil and HasPointAndRange() returns true for hasPoint.
2311 : //
2312 : // The caller should not modify the contents of the returned slice, and its
2313 : // contents may change on the next call to Next.
2314 1 : func (i *Iterator) ValueAndErr() ([]byte, error) {
2315 1 : val, callerOwned, err := i.value.Value(i.lazyValueBuf)
2316 1 : if err != nil {
2317 1 : i.err = err
2318 1 : i.iterValidityState = IterExhausted
2319 1 : }
2320 1 : if callerOwned {
2321 1 : i.lazyValueBuf = val[:0]
2322 1 : }
2323 1 : return val, err
2324 : }
2325 :
2326 : // LazyValue returns the LazyValue. Only for advanced use cases.
2327 : // REQUIRES: i.Error()==nil and HasPointAndRange() returns true for hasPoint.
2328 0 : func (i *Iterator) LazyValue() LazyValue {
2329 0 : return i.value.LazyValue()
2330 0 : }
2331 :
2332 : // RangeKeys returns the range key values and their suffixes covering the
2333 : // current iterator position. The range bounds may be retrieved separately
2334 : // through Iterator.RangeBounds().
2335 1 : func (i *Iterator) RangeKeys() []RangeKeyData {
2336 1 : if i.rangeKey == nil || !i.opts.rangeKeys() || !i.rangeKey.hasRangeKey {
2337 0 : return nil
2338 0 : }
2339 1 : return i.rangeKey.keys
2340 : }
2341 :
2342 : // Valid returns true if the iterator is positioned at a valid key/value pair
2343 : // and false otherwise.
2344 1 : func (i *Iterator) Valid() bool {
2345 1 : valid := i.iterValidityState == IterValid && !i.requiresReposition
2346 1 : if invariants.Enabled {
2347 1 : if err := i.Error(); valid && err != nil {
2348 0 : panic(errors.AssertionFailedf("pebble: iterator is valid with non-nil Error: %+v", err))
2349 : }
2350 : }
2351 1 : return valid
2352 : }
2353 :
2354 : // Error returns any accumulated error.
2355 1 : func (i *Iterator) Error() error {
2356 1 : if i.err != nil {
2357 1 : return i.err
2358 1 : }
2359 1 : if i.iter != nil {
2360 1 : return i.iter.Error()
2361 1 : }
2362 0 : return nil
2363 : }
2364 :
2365 : const maxKeyBufCacheSize = 4 << 10 // 4 KB
2366 :
2367 : // Close closes the iterator and returns any accumulated error. Exhausting
2368 : // all the key/value pairs in a table is not considered to be an error.
2369 : // It is not valid to call any method, including Close, after the iterator
2370 : // has been closed.
2371 1 : func (i *Iterator) Close() error {
2372 1 : if i.trackerHandle.IsValid() {
2373 1 : i.tracker.Stop(i.trackerHandle)
2374 1 : i.trackerHandle = 0
2375 1 : }
2376 : // Close the child iterator before releasing the readState because when the
2377 : // readState is released sstables referenced by the readState may be deleted
2378 : // which will fail on Windows if the sstables are still open by the child
2379 : // iterator.
2380 1 : if i.iter != nil {
2381 1 : i.err = firstError(i.err, i.iter.Close())
2382 1 :
2383 1 : // Closing i.iter did not necessarily close the point and range key
2384 1 : // iterators. Calls to SetOptions may have 'disconnected' either one
2385 1 : // from i.iter if iteration key types were changed. Both point and range
2386 1 : // key iterators are preserved in case the iterator needs to switch key
2387 1 : // types again. We explicitly close both of these iterators here.
2388 1 : //
2389 1 : // NB: If the iterators were still connected to i.iter, they may be
2390 1 : // closed, but calling Close on a closed internal iterator or fragment
2391 1 : // iterator is allowed.
2392 1 : if i.pointIter != nil {
2393 1 : i.err = firstError(i.err, i.pointIter.Close())
2394 1 : }
2395 1 : if i.rangeKey != nil && i.rangeKey.rangeKeyIter != nil {
2396 1 : i.rangeKey.rangeKeyIter.Close()
2397 1 : }
2398 1 : i.err = firstError(i.err, i.blobValueFetcher.Close())
2399 : }
2400 1 : err := i.err
2401 1 :
2402 1 : if i.readState != nil {
2403 1 : if i.readSampling.pendingCompactions.size > 0 {
2404 1 : // Copy pending read compactions using db.mu.Lock()
2405 1 : i.readState.db.mu.Lock()
2406 1 : i.readState.db.mu.compact.readCompactions.combine(&i.readSampling.pendingCompactions, i.cmp)
2407 1 : reschedule := i.readState.db.mu.compact.rescheduleReadCompaction
2408 1 : i.readState.db.mu.compact.rescheduleReadCompaction = false
2409 1 : concurrentCompactions := i.readState.db.mu.compact.compactingCount
2410 1 : i.readState.db.mu.Unlock()
2411 1 :
2412 1 : if reschedule && concurrentCompactions == 0 {
2413 0 : // In a read heavy workload, flushes may not happen frequently enough to
2414 0 : // schedule compactions.
2415 0 : i.readState.db.compactionSchedulers.Add(1)
2416 0 : go i.readState.db.maybeScheduleCompactionAsync()
2417 0 : }
2418 : }
2419 :
2420 1 : i.readState.unref()
2421 1 : i.readState = nil
2422 : }
2423 :
2424 1 : if i.version != nil {
2425 1 : i.version.Unref()
2426 1 : }
2427 1 : if i.externalIter != nil {
2428 1 : err = firstError(err, i.externalIter.Close())
2429 1 : }
2430 :
2431 : // Close the closer for the current value if one was open.
2432 1 : if i.valueCloser != nil {
2433 1 : err = firstError(err, i.valueCloser.Close())
2434 1 : i.valueCloser = nil
2435 1 : }
2436 :
2437 1 : if i.rangeKey != nil {
2438 1 : i.rangeKey.rangeKeyBuffers.PrepareForReuse()
2439 1 : *i.rangeKey = iteratorRangeKeyState{
2440 1 : rangeKeyBuffers: i.rangeKey.rangeKeyBuffers,
2441 1 : }
2442 1 : iterRangeKeyStateAllocPool.Put(i.rangeKey)
2443 1 : i.rangeKey = nil
2444 1 : }
2445 :
2446 1 : alloc := i.alloc
2447 1 : if alloc == nil {
2448 1 : // NB: When the Iterator is used as a part of a Get(), Close() is called by
2449 1 : // getIterAlloc.Close which handles recycling the appropriate structure and
2450 1 : // fields.
2451 1 : return err
2452 1 : }
2453 :
2454 : // Reset the alloc struct, retaining any buffers within their maximum
2455 : // bounds, and return the iterAlloc struct to the pool.
2456 :
2457 : // Avoid caching the key buf if it is overly large. The constant is fairly
2458 : // arbitrary.
2459 1 : if cap(i.keyBuf) < maxKeyBufCacheSize {
2460 1 : alloc.keyBuf = i.keyBuf
2461 1 : }
2462 1 : if cap(i.prefixOrFullSeekKey) < maxKeyBufCacheSize {
2463 1 : alloc.prefixOrFullSeekKey = i.prefixOrFullSeekKey
2464 1 : }
2465 1 : for j := range i.boundsBuf {
2466 1 : if cap(i.boundsBuf[j]) < maxKeyBufCacheSize {
2467 1 : alloc.boundsBuf[j] = i.boundsBuf[j]
2468 1 : }
2469 : }
2470 1 : mergingIterHeapItems := alloc.merging.heap.items
2471 1 :
2472 1 : // We zero each field piecemeal in part to avoid the use of a stack
2473 1 : // allocated autotmp iterAlloc variable, and in part to make it easier
2474 1 : // to make zeroing conditional on whether the field was actually used.
2475 1 : //
2476 1 : // TODO(jackson,nvanbenschoten): Iterator.Close continues to appear in
2477 1 : // profiles in part because of this zeroing.
2478 1 : if i.batch != nil {
2479 1 : alloc.batchState = iteratorBatchState{}
2480 1 : }
2481 1 : alloc.dbi.clearForReuse()
2482 1 : alloc.merging = mergingIter{}
2483 1 : clear(mergingIterHeapItems[:cap(mergingIterHeapItems)])
2484 1 : alloc.merging.heap.items = mergingIterHeapItems
2485 1 : clear(alloc.mlevels[:])
2486 1 : clear(alloc.levels[:])
2487 1 : clear(alloc.levelsPositioned[:])
2488 1 :
2489 1 : iterAllocPool.Put(alloc)
2490 1 : return err
2491 : }
2492 :
2493 : // SetBounds sets the lower and upper bounds for the iterator. Once SetBounds
2494 : // returns, the caller is free to mutate the provided slices.
2495 : //
2496 : // The iterator will always be invalidated and must be repositioned with a call
2497 : // to SeekGE, SeekPrefixGE, SeekLT, First, or Last.
2498 1 : func (i *Iterator) SetBounds(lower, upper []byte) {
2499 1 : // Ensure that the Iterator appears exhausted, regardless of whether we
2500 1 : // actually have to invalidate the internal iterator. Optimizations that
2501 1 : // avoid exhaustion are an internal implementation detail that shouldn't
2502 1 : // leak through the interface. The caller should still call an absolute
2503 1 : // positioning method to reposition the iterator.
2504 1 : i.requiresReposition = true
2505 1 :
2506 1 : if ((i.opts.LowerBound == nil) == (lower == nil)) &&
2507 1 : ((i.opts.UpperBound == nil) == (upper == nil)) &&
2508 1 : i.equal(i.opts.LowerBound, lower) &&
2509 1 : i.equal(i.opts.UpperBound, upper) {
2510 1 : // Unchanged, noop.
2511 1 : return
2512 1 : }
2513 :
2514 : // Copy the user-provided bounds into an Iterator-owned buffer, and set them
2515 : // on i.opts.{Lower,Upper}Bound.
2516 1 : i.processBounds(lower, upper)
2517 1 :
2518 1 : i.iter.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2519 1 : // If the iterator has an open point iterator that's not currently being
2520 1 : // used, propagate the new bounds to it.
2521 1 : if i.pointIter != nil && !i.opts.pointKeys() {
2522 1 : i.pointIter.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2523 1 : }
2524 : // If the iterator has a range key iterator, propagate bounds to it. The
2525 : // top-level SetBounds on the interleaving iterator (i.iter) won't propagate
2526 : // bounds to the range key iterator stack, because the FragmentIterator
2527 : // interface doesn't define a SetBounds method. We need to directly inform
2528 : // the iterConfig stack.
2529 1 : if i.rangeKey != nil {
2530 1 : i.rangeKey.iterConfig.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2531 1 : }
2532 :
2533 : // Even though this is not a positioning operation, the alteration of the
2534 : // bounds means we cannot optimize Seeks by using Next.
2535 1 : i.invalidate()
2536 : }
2537 :
2538 : // SetContext replaces the context provided at iterator creation, or the last
2539 : // one provided by SetContext. Even though iterators are expected to be
2540 : // short-lived, there are some cases where either (a) iterators are used far
2541 : // from the code that created them, (b) iterators are reused (while being
2542 : // short-lived) for processing different requests. For such scenarios, we
2543 : // allow the caller to replace the context.
2544 0 : func (i *Iterator) SetContext(ctx context.Context) {
2545 0 : i.ctx = ctx
2546 0 : i.iter.SetContext(ctx)
2547 0 : // If the iterator has an open point iterator that's not currently being
2548 0 : // used, propagate the new context to it.
2549 0 : if i.pointIter != nil && !i.opts.pointKeys() {
2550 0 : i.pointIter.SetContext(i.ctx)
2551 0 : }
2552 : }
2553 :
2554 : // Initialization and changing of the bounds must call processBounds.
2555 : // processBounds saves the bounds and computes derived state from those
2556 : // bounds.
2557 1 : func (i *Iterator) processBounds(lower, upper []byte) {
2558 1 : // Copy the user-provided bounds into an Iterator-owned buffer. We can't
2559 1 : // overwrite the current bounds, because some internal iterators compare old
2560 1 : // and new bounds for optimizations.
2561 1 :
2562 1 : buf := i.boundsBuf[i.boundsBufIdx][:0]
2563 1 : if lower != nil {
2564 1 : buf = append(buf, lower...)
2565 1 : i.opts.LowerBound = buf
2566 1 : } else {
2567 1 : i.opts.LowerBound = nil
2568 1 : }
2569 1 : i.nextPrefixNotPermittedByUpperBound = false
2570 1 : if upper != nil {
2571 1 : buf = append(buf, upper...)
2572 1 : i.opts.UpperBound = buf[len(buf)-len(upper):]
2573 1 : if i.comparer.Split(i.opts.UpperBound) != len(i.opts.UpperBound) {
2574 1 : // Setting an upper bound that is a versioned MVCC key. This means
2575 1 : // that a key can have some MVCC versions before the upper bound and
2576 1 : // some after. This causes significant complications for NextPrefix,
2577 1 : // so we bar the user of NextPrefix.
2578 1 : i.nextPrefixNotPermittedByUpperBound = true
2579 1 : }
2580 1 : } else {
2581 1 : i.opts.UpperBound = nil
2582 1 : }
2583 1 : i.boundsBuf[i.boundsBufIdx] = buf
2584 1 : i.boundsBufIdx = 1 - i.boundsBufIdx
2585 : }
2586 :
2587 : // SetOptions sets new iterator options for the iterator. Note that the lower
2588 : // and upper bounds applied here will supersede any bounds set by previous calls
2589 : // to SetBounds.
2590 : //
2591 : // Note that the slices provided in this SetOptions must not be changed by the
2592 : // caller until the iterator is closed, or a subsequent SetBounds or SetOptions
2593 : // has returned. This is because comparisons between the existing and new bounds
2594 : // are sometimes used to optimize seeking. See the extended commentary on
2595 : // SetBounds.
2596 : //
2597 : // If the iterator was created over an indexed mutable batch, the iterator's
2598 : // view of the mutable batch is refreshed.
2599 : //
2600 : // The iterator will always be invalidated and must be repositioned with a call
2601 : // to SeekGE, SeekPrefixGE, SeekLT, First, or Last.
2602 : //
2603 : // If only lower and upper bounds need to be modified, prefer SetBounds.
2604 1 : func (i *Iterator) SetOptions(o *IterOptions) {
2605 1 : if i.externalIter != nil {
2606 1 : if err := validateExternalIterOpts(o); err != nil {
2607 0 : panic(err)
2608 : }
2609 : }
2610 :
2611 : // Ensure that the Iterator appears exhausted, regardless of whether we
2612 : // actually have to invalidate the internal iterator. Optimizations that
2613 : // avoid exhaustion are an internal implementation detail that shouldn't
2614 : // leak through the interface. The caller should still call an absolute
2615 : // positioning method to reposition the iterator.
2616 1 : i.requiresReposition = true
2617 1 :
2618 1 : // Check if global state requires we close all internal iterators.
2619 1 : //
2620 1 : // If the Iterator is in an error state, invalidate the existing iterators
2621 1 : // so that we reconstruct an iterator state from scratch.
2622 1 : //
2623 1 : // If OnlyReadGuaranteedDurable changed, the iterator stacks are incorrect,
2624 1 : // improperly including or excluding memtables. Invalidate them so that
2625 1 : // finishInitializingIter will reconstruct them.
2626 1 : closeBoth := i.err != nil ||
2627 1 : o.OnlyReadGuaranteedDurable != i.opts.OnlyReadGuaranteedDurable
2628 1 :
2629 1 : // If either options specify block property filters for an iterator stack,
2630 1 : // reconstruct it.
2631 1 : if i.pointIter != nil && (closeBoth || len(o.PointKeyFilters) > 0 || len(i.opts.PointKeyFilters) > 0 ||
2632 1 : o.RangeKeyMasking.Filter != nil || i.opts.RangeKeyMasking.Filter != nil || o.SkipPoint != nil ||
2633 1 : i.opts.SkipPoint != nil) {
2634 1 : i.err = firstError(i.err, i.pointIter.Close())
2635 1 : i.pointIter = nil
2636 1 : }
2637 1 : if i.rangeKey != nil {
2638 1 : if closeBoth || len(o.RangeKeyFilters) > 0 || len(i.opts.RangeKeyFilters) > 0 {
2639 1 : i.rangeKey.rangeKeyIter.Close()
2640 1 : i.rangeKey = nil
2641 1 : } else {
2642 1 : // If there's still a range key iterator stack, invalidate the
2643 1 : // iterator. This ensures RangeKeyChanged() returns true if a
2644 1 : // subsequent positioning operation discovers a range key. It also
2645 1 : // prevents seek no-op optimizations.
2646 1 : i.invalidate()
2647 1 : }
2648 : }
2649 :
2650 : // If the iterator is backed by a batch that's been mutated, refresh its
2651 : // existing point and range-key iterators, and invalidate the iterator to
2652 : // prevent seek-using-next optimizations. If we don't yet have a point-key
2653 : // iterator or range-key iterator but we require one, it'll be created in
2654 : // the slow path that reconstructs the iterator in finishInitializingIter.
2655 1 : if i.batch != nil {
2656 1 : nextBatchSeqNum := (base.SeqNum(len(i.batch.batch.data)) | base.SeqNumBatchBit)
2657 1 : if nextBatchSeqNum != i.batch.batchSeqNum {
2658 1 : i.batch.batchSeqNum = nextBatchSeqNum
2659 1 : if i.merging != nil {
2660 1 : i.merging.batchSnapshot = nextBatchSeqNum
2661 1 : }
2662 : // Prevent a no-op seek optimization on the next seek. We won't be
2663 : // able to reuse the top-level Iterator state, because it may be
2664 : // incorrect after the inclusion of new batch mutations.
2665 1 : i.batchJustRefreshed = true
2666 1 : if i.pointIter != nil && i.batch.batch.countRangeDels > 0 {
2667 1 : if i.batch.rangeDelIter.Count() == 0 {
2668 1 : // When we constructed this iterator, there were no
2669 1 : // rangedels in the batch. Iterator construction will
2670 1 : // have excluded the batch rangedel iterator from the
2671 1 : // point iterator stack. We need to reconstruct the
2672 1 : // point iterator to add i.batchRangeDelIter into the
2673 1 : // iterator stack.
2674 1 : i.err = firstError(i.err, i.pointIter.Close())
2675 1 : i.pointIter = nil
2676 1 : } else {
2677 1 : // There are range deletions in the batch and we already
2678 1 : // have a batch rangedel iterator. We can update the
2679 1 : // batch rangedel iterator in place.
2680 1 : //
2681 1 : // NB: There may or may not be new range deletions. We
2682 1 : // can't tell based on i.batchRangeDelIter.Count(),
2683 1 : // which is the count of fragmented range deletions, NOT
2684 1 : // the number of range deletions written to the batch
2685 1 : // [i.batch.countRangeDels].
2686 1 : i.batch.batch.initRangeDelIter(&i.opts, &i.batch.rangeDelIter, nextBatchSeqNum)
2687 1 : }
2688 : }
2689 1 : if i.rangeKey != nil && i.batch.batch.countRangeKeys > 0 {
2690 1 : if i.batch.rangeKeyIter.Count() == 0 {
2691 1 : // When we constructed this iterator, there were no range
2692 1 : // keys in the batch. Iterator construction will have
2693 1 : // excluded the batch rangekey iterator from the range key
2694 1 : // iterator stack. We need to reconstruct the range key
2695 1 : // iterator to add i.batchRangeKeyIter into the iterator
2696 1 : // stack.
2697 1 : i.rangeKey.rangeKeyIter.Close()
2698 1 : i.rangeKey = nil
2699 1 : } else {
2700 1 : // There are range keys in the batch and we already
2701 1 : // have a batch rangekey iterator. We can update the batch
2702 1 : // rangekey iterator in place.
2703 1 : //
2704 1 : // NB: There may or may not be new range keys. We can't
2705 1 : // tell based on i.batchRangeKeyIter.Count(), which is the
2706 1 : // count of fragmented range keys, NOT the number of
2707 1 : // range keys written to the batch [i.batch.countRangeKeys].
2708 1 : i.batch.batch.initRangeKeyIter(&i.opts, &i.batch.rangeKeyIter, nextBatchSeqNum)
2709 1 : i.invalidate()
2710 1 : }
2711 : }
2712 : }
2713 : }
2714 :
2715 : // Reset combinedIterState.initialized in case the iterator key types
2716 : // changed. If there's already a range key iterator stack, the combined
2717 : // iterator is already initialized. Additionally, if the iterator is not
2718 : // configured to include range keys, mark it as initialized to signal that
2719 : // lower level iterators should not trigger a switch to combined iteration.
2720 1 : i.lazyCombinedIter.combinedIterState = combinedIterState{
2721 1 : initialized: i.rangeKey != nil || !i.opts.rangeKeys(),
2722 1 : }
2723 1 :
2724 1 : boundsEqual := ((i.opts.LowerBound == nil) == (o.LowerBound == nil)) &&
2725 1 : ((i.opts.UpperBound == nil) == (o.UpperBound == nil)) &&
2726 1 : i.equal(i.opts.LowerBound, o.LowerBound) &&
2727 1 : i.equal(i.opts.UpperBound, o.UpperBound)
2728 1 :
2729 1 : if boundsEqual && o.KeyTypes == i.opts.KeyTypes &&
2730 1 : (i.pointIter != nil || !i.opts.pointKeys()) &&
2731 1 : (i.rangeKey != nil || !i.opts.rangeKeys() || i.opts.KeyTypes == IterKeyTypePointsAndRanges) &&
2732 1 : i.comparer.CompareRangeSuffixes(o.RangeKeyMasking.Suffix, i.opts.RangeKeyMasking.Suffix) == 0 &&
2733 1 : o.UseL6Filters == i.opts.UseL6Filters {
2734 1 : // The options are identical, so we can likely use the fast path. In
2735 1 : // addition to all the above constraints, we cannot use the fast path if
2736 1 : // configured to perform lazy combined iteration but an indexed batch
2737 1 : // used by the iterator now contains range keys. Lazy combined iteration
2738 1 : // is not compatible with batch range keys because we always need to
2739 1 : // merge the batch's range keys into iteration.
2740 1 : if i.rangeKey != nil || !i.opts.rangeKeys() || i.batch == nil || i.batch.batch.countRangeKeys == 0 {
2741 1 : // Fast path. This preserves the Seek-using-Next optimizations as
2742 1 : // long as the iterator wasn't already invalidated up above.
2743 1 : return
2744 1 : }
2745 : }
2746 : // Slow path.
2747 :
2748 : // The options changed. Save the new ones to i.opts.
2749 1 : if boundsEqual {
2750 1 : // Copying the options into i.opts will overwrite LowerBound and
2751 1 : // UpperBound fields with the user-provided slices. We need to hold on
2752 1 : // to the Pebble-owned slices, so save them and re-set them after the
2753 1 : // copy.
2754 1 : lower, upper := i.opts.LowerBound, i.opts.UpperBound
2755 1 : i.opts = *o
2756 1 : i.opts.LowerBound, i.opts.UpperBound = lower, upper
2757 1 : } else {
2758 1 : i.opts = *o
2759 1 : i.processBounds(o.LowerBound, o.UpperBound)
2760 1 : // Propagate the changed bounds to the existing point iterator.
2761 1 : // NB: We propagate i.opts.{Lower,Upper}Bound, not o.{Lower,Upper}Bound
2762 1 : // because i.opts now point to buffers owned by Pebble.
2763 1 : if i.pointIter != nil {
2764 1 : i.pointIter.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2765 1 : }
2766 1 : if i.rangeKey != nil {
2767 1 : i.rangeKey.iterConfig.SetBounds(i.opts.LowerBound, i.opts.UpperBound)
2768 1 : }
2769 : }
2770 :
2771 : // Even though this is not a positioning operation, the invalidation of the
2772 : // iterator stack means we cannot optimize Seeks by using Next.
2773 1 : i.invalidate()
2774 1 :
2775 1 : // Iterators created through NewExternalIter have a different iterator
2776 1 : // initialization process.
2777 1 : if i.externalIter != nil {
2778 1 : _ = finishInitializingExternal(i.ctx, i)
2779 1 : return
2780 1 : }
2781 1 : finishInitializingIter(i.ctx, i.alloc)
2782 : }
2783 :
2784 1 : func (i *Iterator) invalidate() {
2785 1 : i.lastPositioningOp = unknownLastPositionOp
2786 1 : i.hasPrefix = false
2787 1 : i.iterKV = nil
2788 1 : i.err = nil
2789 1 : // This switch statement isn't necessary for correctness since callers
2790 1 : // should call a repositioning method. We could have arbitrarily set i.pos
2791 1 : // to one of the values. But it results in more intuitive behavior in
2792 1 : // tests, which do not always reposition.
2793 1 : switch i.pos {
2794 1 : case iterPosCurForward, iterPosNext, iterPosCurForwardPaused:
2795 1 : i.pos = iterPosCurForward
2796 1 : case iterPosCurReverse, iterPosPrev, iterPosCurReversePaused:
2797 1 : i.pos = iterPosCurReverse
2798 : }
2799 1 : i.iterValidityState = IterExhausted
2800 1 : if i.rangeKey != nil {
2801 1 : i.rangeKey.iiter.Invalidate()
2802 1 : i.rangeKey.prevPosHadRangeKey = false
2803 1 : }
2804 : }
2805 :
2806 : // Metrics returns per-iterator metrics.
2807 0 : func (i *Iterator) Metrics() IteratorMetrics {
2808 0 : m := IteratorMetrics{
2809 0 : ReadAmp: 1,
2810 0 : }
2811 0 : if mi, ok := i.iter.(*mergingIter); ok {
2812 0 : m.ReadAmp = len(mi.levels)
2813 0 : }
2814 0 : return m
2815 : }
2816 :
2817 : // ResetStats resets the stats to 0.
2818 0 : func (i *Iterator) ResetStats() {
2819 0 : i.stats = IteratorStats{}
2820 0 : }
2821 :
2822 : // Stats returns the current stats.
2823 1 : func (i *Iterator) Stats() IteratorStats {
2824 1 : return i.stats
2825 1 : }
2826 :
2827 : // CloneOptions configures an iterator constructed through Iterator.Clone.
2828 : type CloneOptions struct {
2829 : // IterOptions, if non-nil, define the iterator options to configure a
2830 : // cloned iterator. If nil, the clone adopts the same IterOptions as the
2831 : // iterator being cloned.
2832 : IterOptions *IterOptions
2833 : // RefreshBatchView may be set to true when cloning an Iterator over an
2834 : // indexed batch. When false, the clone adopts the same (possibly stale)
2835 : // view of the indexed batch as the cloned Iterator. When true, the clone is
2836 : // constructed with a refreshed view of the batch, observing all of the
2837 : // batch's mutations at the time of the Clone. If the cloned iterator was
2838 : // not constructed to read over an indexed batch, RefreshVatchView has no
2839 : // effect.
2840 : RefreshBatchView bool
2841 : }
2842 :
2843 : // Clone creates a new Iterator over the same underlying data, i.e., over the
2844 : // same {batch, memtables, sstables}). The resulting iterator is not positioned.
2845 : // It starts with the same IterOptions, unless opts.IterOptions is set.
2846 : //
2847 : // When called on an Iterator over an indexed batch, the clone's visibility of
2848 : // the indexed batch is determined by CloneOptions.RefreshBatchView. If false,
2849 : // the clone inherits the iterator's current (possibly stale) view of the batch,
2850 : // and callers may call SetOptions to subsequently refresh the clone's view to
2851 : // include all batch mutations. If true, the clone is constructed with a
2852 : // complete view of the indexed batch's mutations at the time of the Clone.
2853 : //
2854 : // Callers can use Clone if they need multiple iterators that need to see
2855 : // exactly the same underlying state of the DB. This should not be used to
2856 : // extend the lifetime of the data backing the original Iterator since that
2857 : // will cause an increase in memory and disk usage (use NewSnapshot for that
2858 : // purpose).
2859 1 : func (i *Iterator) Clone(opts CloneOptions) (*Iterator, error) {
2860 1 : return i.CloneWithContext(context.Background(), opts)
2861 1 : }
2862 :
2863 : // CloneWithContext is like Clone, and additionally accepts a context for
2864 : // tracing.
2865 1 : func (i *Iterator) CloneWithContext(ctx context.Context, opts CloneOptions) (*Iterator, error) {
2866 1 : if opts.IterOptions == nil {
2867 1 : opts.IterOptions = &i.opts
2868 1 : }
2869 1 : if i.batchOnlyIter {
2870 1 : return nil, errors.Errorf("cannot Clone a batch-only Iterator")
2871 1 : }
2872 1 : readState := i.readState
2873 1 : vers := i.version
2874 1 : if readState == nil && vers == nil {
2875 1 : return nil, errors.Errorf("cannot Clone a closed Iterator")
2876 1 : }
2877 : // i is already holding a ref, so there is no race with unref here.
2878 : //
2879 : // TODO(bilal): If the underlying iterator was created on a snapshot, we could
2880 : // grab a reference to the current readState instead of reffing the original
2881 : // readState. This allows us to release references to some zombie sstables
2882 : // and memtables.
2883 1 : if readState != nil {
2884 1 : readState.ref()
2885 1 : }
2886 1 : if vers != nil {
2887 1 : vers.Ref()
2888 1 : }
2889 : // Bundle various structures under a single umbrella in order to allocate
2890 : // them together.
2891 1 : buf := newIterAlloc()
2892 1 : dbi := &buf.dbi
2893 1 : *dbi = Iterator{
2894 1 : ctx: ctx,
2895 1 : opts: *opts.IterOptions,
2896 1 : alloc: buf,
2897 1 : merge: i.merge,
2898 1 : comparer: i.comparer,
2899 1 : readState: readState,
2900 1 : version: vers,
2901 1 : keyBuf: buf.keyBuf,
2902 1 : prefixOrFullSeekKey: buf.prefixOrFullSeekKey,
2903 1 : boundsBuf: buf.boundsBuf,
2904 1 : fc: i.fc,
2905 1 : newIters: i.newIters,
2906 1 : newIterRangeKey: i.newIterRangeKey,
2907 1 : valueRetrievalProfile: i.valueRetrievalProfile,
2908 1 : seqNum: i.seqNum,
2909 1 : tracker: i.tracker,
2910 1 : }
2911 1 : if i.tracker != nil && !dbi.opts.ExemptFromTracking {
2912 1 : dbi.trackerHandle = i.tracker.Start()
2913 1 : }
2914 1 : if i.batch != nil {
2915 1 : dbi.batch = &buf.batchState
2916 1 : dbi.batch.batch = i.batch.batch
2917 1 : dbi.batch.batchSeqNum = i.batch.batchSeqNum
2918 1 : // If the caller requested the clone have a current view of the indexed
2919 1 : // batch, set the clone's batch sequence number appropriately.
2920 1 : if opts.RefreshBatchView {
2921 1 : dbi.batch.batchSeqNum = (base.SeqNum(len(i.batch.batch.data)) | base.SeqNumBatchBit)
2922 1 : }
2923 : }
2924 1 : dbi.processBounds(dbi.opts.LowerBound, dbi.opts.UpperBound)
2925 1 : return finishInitializingIter(ctx, buf), nil
2926 : }
2927 :
2928 : // Merge adds all of the argument's statistics to the receiver. It may be used
2929 : // to accumulate stats across multiple iterators.
2930 1 : func (stats *IteratorStats) Merge(o IteratorStats) {
2931 1 : for i := InterfaceCall; i < NumStatsKind; i++ {
2932 1 : stats.ForwardSeekCount[i] += o.ForwardSeekCount[i]
2933 1 : stats.ReverseSeekCount[i] += o.ReverseSeekCount[i]
2934 1 : stats.ForwardStepCount[i] += o.ForwardStepCount[i]
2935 1 : stats.ReverseStepCount[i] += o.ReverseStepCount[i]
2936 1 : }
2937 1 : stats.InternalStats.Merge(o.InternalStats)
2938 1 : stats.RangeKeyStats.Merge(o.RangeKeyStats)
2939 : }
2940 :
2941 1 : func (stats *IteratorStats) String() string {
2942 1 : return redact.StringWithoutMarkers(stats)
2943 1 : }
2944 :
2945 : // SafeFormat implements the redact.SafeFormatter interface.
2946 1 : func (stats *IteratorStats) SafeFormat(s redact.SafePrinter, verb rune) {
2947 1 : if stats.ReverseSeekCount[InterfaceCall] == 0 && stats.ReverseSeekCount[InternalIterCall] == 0 {
2948 1 : s.Printf("seeked %s times (%s internal)",
2949 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InterfaceCall])),
2950 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InternalIterCall])),
2951 1 : )
2952 1 : } else {
2953 1 : s.Printf("seeked %s times (%s fwd/%s rev, internal: %s fwd/%s rev)",
2954 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InterfaceCall]+stats.ReverseSeekCount[InterfaceCall])),
2955 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InterfaceCall])),
2956 1 : humanize.Count.Uint64(uint64(stats.ReverseSeekCount[InterfaceCall])),
2957 1 : humanize.Count.Uint64(uint64(stats.ForwardSeekCount[InternalIterCall])),
2958 1 : humanize.Count.Uint64(uint64(stats.ReverseSeekCount[InternalIterCall])),
2959 1 : )
2960 1 : }
2961 1 : s.SafeString("; ")
2962 1 :
2963 1 : if stats.ReverseStepCount[InterfaceCall] == 0 && stats.ReverseStepCount[InternalIterCall] == 0 {
2964 1 : s.Printf("stepped %s times (%s internal)",
2965 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InterfaceCall])),
2966 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InternalIterCall])),
2967 1 : )
2968 1 : } else {
2969 1 : s.Printf("stepped %s times (%s fwd/%s rev, internal: %s fwd/%s rev)",
2970 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InterfaceCall]+stats.ReverseStepCount[InterfaceCall])),
2971 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InterfaceCall])),
2972 1 : humanize.Count.Uint64(uint64(stats.ReverseStepCount[InterfaceCall])),
2973 1 : humanize.Count.Uint64(uint64(stats.ForwardStepCount[InternalIterCall])),
2974 1 : humanize.Count.Uint64(uint64(stats.ReverseStepCount[InternalIterCall])),
2975 1 : )
2976 1 : }
2977 :
2978 1 : if stats.InternalStats != (InternalIteratorStats{}) {
2979 1 : s.SafeString("; ")
2980 1 : stats.InternalStats.SafeFormat(s, verb)
2981 1 : }
2982 1 : if stats.RangeKeyStats != (RangeKeyIteratorStats{}) {
2983 1 : s.SafeString(", ")
2984 1 : stats.RangeKeyStats.SafeFormat(s, verb)
2985 1 : }
2986 : }
2987 :
2988 : // CanDeterministicallySingleDelete takes a valid iterator and examines internal
2989 : // state to determine if a SingleDelete deleting Iterator.Key() would
2990 : // deterministically delete the key. CanDeterministicallySingleDelete requires
2991 : // the iterator to be oriented in the forward direction (eg, the last
2992 : // positioning operation must've been a First, a Seek[Prefix]GE, or a
2993 : // Next[Prefix][WithLimit]).
2994 : //
2995 : // This function does not change the external position of the iterator, and all
2996 : // positioning methods should behave the same as if it was never called. This
2997 : // function will only return a meaningful result the first time it's invoked at
2998 : // an iterator position. This function invalidates the iterator Value's memory,
2999 : // and the caller must not rely on the memory safety of the previous Iterator
3000 : // position.
3001 : //
3002 : // If CanDeterministicallySingleDelete returns true AND the key at the iterator
3003 : // position is not modified between the creation of the Iterator and the commit
3004 : // of a batch containing a SingleDelete over the key, then the caller can be
3005 : // assured that SingleDelete is equivalent to Delete on the local engine, but it
3006 : // may not be true on another engine that received the same writes and with
3007 : // logically equivalent state since this engine may have collapsed multiple SETs
3008 : // into one.
3009 1 : func CanDeterministicallySingleDelete(it *Iterator) (bool, error) {
3010 1 : // This function may only be called once per external iterator position. We
3011 1 : // can validate this by checking the last positioning operation.
3012 1 : if it.lastPositioningOp == internalNextOp {
3013 1 : return false, errors.New("pebble: CanDeterministicallySingleDelete called twice")
3014 1 : }
3015 1 : validity, kind := it.internalNext()
3016 1 : var shadowedBySingleDelete bool
3017 1 : for validity == internalNextValid {
3018 1 : switch kind {
3019 1 : case InternalKeyKindDelete, InternalKeyKindDeleteSized:
3020 1 : // A DEL or DELSIZED tombstone is okay. An internal key
3021 1 : // sequence like SINGLEDEL; SET; DEL; SET can be handled
3022 1 : // deterministically. If there are SETs further down, we
3023 1 : // don't care about them.
3024 1 : return true, nil
3025 1 : case InternalKeyKindSingleDelete:
3026 1 : // A SingleDelete is okay as long as when that SingleDelete was
3027 1 : // written, it was written deterministically (eg, with its own
3028 1 : // CanDeterministicallySingleDelete check). Validate that it was
3029 1 : // written deterministically. We'll allow one set to appear after
3030 1 : // the SingleDelete.
3031 1 : shadowedBySingleDelete = true
3032 1 : validity, kind = it.internalNext()
3033 1 : continue
3034 1 : case InternalKeyKindSet, InternalKeyKindSetWithDelete, InternalKeyKindMerge:
3035 1 : // If we observed a single delete, it's allowed to delete 1 key.
3036 1 : // We'll keep looping to validate that the internal keys beneath the
3037 1 : // already-written single delete are copacetic.
3038 1 : if shadowedBySingleDelete {
3039 1 : shadowedBySingleDelete = false
3040 1 : validity, kind = it.internalNext()
3041 1 : continue
3042 : }
3043 : // We encountered a shadowed SET, SETWITHDEL, MERGE. A SINGLEDEL
3044 : // that deleted the KV at the original iterator position could
3045 : // result in this key becoming visible.
3046 1 : return false, nil
3047 0 : case InternalKeyKindRangeDelete:
3048 0 : // RangeDeletes are handled by the merging iterator and should never
3049 0 : // be observed by the top-level Iterator.
3050 0 : panic(errors.AssertionFailedf("pebble: unexpected range delete"))
3051 0 : case InternalKeyKindRangeKeySet, InternalKeyKindRangeKeyUnset, InternalKeyKindRangeKeyDelete:
3052 0 : // Range keys are interleaved at the maximal sequence number and
3053 0 : // should never be observed within a user key.
3054 0 : panic(errors.AssertionFailedf("pebble: unexpected range key"))
3055 0 : default:
3056 0 : panic(errors.AssertionFailedf("pebble: unexpected key kind: %s", errors.Safe(kind)))
3057 : }
3058 : }
3059 1 : if validity == internalNextError {
3060 1 : return false, it.Error()
3061 1 : }
3062 1 : return true, nil
3063 : }
3064 :
3065 : // internalNextValidity enumerates the potential outcomes of a call to
3066 : // internalNext.
3067 : type internalNextValidity int8
3068 :
3069 : const (
3070 : // internalNextError is returned by internalNext when an error occurred and
3071 : // the caller is responsible for checking iter.Error().
3072 : internalNextError internalNextValidity = iota
3073 : // internalNextExhausted is returned by internalNext when the next internal
3074 : // key is an internal key with a different user key than Iterator.Key().
3075 : internalNextExhausted
3076 : // internalNextValid is returned by internalNext when the internal next
3077 : // found a shadowed internal key with a user key equal to Iterator.Key().
3078 : internalNextValid
3079 : )
3080 :
3081 : // internalNext advances internal Iterator state forward to expose the
3082 : // InternalKeyKind of the next internal key with a user key equal to Key().
3083 : //
3084 : // internalNext is a highly specialized operation and is unlikely to be
3085 : // generally useful. See Iterator.Next for how to reposition the iterator to the
3086 : // next key. internalNext requires the Iterator to be at a valid position in the
3087 : // forward direction (the last positioning operation must've been a First, a
3088 : // Seek[Prefix]GE, or a Next[Prefix][WithLimit] and Valid() must return true).
3089 : //
3090 : // internalNext, unlike all other Iterator methods, exposes internal LSM state.
3091 : // internalNext advances the Iterator's internal iterator to the next shadowed
3092 : // key with a user key equal to Key(). When a key is overwritten or deleted, its
3093 : // removal from the LSM occurs lazily as a part of compactions. internalNext
3094 : // allows the caller to see whether an obsolete internal key exists with the
3095 : // current Key(), and what it's key kind is. Note that the existence of an
3096 : // internal key is nondeterministic and dependent on internal LSM state. These
3097 : // semantics are unlikely to be applicable to almost all use cases.
3098 : //
3099 : // If internalNext finds a key that shares the same user key as Key(), it
3100 : // returns internalNextValid and the internal key's kind. If internalNext
3101 : // encounters an error, it returns internalNextError and the caller is expected
3102 : // to call Iterator.Error() to retrieve it. In all other circumstances,
3103 : // internalNext returns internalNextExhausted, indicating that there are no more
3104 : // additional internal keys with the user key Key().
3105 : //
3106 : // internalNext does not change the external position of the iterator, and a
3107 : // Next operation should behave the same as if internalNext was never called.
3108 : // internalNext does invalidate the iterator Value's memory, and the caller must
3109 : // not rely on the memory safety of the previous Iterator position.
3110 1 : func (i *Iterator) internalNext() (internalNextValidity, base.InternalKeyKind) {
3111 1 : i.stats.ForwardStepCount[InterfaceCall]++
3112 1 : if i.err != nil {
3113 1 : return internalNextError, base.InternalKeyKindInvalid
3114 1 : } else if i.iterValidityState != IterValid {
3115 1 : return internalNextExhausted, base.InternalKeyKindInvalid
3116 1 : }
3117 1 : i.lastPositioningOp = internalNextOp
3118 1 :
3119 1 : switch i.pos {
3120 1 : case iterPosCurForward:
3121 1 : i.iterKV = i.iter.Next()
3122 1 : if i.iterKV == nil {
3123 1 : // We check i.iter.Error() here and return an internalNextError enum
3124 1 : // variant so that the caller does not need to check i.iter.Error()
3125 1 : // in the common case that the next internal key has a new user key.
3126 1 : if i.err = i.iter.Error(); i.err != nil {
3127 0 : return internalNextError, base.InternalKeyKindInvalid
3128 0 : }
3129 1 : i.pos = iterPosNext
3130 1 : return internalNextExhausted, base.InternalKeyKindInvalid
3131 1 : } else if i.comparer.Equal(i.iterKV.K.UserKey, i.key) {
3132 1 : return internalNextValid, i.iterKV.Kind()
3133 1 : }
3134 1 : i.pos = iterPosNext
3135 1 : return internalNextExhausted, base.InternalKeyKindInvalid
3136 1 : case iterPosCurReverse, iterPosCurReversePaused, iterPosPrev:
3137 1 : i.err = errors.New("switching from reverse to forward via internalNext is prohibited")
3138 1 : i.iterValidityState = IterExhausted
3139 1 : return internalNextError, base.InternalKeyKindInvalid
3140 1 : case iterPosNext, iterPosCurForwardPaused:
3141 1 : // The previous method already moved onto the next user key. This is
3142 1 : // only possible if
3143 1 : // - the last positioning method was a call to internalNext, and we
3144 1 : // advanced to a new user key.
3145 1 : // - the previous non-internalNext iterator operation encountered a
3146 1 : // range key or merge, forcing an internal Next that found a new
3147 1 : // user key that's not equal to i.Iterator.Key().
3148 1 : return internalNextExhausted, base.InternalKeyKindInvalid
3149 0 : default:
3150 0 : panic("unreachable")
3151 : }
3152 : }
3153 :
3154 : var _ base.IteratorDebug = (*Iterator)(nil)
3155 :
3156 : // DebugTree implements the base.IteratorDebug interface.
3157 0 : func (i *Iterator) DebugTree(tp treeprinter.Node) {
3158 0 : n := tp.Childf("%T(%p)", i, i)
3159 0 : if i.iter != nil {
3160 0 : i.iter.DebugTree(n)
3161 0 : }
3162 0 : if i.pointIter != nil {
3163 0 : i.pointIter.DebugTree(n)
3164 0 : }
3165 : }
|
