Line data Source code
1 : // Copyright 2018 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "bytes"
9 : "context"
10 : "fmt"
11 : "runtime/debug"
12 : "unsafe"
13 :
14 : "github.com/cockroachdb/errors"
15 : "github.com/cockroachdb/pebble/internal/base"
16 : "github.com/cockroachdb/pebble/internal/invariants"
17 : "github.com/cockroachdb/pebble/internal/keyspan"
18 : "github.com/cockroachdb/pebble/internal/treeprinter"
19 : )
20 :
21 : type mergingIterLevel struct {
22 : index int
23 : iter internalIterator
24 : // rangeDelIter is set to the range-deletion iterator for the level. When
25 : // configured with a levelIter, this pointer changes as sstable boundaries
26 : // are crossed. See levelIter.initRangeDel and the Range Deletions comment
27 : // below.
28 : rangeDelIter keyspan.FragmentIterator
29 : // rangeDelIterGeneration is incremented whenever rangeDelIter changes.
30 : rangeDelIterGeneration int
31 : // iterKV caches the current key-value pair iter points to.
32 : iterKV *base.InternalKV
33 : // levelIter is non-nil if this level's iter is ultimately backed by a
34 : // *levelIter. The handle in iter may have wrapped the levelIter with
35 : // intermediary internalIterator implementations.
36 : levelIter *levelIter
37 :
38 : // tombstone caches the tombstone rangeDelIter is currently pointed at. If
39 : // tombstone is nil, there are no further tombstones within the
40 : // current sstable in the current iterator direction. The cached tombstone is
41 : // only valid for the levels in the range [0,heap[0].index]. This avoids
42 : // positioning tombstones at lower levels which cannot possibly shadow the
43 : // current key.
44 : tombstone *keyspan.Span
45 : }
46 :
47 1 : func (ml *mergingIterLevel) setRangeDelIter(iter keyspan.FragmentIterator) {
48 1 : ml.tombstone = nil
49 1 : if ml.rangeDelIter != nil {
50 1 : ml.rangeDelIter.Close()
51 1 : }
52 1 : ml.rangeDelIter = iter
53 1 : ml.rangeDelIterGeneration++
54 : }
55 :
56 : // mergingIter provides a merged view of multiple iterators from different
57 : // levels of the LSM.
58 : //
59 : // The core of a mergingIter is a heap of internalIterators (see
60 : // mergingIterHeap). The heap can operate as either a min-heap, used during
61 : // forward iteration (First, SeekGE, Next) or a max-heap, used during reverse
62 : // iteration (Last, SeekLT, Prev). The heap is initialized in calls to First,
63 : // Last, SeekGE, and SeekLT. A call to Next or Prev takes the current top
64 : // element on the heap, advances its iterator, and then "fixes" the heap
65 : // property. When one of the child iterators is exhausted during Next/Prev
66 : // iteration, it is removed from the heap.
67 : //
68 : // # Range Deletions
69 : //
70 : // A mergingIter can optionally be configured with a slice of range deletion
71 : // iterators. The range deletion iterator slice must exactly parallel the point
72 : // iterators and the range deletion iterator must correspond to the same level
73 : // in the LSM as the point iterator. Note that each memtable and each table in
74 : // L0 is a different "level" from the mergingIter perspective. So level 0 below
75 : // does not correspond to L0 in the LSM.
76 : //
77 : // A range deletion iterator iterates over fragmented range tombstones. Range
78 : // tombstones are fragmented by splitting them at any overlapping points. This
79 : // fragmentation guarantees that within an sstable tombstones will either be
80 : // distinct or will have identical start and end user keys. While range
81 : // tombstones are fragmented within an sstable, the start and end keys are not truncated
82 : // to sstable boundaries. This is necessary because the tombstone end key is
83 : // exclusive and does not have a sequence number. Consider an sstable
84 : // containing the range tombstone [a,c)#9 and the key "b#8". The tombstone must
85 : // delete "b#8", yet older versions of "b" might spill over to the next
86 : // sstable. So the boundary key for this sstable must be "b#8". Adjusting the
87 : // end key of tombstones to be optionally inclusive or contain a sequence
88 : // number would be possible solutions (such solutions have potentially serious
89 : // issues: tombstones have exclusive end keys since an inclusive deletion end can
90 : // be converted to an exclusive one while the reverse transformation is not possible;
91 : // the semantics of a sequence number for the end key of a range tombstone are murky).
92 : //
93 : // The approach taken here performs an
94 : // implicit truncation of the tombstone to the sstable boundaries.
95 : //
96 : // During initialization of a mergingIter, the range deletion iterators for
97 : // batches, memtables, and L0 tables are populated up front. Note that Batches
98 : // and memtables index unfragmented tombstones. Batch.newRangeDelIter() and
99 : // memTable.newRangeDelIter() fragment and cache the tombstones on demand. The
100 : // L1-L6 range deletion iterators are populated by levelIter. When configured
101 : // to load range deletion iterators, whenever a levelIter loads a table it
102 : // loads both the point iterator and the range deletion
103 : // iterator. levelIter.rangeDelIter is configured to point to the right entry
104 : // in mergingIter.levels. The effect of this setup is that
105 : // mergingIter.levels[i].rangeDelIter always contains the fragmented range
106 : // tombstone for the current table in level i that the levelIter has open.
107 : //
108 : // Another crucial mechanism of levelIter is that it materializes fake point
109 : // entries for the table boundaries if the boundary is range deletion
110 : // key. Consider a table that contains only a range tombstone [a-e)#10. The
111 : // sstable boundaries for this table will be a#10,15 and
112 : // e#72057594037927935,15. During forward iteration levelIter will return
113 : // e#72057594037927935,15 as a key. During reverse iteration levelIter will
114 : // return a#10,15 as a key. These sentinel keys act as bookends to point
115 : // iteration and allow mergingIter to keep a table and its associated range
116 : // tombstones loaded as long as there are keys at lower levels that are within
117 : // the bounds of the table.
118 : //
119 : // The final piece to the range deletion puzzle is the LSM invariant that for a
120 : // given key K newer versions of K can only exist earlier in the level, or at
121 : // higher levels of the tree. For example, if K#4 exists in L3, k#5 can only
122 : // exist earlier in the L3 or in L0, L1, L2 or a memtable. Get very explicitly
123 : // uses this invariant to find the value for a key by walking the LSM level by
124 : // level. For range deletions, this invariant means that a range deletion at
125 : // level N will necessarily shadow any keys within its bounds in level Y where
126 : // Y > N. One wrinkle to this statement is that it only applies to keys that
127 : // lie within the sstable bounds as well, but we get that guarantee due to the
128 : // way the range deletion iterator and point iterator are bound together by a
129 : // levelIter.
130 : //
131 : // Tying the above all together, we get a picture where each level (index in
132 : // mergingIter.levels) is composed of both point operations (pX) and range
133 : // deletions (rX). The range deletions for level X shadow both the point
134 : // operations and range deletions for level Y where Y > X allowing mergingIter
135 : // to skip processing entries in that shadow. For example, consider the
136 : // scenario:
137 : //
138 : // r0: a---e
139 : // r1: d---h
140 : // r2: g---k
141 : // r3: j---n
142 : // r4: m---q
143 : //
144 : // This is showing 5 levels of range deletions. Consider what happens upon
145 : // SeekGE("b"). We first seek the point iterator for level 0 (the point values
146 : // are not shown above) and we then seek the range deletion iterator. That
147 : // returns the tombstone [a,e). This tombstone tells us that all keys in the
148 : // range [a,e) in lower levels are deleted so we can skip them. So we can
149 : // adjust the seek key to "e", the tombstone end key. For level 1 we seek to
150 : // "e" and find the range tombstone [d,h) and similar logic holds. By the time
151 : // we get to level 4 we're seeking to "n".
152 : //
153 : // One consequence of not truncating tombstone end keys to sstable boundaries
154 : // is the seeking process described above cannot always seek to the tombstone
155 : // end key in the older level. For example, imagine in the above example r3 is
156 : // a partitioned level (i.e., L1+ in our LSM), and the sstable containing [j,
157 : // n) has "k" as its upper boundary. In this situation, compactions involving
158 : // keys at or after "k" can output those keys to r4+, even if they're newer
159 : // than our tombstone [j, n). So instead of seeking to "n" in r4 we can only
160 : // seek to "k". To achieve this, the instance variable `largestUserKey.`
161 : // maintains the upper bounds of the current sstables in the partitioned
162 : // levels. In this example, `levels[3].largestUserKey` holds "k", telling us to
163 : // limit the seek triggered by a tombstone in r3 to "k".
164 : //
165 : // During actual iteration levels can contain both point operations and range
166 : // deletions. Within a level, when a range deletion contains a point operation
167 : // the sequence numbers must be checked to determine if the point operation is
168 : // newer or older than the range deletion tombstone. The mergingIter maintains
169 : // the invariant that the range deletion iterators for all levels newer that
170 : // the current iteration key (L < m.heap.items[0].index) are positioned at the
171 : // next (or previous during reverse iteration) range deletion tombstone. We
172 : // know those levels don't contain a range deletion tombstone that covers the
173 : // current key because if they did the current key would be deleted. The range
174 : // deletion iterator for the current key's level is positioned at a range
175 : // tombstone covering or past the current key. The position of all of other
176 : // range deletion iterators is unspecified. Whenever a key from those levels
177 : // becomes the current key, their range deletion iterators need to be
178 : // positioned. This lazy positioning avoids seeking the range deletion
179 : // iterators for keys that are never considered. (A similar bit of lazy
180 : // evaluation can be done for the point iterators, but is still TBD).
181 : //
182 : // For a full example, consider the following setup:
183 : //
184 : // p0: o
185 : // r0: m---q
186 : //
187 : // p1: n p
188 : // r1: g---k
189 : //
190 : // p2: b d i
191 : // r2: a---e q----v
192 : //
193 : // p3: e
194 : // r3:
195 : //
196 : // If we start iterating from the beginning, the first key we encounter is "b"
197 : // in p2. When the mergingIter is pointing at a valid entry, the range deletion
198 : // iterators for all of the levels < m.heap.items[0].index are positioned at
199 : // the next range tombstone past the current key. So r0 will point at [m,q) and
200 : // r1 at [g,k). When the key "b" is encountered, we check to see if the current
201 : // tombstone for r0 or r1 contains it, and whether the tombstone for r2, [a,e),
202 : // contains and is newer than "b".
203 : //
204 : // Advancing the iterator finds the next key at "d". This is in the same level
205 : // as the previous key "b" so we don't have to reposition any of the range
206 : // deletion iterators, but merely check whether "d" is now contained by any of
207 : // the range tombstones at higher levels or has stepped past the range
208 : // tombstone in its own level or higher levels. In this case, there is nothing to be done.
209 : //
210 : // Advancing the iterator again finds "e". Since "e" comes from p3, we have to
211 : // position the r3 range deletion iterator, which is empty. "e" is past the r2
212 : // tombstone of [a,e) so we need to advance the r2 range deletion iterator to
213 : // [q,v).
214 : //
215 : // The next key is "i". Because this key is in p2, a level above "e", we don't
216 : // have to reposition any range deletion iterators and instead see that "i" is
217 : // covered by the range tombstone [g,k). The iterator is immediately advanced
218 : // to "n" which is covered by the range tombstone [m,q) causing the iterator to
219 : // advance to "o" which is visible.
220 : //
221 : // # Error handling
222 : //
223 : // Any iterator operation may fail. The InternalIterator contract dictates that
224 : // an iterator must return a nil internal key when an error occurs, and a
225 : // subsequent call to Error() should return the error value. The exported
226 : // merging iterator positioning methods must adhere to this contract by setting
227 : // m.err to hold any error encountered by the individual level iterators and
228 : // returning a nil internal key. Some internal helpers (eg,
229 : // find[Next|Prev]Entry) also adhere to this contract, setting m.err directly).
230 : // Other internal functions return an explicit error return value and DO NOT set
231 : // m.err, relying on the caller to set m.err appropriately.
232 : //
233 : // TODO(jackson): Update the InternalIterator interface to return explicit error
234 : // return values (and an *InternalKV pointer).
235 : //
236 : // TODO(peter,rangedel): For testing, advance the iterator through various
237 : // scenarios and have each step display the current state (i.e. the current
238 : // heap and range-del iterator positioning).
239 : type mergingIter struct {
240 : logger Logger
241 : split Split
242 : dir int
243 : snapshot base.SeqNum
244 : batchSnapshot base.SeqNum
245 : levels []mergingIterLevel
246 : heap mergingIterHeap
247 : err error
248 : prefix []byte
249 : lower []byte
250 : upper []byte
251 : stats *InternalIteratorStats
252 : seekKeyBuf []byte
253 :
254 : // levelsPositioned, if non-nil, is a slice of the same length as levels.
255 : // It's used by NextPrefix to record which levels have already been
256 : // repositioned. It's created lazily by the first call to NextPrefix.
257 : levelsPositioned []bool
258 :
259 : combinedIterState *combinedIterState
260 :
261 : // Used in some tests to disable the random disabling of seek optimizations.
262 : forceEnableSeekOpt bool
263 : }
264 :
265 : // mergingIter implements the base.InternalIterator interface.
266 : var _ base.InternalIterator = (*mergingIter)(nil)
267 :
268 : // newMergingIter returns an iterator that merges its input. Walking the
269 : // resultant iterator will return all key/value pairs of all input iterators
270 : // in strictly increasing key order, as defined by cmp. It is permissible to
271 : // pass a nil split parameter if the caller is never going to call
272 : // SeekPrefixGE.
273 : //
274 : // The input's key ranges may overlap, but there are assumed to be no duplicate
275 : // keys: if iters[i] contains a key k then iters[j] will not contain that key k.
276 : //
277 : // None of the iters may be nil.
278 : func newMergingIter(
279 : logger Logger,
280 : stats *base.InternalIteratorStats,
281 : cmp Compare,
282 : split Split,
283 : iters ...internalIterator,
284 1 : ) *mergingIter {
285 1 : m := &mergingIter{}
286 1 : levels := make([]mergingIterLevel, len(iters))
287 1 : for i := range levels {
288 1 : levels[i].iter = iters[i]
289 1 : }
290 1 : m.init(&IterOptions{logger: logger}, stats, cmp, split, levels...)
291 1 : return m
292 : }
293 :
294 : func (m *mergingIter) init(
295 : opts *IterOptions,
296 : stats *base.InternalIteratorStats,
297 : cmp Compare,
298 : split Split,
299 : levels ...mergingIterLevel,
300 1 : ) {
301 1 : m.err = nil // clear cached iteration error
302 1 : m.logger = opts.getLogger()
303 1 : if opts != nil {
304 1 : m.lower = opts.LowerBound
305 1 : m.upper = opts.UpperBound
306 1 : }
307 1 : m.snapshot = base.SeqNumMax
308 1 : m.batchSnapshot = base.SeqNumMax
309 1 : m.levels = levels
310 1 : m.heap.cmp = cmp
311 1 : m.split = split
312 1 : m.stats = stats
313 1 : if cap(m.heap.items) < len(levels) {
314 1 : m.heap.items = make([]*mergingIterLevel, 0, len(levels))
315 1 : } else {
316 1 : m.heap.items = m.heap.items[:0]
317 1 : }
318 1 : for l := range m.levels {
319 1 : m.levels[l].index = l
320 1 : }
321 : }
322 :
323 1 : func (m *mergingIter) initHeap() {
324 1 : m.heap.items = m.heap.items[:0]
325 1 : for i := range m.levels {
326 1 : if l := &m.levels[i]; l.iterKV != nil {
327 1 : m.heap.items = append(m.heap.items, l)
328 1 : }
329 : }
330 1 : m.heap.init()
331 : }
332 :
333 1 : func (m *mergingIter) initMinHeap() error {
334 1 : m.dir = 1
335 1 : m.heap.reverse = false
336 1 : m.initHeap()
337 1 : return m.initMinRangeDelIters(-1)
338 1 : }
339 :
340 : // The level of the previous top element was oldTopLevel. Note that all range delete
341 : // iterators < oldTopLevel are positioned past the key of the previous top element and
342 : // the range delete iterator == oldTopLevel is positioned at or past the key of the
343 : // previous top element. We need to position the range delete iterators from oldTopLevel + 1
344 : // to the level of the current top element.
345 1 : func (m *mergingIter) initMinRangeDelIters(oldTopLevel int) error {
346 1 : if m.heap.len() == 0 {
347 1 : return nil
348 1 : }
349 :
350 : // Position the range-del iterators at levels <= m.heap.items[0].index.
351 1 : item := m.heap.items[0]
352 1 : for level := oldTopLevel + 1; level <= item.index; level++ {
353 1 : l := &m.levels[level]
354 1 : if l.rangeDelIter == nil {
355 1 : continue
356 : }
357 1 : var err error
358 1 : l.tombstone, err = l.rangeDelIter.SeekGE(item.iterKV.K.UserKey)
359 1 : if err != nil {
360 0 : return err
361 0 : }
362 : }
363 1 : return nil
364 : }
365 :
366 1 : func (m *mergingIter) initMaxHeap() error {
367 1 : m.dir = -1
368 1 : m.heap.reverse = true
369 1 : m.initHeap()
370 1 : return m.initMaxRangeDelIters(-1)
371 1 : }
372 :
373 : // The level of the previous top element was oldTopLevel. Note that all range delete
374 : // iterators < oldTopLevel are positioned before the key of the previous top element and
375 : // the range delete iterator == oldTopLevel is positioned at or before the key of the
376 : // previous top element. We need to position the range delete iterators from oldTopLevel + 1
377 : // to the level of the current top element.
378 1 : func (m *mergingIter) initMaxRangeDelIters(oldTopLevel int) error {
379 1 : if m.heap.len() == 0 {
380 1 : return nil
381 1 : }
382 : // Position the range-del iterators at levels <= m.heap.items[0].index.
383 1 : item := m.heap.items[0]
384 1 : for level := oldTopLevel + 1; level <= item.index; level++ {
385 1 : l := &m.levels[level]
386 1 : if l.rangeDelIter == nil {
387 1 : continue
388 : }
389 1 : tomb, err := keyspan.SeekLE(m.heap.cmp, l.rangeDelIter, item.iterKV.K.UserKey)
390 1 : if err != nil {
391 0 : return err
392 0 : }
393 1 : l.tombstone = tomb
394 : }
395 1 : return nil
396 : }
397 :
398 1 : func (m *mergingIter) switchToMinHeap() error {
399 1 : if m.heap.len() == 0 {
400 1 : if m.lower != nil {
401 1 : m.SeekGE(m.lower, base.SeekGEFlagsNone)
402 1 : } else {
403 1 : m.First()
404 1 : }
405 1 : return m.err
406 : }
407 :
408 : // We're switching from using a max heap to a min heap. We need to advance
409 : // any iterator that is less than or equal to the current key. Consider the
410 : // scenario where we have 2 iterators being merged (user-key:seq-num):
411 : //
412 : // i1: *a:2 b:2
413 : // i2: a:1 b:1
414 : //
415 : // The current key is a:2 and i2 is pointed at a:1. When we switch to forward
416 : // iteration, we want to return a key that is greater than a:2.
417 :
418 1 : key := m.heap.items[0].iterKV.K
419 1 : cur := m.heap.items[0]
420 1 :
421 1 : for i := range m.levels {
422 1 : l := &m.levels[i]
423 1 : if l == cur {
424 1 : continue
425 : }
426 1 : for l.iterKV = l.iter.Next(); l.iterKV != nil; l.iterKV = l.iter.Next() {
427 1 : if base.InternalCompare(m.heap.cmp, key, l.iterKV.K) < 0 {
428 1 : // key < iter-key
429 1 : break
430 : }
431 : // key >= iter-key
432 : }
433 1 : if l.iterKV == nil {
434 1 : if err := l.iter.Error(); err != nil {
435 1 : return err
436 1 : }
437 : }
438 : }
439 :
440 : // Special handling for the current iterator because we were using its key
441 : // above.
442 1 : cur.iterKV = cur.iter.Next()
443 1 : if cur.iterKV == nil {
444 1 : if err := cur.iter.Error(); err != nil {
445 1 : return err
446 1 : }
447 : }
448 1 : return m.initMinHeap()
449 : }
450 :
451 1 : func (m *mergingIter) switchToMaxHeap() error {
452 1 : if m.heap.len() == 0 {
453 1 : if m.upper != nil {
454 1 : m.SeekLT(m.upper, base.SeekLTFlagsNone)
455 1 : } else {
456 1 : m.Last()
457 1 : }
458 1 : return m.err
459 : }
460 :
461 : // We're switching from using a min heap to a max heap. We need to backup any
462 : // iterator that is greater than or equal to the current key. Consider the
463 : // scenario where we have 2 iterators being merged (user-key:seq-num):
464 : //
465 : // i1: a:2 *b:2
466 : // i2: a:1 b:1
467 : //
468 : // The current key is b:2 and i2 is pointing at b:1. When we switch to
469 : // reverse iteration, we want to return a key that is less than b:2.
470 1 : key := m.heap.items[0].iterKV.K
471 1 : cur := m.heap.items[0]
472 1 :
473 1 : for i := range m.levels {
474 1 : l := &m.levels[i]
475 1 : if l == cur {
476 1 : continue
477 : }
478 :
479 1 : for l.iterKV = l.iter.Prev(); l.iterKV != nil; l.iterKV = l.iter.Prev() {
480 1 : if base.InternalCompare(m.heap.cmp, key, l.iterKV.K) > 0 {
481 1 : // key > iter-key
482 1 : break
483 : }
484 : // key <= iter-key
485 : }
486 1 : if l.iterKV == nil {
487 1 : if err := l.iter.Error(); err != nil {
488 0 : return err
489 0 : }
490 : }
491 : }
492 :
493 : // Special handling for the current iterator because we were using its key
494 : // above.
495 1 : cur.iterKV = cur.iter.Prev()
496 1 : if cur.iterKV == nil {
497 1 : if err := cur.iter.Error(); err != nil {
498 1 : return err
499 1 : }
500 : }
501 1 : return m.initMaxHeap()
502 : }
503 :
504 : // nextEntry unconditionally steps to the next entry. item is the current top
505 : // item in the heap.
506 1 : func (m *mergingIter) nextEntry(l *mergingIterLevel, succKey []byte) error {
507 1 : // INVARIANT: If in prefix iteration mode, item.iterKey must have a prefix equal
508 1 : // to m.prefix. This invariant is important for ensuring TrySeekUsingNext
509 1 : // optimizations behave correctly.
510 1 : //
511 1 : // During prefix iteration, the iterator does not have a full view of the
512 1 : // LSM. Some level iterators may omit keys that are known to fall outside
513 1 : // the seek prefix (eg, due to sstable bloom filter exclusion). It's
514 1 : // important that in such cases we don't position any iterators beyond
515 1 : // m.prefix, because doing so may interfere with future seeks.
516 1 : //
517 1 : // Let prefixes P1 < P2 < P3. Imagine a SeekPrefixGE to prefix P1, followed
518 1 : // by a SeekPrefixGE to prefix P2. Imagine there exist live keys at prefix
519 1 : // P2, but they're not visible to the SeekPrefixGE(P1) (because of
520 1 : // bloom-filter exclusion or a range tombstone that deletes prefix P1 but
521 1 : // not P2). If the SeekPrefixGE(P1) is allowed to move any level iterators
522 1 : // to P3, the SeekPrefixGE(P2, TrySeekUsingNext=true) may mistakenly think
523 1 : // the level contains no point keys or range tombstones within the prefix
524 1 : // P2. Care is taken to avoid ever advancing the iterator beyond the current
525 1 : // prefix. If nextEntry is ever invoked while we're already beyond the
526 1 : // current prefix, we're violating the invariant.
527 1 : if invariants.Enabled && m.prefix != nil {
528 1 : if p := m.split.Prefix(l.iterKV.K.UserKey); !bytes.Equal(m.prefix, p) {
529 0 : m.logger.Fatalf("mergingIter: prefix violation: nexting beyond prefix %q; existing heap root %q\n%s",
530 0 : m.prefix, l.iterKV, debug.Stack())
531 0 : }
532 : }
533 :
534 1 : oldTopLevel := l.index
535 1 : oldRangeDelIterGeneration := l.rangeDelIterGeneration
536 1 :
537 1 : if succKey == nil {
538 1 : l.iterKV = l.iter.Next()
539 1 : } else {
540 1 : l.iterKV = l.iter.NextPrefix(succKey)
541 1 : }
542 :
543 1 : if l.iterKV == nil {
544 1 : if err := l.iter.Error(); err != nil {
545 1 : return err
546 1 : }
547 1 : m.heap.pop()
548 1 : } else {
549 1 : if m.prefix != nil && !bytes.Equal(m.prefix, m.split.Prefix(l.iterKV.K.UserKey)) {
550 1 : // Set keys without a matching prefix to their zero values when in prefix
551 1 : // iteration mode and remove iterated level from heap.
552 1 : l.iterKV = nil
553 1 : m.heap.pop()
554 1 : } else if m.heap.len() > 1 {
555 1 : m.heap.fix(0)
556 1 : }
557 1 : if l.rangeDelIterGeneration != oldRangeDelIterGeneration {
558 1 : // The rangeDelIter changed which indicates that the l.iter moved to the
559 1 : // next sstable. We have to update the tombstone for oldTopLevel as well.
560 1 : oldTopLevel--
561 1 : }
562 : }
563 :
564 : // The cached tombstones are only valid for the levels
565 : // [0,oldTopLevel]. Updated the cached tombstones for any levels in the range
566 : // [oldTopLevel+1,heap[0].index].
567 1 : return m.initMinRangeDelIters(oldTopLevel)
568 : }
569 :
570 : // isNextEntryDeleted starts from the current entry (as the next entry) and if
571 : // it is deleted, moves the iterators forward as needed and returns true, else
572 : // it returns false. item is the top item in the heap. If any of the required
573 : // iterator operations error, the error is returned without updating m.err.
574 : //
575 : // During prefix iteration mode, isNextEntryDeleted will exhaust the iterator by
576 : // clearing the heap if the deleted key(s) extend beyond the iteration prefix
577 : // during prefix-iteration mode.
578 1 : func (m *mergingIter) isNextEntryDeleted(item *mergingIterLevel) (bool, error) {
579 1 : // Look for a range deletion tombstone containing item.iterKV at higher
580 1 : // levels (level < item.index). If we find such a range tombstone we know
581 1 : // it deletes the key in the current level. Also look for a range
582 1 : // deletion at the current level (level == item.index). If we find such a
583 1 : // range deletion we need to check whether it is newer than the current
584 1 : // entry.
585 1 : for level := 0; level <= item.index; level++ {
586 1 : l := &m.levels[level]
587 1 : if l.rangeDelIter == nil || l.tombstone == nil {
588 1 : // If l.tombstone is nil, there are no further tombstones
589 1 : // in the current sstable in the current (forward) iteration
590 1 : // direction.
591 1 : continue
592 : }
593 1 : if m.heap.cmp(l.tombstone.End, item.iterKV.K.UserKey) <= 0 {
594 1 : // The current key is at or past the tombstone end key.
595 1 : //
596 1 : // NB: for the case that this l.rangeDelIter is provided by a levelIter we know that
597 1 : // the levelIter must be positioned at a key >= item.iterKV. So it is sufficient to seek the
598 1 : // current l.rangeDelIter (since any range del iterators that will be provided by the
599 1 : // levelIter in the future cannot contain item.iterKV). Also, it is possible that we
600 1 : // will encounter parts of the range delete that should be ignored -- we handle that
601 1 : // below.
602 1 : var err error
603 1 : l.tombstone, err = l.rangeDelIter.SeekGE(item.iterKV.K.UserKey)
604 1 : if err != nil {
605 1 : return false, err
606 1 : }
607 : }
608 1 : if l.tombstone == nil {
609 1 : continue
610 : }
611 :
612 1 : if l.tombstone.VisibleAt(m.snapshot) && m.heap.cmp(l.tombstone.Start, item.iterKV.K.UserKey) <= 0 {
613 1 : if level < item.index {
614 1 : // We could also do m.seekGE(..., level + 1). The levels from
615 1 : // [level + 1, item.index) are already after item.iterKV so seeking them may be
616 1 : // wasteful.
617 1 :
618 1 : // We can seek up to tombstone.End.
619 1 : //
620 1 : // Progress argument: Since this file is at a higher level than item.iterKV we know
621 1 : // that the iterator in this file must be positioned within its bounds and at a key
622 1 : // X > item.iterKV (otherwise it would be the min of the heap). It is not
623 1 : // possible for X.UserKey == item.iterKV.UserKey, since it is incompatible with
624 1 : // X > item.iterKV (a lower version cannot be in a higher sstable), so it must be that
625 1 : // X.UserKey > item.iterKV.UserKey. Which means l.largestUserKey > item.key.UserKey.
626 1 : // We also know that l.tombstone.End > item.iterKV.UserKey. So the min of these,
627 1 : // seekKey, computed below, is > item.iterKV.UserKey, so the call to seekGE() will
628 1 : // make forward progress.
629 1 : m.seekKeyBuf = append(m.seekKeyBuf[:0], l.tombstone.End...)
630 1 : seekKey := m.seekKeyBuf
631 1 : // This seek is not directly due to a SeekGE call, so we don't know
632 1 : // enough about the underlying iterator positions, and so we keep the
633 1 : // try-seek-using-next optimization disabled. Additionally, if we're in
634 1 : // prefix-seek mode and a re-seek would have moved us past the original
635 1 : // prefix, we can remove all merging iter levels below the rangedel
636 1 : // tombstone's level and return immediately instead of re-seeking. This
637 1 : // is correct since those levels cannot provide a key that matches the
638 1 : // prefix, and is also visible. Additionally, this is important to make
639 1 : // subsequent `TrySeekUsingNext` work correctly, as a re-seek on a
640 1 : // different prefix could have resulted in this iterator skipping visible
641 1 : // keys at prefixes in between m.prefix and seekKey, that are currently
642 1 : // not in the heap due to a bloom filter mismatch.
643 1 : //
644 1 : // Additionally, we set the relative-seek flag. This is
645 1 : // important when iterating with lazy combined iteration. If
646 1 : // there's a range key between this level's current file and the
647 1 : // file the seek will land on, we need to detect it in order to
648 1 : // trigger construction of the combined iterator.
649 1 : if m.prefix != nil {
650 1 : if !bytes.Equal(m.prefix, m.split.Prefix(seekKey)) {
651 1 : for i := item.index; i < len(m.levels); i++ {
652 1 : // Remove this level from the heap. Setting iterKV
653 1 : // to nil should be sufficient for initMinHeap to
654 1 : // not re-initialize the heap with them in it. Other
655 1 : // fields in mergingIterLevel can remain as-is; the
656 1 : // iter/rangeDelIter needs to stay intact for future
657 1 : // trySeekUsingNexts to work, the level iter
658 1 : // boundary context is owned by the levelIter which
659 1 : // is not being repositioned, and any tombstones in
660 1 : // these levels will be irrelevant for us anyway.
661 1 : m.levels[i].iterKV = nil
662 1 : }
663 : // TODO(bilal): Consider a more efficient way of removing levels from
664 : // the heap without reinitializing all of it. This would likely
665 : // necessitate tracking the heap positions of each mergingIterHeap
666 : // item in the mergingIterLevel, and then swapping that item in the
667 : // heap with the last-positioned heap item, and shrinking the heap by
668 : // one.
669 1 : if err := m.initMinHeap(); err != nil {
670 0 : return false, err
671 0 : }
672 1 : return true, nil
673 : }
674 : }
675 1 : if err := m.seekGE(seekKey, item.index, base.SeekGEFlagsNone.EnableRelativeSeek()); err != nil {
676 0 : return false, err
677 0 : }
678 1 : return true, nil
679 : }
680 1 : if l.tombstone.CoversAt(m.snapshot, item.iterKV.SeqNum()) {
681 1 : if err := m.nextEntry(item, nil /* succKey */); err != nil {
682 0 : return false, err
683 0 : }
684 1 : return true, nil
685 : }
686 : }
687 : }
688 1 : return false, nil
689 : }
690 :
691 : // Starting from the current entry, finds the first (next) entry that can be returned.
692 : //
693 : // If an error occurs, m.err is updated to hold the error and findNextentry
694 : // returns a nil internal key.
695 1 : func (m *mergingIter) findNextEntry() *base.InternalKV {
696 1 : for m.heap.len() > 0 && m.err == nil {
697 1 : item := m.heap.items[0]
698 1 :
699 1 : // The levelIter internal iterator will interleave exclusive sentinel
700 1 : // keys to keep files open until their range deletions are no longer
701 1 : // necessary. Sometimes these are interleaved with the user key of a
702 1 : // file's largest key, in which case they may simply be stepped over to
703 1 : // move to the next file in the forward direction. Other times they're
704 1 : // interleaved at the user key of the user-iteration boundary, if that
705 1 : // falls within the bounds of a file. In the latter case, there are no
706 1 : // more keys < m.upper, and we can stop iterating.
707 1 : //
708 1 : // We perform a key comparison to differentiate between these two cases.
709 1 : // This key comparison is considered okay because it only happens for
710 1 : // sentinel keys. It may be eliminated after #2863.
711 1 : if m.levels[item.index].iterKV.K.IsExclusiveSentinel() {
712 1 : if m.upper != nil && m.heap.cmp(m.levels[item.index].iterKV.K.UserKey, m.upper) >= 0 {
713 1 : break
714 : }
715 : // This key is the largest boundary of a file and can be skipped now
716 : // that the file's range deletions are no longer relevant.
717 1 : m.err = m.nextEntry(item, nil /* succKey */)
718 1 : if m.err != nil {
719 0 : return nil
720 0 : }
721 1 : continue
722 : }
723 :
724 1 : m.addItemStats(item)
725 1 :
726 1 : // Check if the heap root key is deleted by a range tombstone in a
727 1 : // higher level. If it is, isNextEntryDeleted will advance the iterator
728 1 : // to a later key (through seeking or nexting).
729 1 : isDeleted, err := m.isNextEntryDeleted(item)
730 1 : if err != nil {
731 1 : m.err = err
732 1 : return nil
733 1 : } else if isDeleted {
734 1 : m.stats.PointsCoveredByRangeTombstones++
735 1 : continue
736 : }
737 :
738 : // Check if the key is visible at the iterator sequence numbers.
739 1 : if !item.iterKV.Visible(m.snapshot, m.batchSnapshot) {
740 1 : m.err = m.nextEntry(item, nil /* succKey */)
741 1 : if m.err != nil {
742 0 : return nil
743 0 : }
744 1 : continue
745 : }
746 :
747 : // The heap root is visible and not deleted by any range tombstones.
748 : // Return it.
749 1 : return item.iterKV
750 : }
751 1 : return nil
752 : }
753 :
754 : // Steps to the prev entry. item is the current top item in the heap.
755 1 : func (m *mergingIter) prevEntry(l *mergingIterLevel) error {
756 1 : oldTopLevel := l.index
757 1 : oldRangeDelIterGeneration := l.rangeDelIterGeneration
758 1 : if l.iterKV = l.iter.Prev(); l.iterKV != nil {
759 1 : if m.heap.len() > 1 {
760 1 : m.heap.fix(0)
761 1 : }
762 1 : if l.rangeDelIterGeneration != oldRangeDelIterGeneration && l.rangeDelIter != nil {
763 1 : // The rangeDelIter changed which indicates that the l.iter moved to the
764 1 : // previous sstable. We have to update the tombstone for oldTopLevel as
765 1 : // well.
766 1 : oldTopLevel--
767 1 : }
768 1 : } else {
769 1 : if err := l.iter.Error(); err != nil {
770 1 : return err
771 1 : }
772 1 : m.heap.pop()
773 : }
774 :
775 : // The cached tombstones are only valid for the levels
776 : // [0,oldTopLevel]. Updated the cached tombstones for any levels in the range
777 : // [oldTopLevel+1,heap[0].index].
778 1 : return m.initMaxRangeDelIters(oldTopLevel)
779 : }
780 :
781 : // isPrevEntryDeleted() starts from the current entry (as the prev entry) and if it is deleted,
782 : // moves the iterators backward as needed and returns true, else it returns false. item is the top
783 : // item in the heap.
784 1 : func (m *mergingIter) isPrevEntryDeleted(item *mergingIterLevel) (bool, error) {
785 1 : // Look for a range deletion tombstone containing item.iterKV at higher
786 1 : // levels (level < item.index). If we find such a range tombstone we know
787 1 : // it deletes the key in the current level. Also look for a range
788 1 : // deletion at the current level (level == item.index). If we find such a
789 1 : // range deletion we need to check whether it is newer than the current
790 1 : // entry.
791 1 : for level := 0; level <= item.index; level++ {
792 1 : l := &m.levels[level]
793 1 : if l.rangeDelIter == nil || l.tombstone == nil {
794 1 : // If l.tombstone is nil, there are no further tombstones
795 1 : // in the current sstable in the current (reverse) iteration
796 1 : // direction.
797 1 : continue
798 : }
799 1 : if m.heap.cmp(item.iterKV.K.UserKey, l.tombstone.Start) < 0 {
800 1 : // The current key is before the tombstone start key.
801 1 : //
802 1 : // NB: for the case that this l.rangeDelIter is provided by a levelIter we know that
803 1 : // the levelIter must be positioned at a key < item.iterKV. So it is sufficient to seek the
804 1 : // current l.rangeDelIter (since any range del iterators that will be provided by the
805 1 : // levelIter in the future cannot contain item.iterKV). Also, it is it is possible that we
806 1 : // will encounter parts of the range delete that should be ignored -- we handle that
807 1 : // below.
808 1 :
809 1 : tomb, err := keyspan.SeekLE(m.heap.cmp, l.rangeDelIter, item.iterKV.K.UserKey)
810 1 : if err != nil {
811 0 : return false, err
812 0 : }
813 1 : l.tombstone = tomb
814 : }
815 1 : if l.tombstone == nil {
816 1 : continue
817 : }
818 1 : if l.tombstone.VisibleAt(m.snapshot) && m.heap.cmp(l.tombstone.End, item.iterKV.K.UserKey) > 0 {
819 1 : if level < item.index {
820 1 : // We could also do m.seekLT(..., level + 1). The levels from
821 1 : // [level + 1, item.index) are already before item.iterKV so seeking them may be
822 1 : // wasteful.
823 1 :
824 1 : // We can seek up to tombstone.Start.UserKey.
825 1 : //
826 1 : // Progress argument: We know that the iterator in this file is positioned within
827 1 : // its bounds and at a key X < item.iterKV (otherwise it would be the max of the heap).
828 1 : // So smallestUserKey <= item.iterKV.UserKey and we already know that
829 1 : // l.tombstone.Start.UserKey <= item.iterKV.UserKey. So the seekKey computed below
830 1 : // is <= item.iterKV.UserKey, and since we do a seekLT() we will make backwards
831 1 : // progress.
832 1 : m.seekKeyBuf = append(m.seekKeyBuf[:0], l.tombstone.Start...)
833 1 : seekKey := m.seekKeyBuf
834 1 : // We set the relative-seek flag. This is important when
835 1 : // iterating with lazy combined iteration. If there's a range
836 1 : // key between this level's current file and the file the seek
837 1 : // will land on, we need to detect it in order to trigger
838 1 : // construction of the combined iterator.
839 1 : if err := m.seekLT(seekKey, item.index, base.SeekLTFlagsNone.EnableRelativeSeek()); err != nil {
840 0 : return false, err
841 0 : }
842 1 : return true, nil
843 : }
844 1 : if l.tombstone.CoversAt(m.snapshot, item.iterKV.SeqNum()) {
845 1 : if err := m.prevEntry(item); err != nil {
846 0 : return false, err
847 0 : }
848 1 : return true, nil
849 : }
850 : }
851 : }
852 1 : return false, nil
853 : }
854 :
855 : // Starting from the current entry, finds the first (prev) entry that can be returned.
856 : //
857 : // If an error occurs, m.err is updated to hold the error and findNextentry
858 : // returns a nil internal key.
859 1 : func (m *mergingIter) findPrevEntry() *base.InternalKV {
860 1 : for m.heap.len() > 0 && m.err == nil {
861 1 : item := m.heap.items[0]
862 1 :
863 1 : // The levelIter internal iterator will interleave exclusive sentinel
864 1 : // keys to keep files open until their range deletions are no longer
865 1 : // necessary. Sometimes these are interleaved with the user key of a
866 1 : // file's smallest key, in which case they may simply be stepped over to
867 1 : // move to the next file in the backward direction. Other times they're
868 1 : // interleaved at the user key of the user-iteration boundary, if that
869 1 : // falls within the bounds of a file. In the latter case, there are no
870 1 : // more keys ≥ m.lower, and we can stop iterating.
871 1 : //
872 1 : // We perform a key comparison to differentiate between these two cases.
873 1 : // This key comparison is considered okay because it only happens for
874 1 : // sentinel keys. It may be eliminated after #2863.
875 1 : if m.levels[item.index].iterKV.K.IsExclusiveSentinel() {
876 1 : if m.lower != nil && m.heap.cmp(m.levels[item.index].iterKV.K.UserKey, m.lower) <= 0 {
877 1 : break
878 : }
879 : // This key is the smallest boundary of a file and can be skipped
880 : // now that the file's range deletions are no longer relevant.
881 1 : m.err = m.prevEntry(item)
882 1 : if m.err != nil {
883 0 : return nil
884 0 : }
885 1 : continue
886 : }
887 :
888 1 : m.addItemStats(item)
889 1 : if isDeleted, err := m.isPrevEntryDeleted(item); err != nil {
890 0 : m.err = err
891 0 : return nil
892 1 : } else if isDeleted {
893 1 : m.stats.PointsCoveredByRangeTombstones++
894 1 : continue
895 : }
896 1 : if item.iterKV.Visible(m.snapshot, m.batchSnapshot) {
897 1 : return item.iterKV
898 1 : }
899 1 : m.err = m.prevEntry(item)
900 : }
901 1 : return nil
902 : }
903 :
904 : // Seeks levels >= level to >= key. Additionally uses range tombstones to extend the seeks.
905 : //
906 : // If an error occurs, seekGE returns the error without setting m.err.
907 1 : func (m *mergingIter) seekGE(key []byte, level int, flags base.SeekGEFlags) error {
908 1 : // When seeking, we can use tombstones to adjust the key we seek to on each
909 1 : // level. Consider the series of range tombstones:
910 1 : //
911 1 : // 1: a---e
912 1 : // 2: d---h
913 1 : // 3: g---k
914 1 : // 4: j---n
915 1 : // 5: m---q
916 1 : //
917 1 : // If we SeekGE("b") we also find the tombstone "b" resides within in the
918 1 : // first level which is [a,e). Regardless of whether this tombstone deletes
919 1 : // "b" in that level, we know it deletes "b" in all lower levels, so we
920 1 : // adjust the search key in the next level to the tombstone end key "e". We
921 1 : // then SeekGE("e") in the second level and find the corresponding tombstone
922 1 : // [d,h). This process continues and we end up seeking for "h" in the 3rd
923 1 : // level, "k" in the 4th level and "n" in the last level.
924 1 : //
925 1 : // TODO(peter,rangedel): In addition to the above we can delay seeking a
926 1 : // level (and any lower levels) when the current iterator position is
927 1 : // contained within a range tombstone at a higher level.
928 1 :
929 1 : // Deterministically disable the TrySeekUsingNext optimizations sometimes in
930 1 : // invariant builds to encourage the metamorphic tests to surface bugs. Note
931 1 : // that we cannot disable the optimization within individual levels. It must
932 1 : // be disabled for all levels or none. If one lower-level iterator performs
933 1 : // a fresh seek whereas another takes advantage of its current iterator
934 1 : // position, the heap can become inconsistent. Consider the following
935 1 : // example:
936 1 : //
937 1 : // L5: [ [b-c) ] [ d ]*
938 1 : // L6: [ b ] [e]*
939 1 : //
940 1 : // Imagine a SeekGE(a). The [b-c) range tombstone deletes the L6 point key
941 1 : // 'b', resulting in the iterator positioned at d with the heap:
942 1 : //
943 1 : // {L5: d, L6: e}
944 1 : //
945 1 : // A subsequent SeekGE(b) is seeking to a larger key, so the caller may set
946 1 : // TrySeekUsingNext()=true. If the L5 iterator used the TrySeekUsingNext
947 1 : // optimization but the L6 iterator did not, the iterator would have the
948 1 : // heap:
949 1 : //
950 1 : // {L6: b, L5: d}
951 1 : //
952 1 : // Because the L5 iterator has already advanced to the next sstable, the
953 1 : // merging iterator cannot observe the [b-c) range tombstone and will
954 1 : // mistakenly return L6's deleted point key 'b'.
955 1 : if testingDisableSeekOpt(key, uintptr(unsafe.Pointer(m))) && !m.forceEnableSeekOpt {
956 1 : flags = flags.DisableTrySeekUsingNext()
957 1 : }
958 :
959 1 : for ; level < len(m.levels); level++ {
960 1 : if invariants.Enabled && m.lower != nil && m.heap.cmp(key, m.lower) < 0 {
961 0 : m.logger.Fatalf("mergingIter: lower bound violation: %s < %s\n%s", key, m.lower, debug.Stack())
962 0 : }
963 :
964 1 : l := &m.levels[level]
965 1 : if m.prefix != nil {
966 1 : l.iterKV = l.iter.SeekPrefixGE(m.prefix, key, flags)
967 1 : if l.iterKV != nil {
968 1 : if !bytes.Equal(m.prefix, m.split.Prefix(l.iterKV.K.UserKey)) {
969 1 : // Prevent keys without a matching prefix from being added to the heap by setting
970 1 : // iterKey and iterValue to their zero values before calling initMinHeap.
971 1 : l.iterKV = nil
972 1 : }
973 : }
974 1 : } else {
975 1 : l.iterKV = l.iter.SeekGE(key, flags)
976 1 : }
977 1 : if l.iterKV == nil {
978 1 : if err := l.iter.Error(); err != nil {
979 1 : return err
980 1 : }
981 : }
982 :
983 : // If this level contains overlapping range tombstones, alter the seek
984 : // key accordingly. Caveat: If we're performing lazy-combined iteration,
985 : // we cannot alter the seek key: Range tombstones don't delete range
986 : // keys, and there might exist live range keys within the range
987 : // tombstone's span that need to be observed to trigger a switch to
988 : // combined iteration.
989 1 : if rangeDelIter := l.rangeDelIter; rangeDelIter != nil &&
990 1 : (m.combinedIterState == nil || m.combinedIterState.initialized) {
991 1 : // The level has a range-del iterator. Find the tombstone containing
992 1 : // the search key.
993 1 : var err error
994 1 : l.tombstone, err = rangeDelIter.SeekGE(key)
995 1 : if err != nil {
996 0 : return err
997 0 : }
998 1 : if l.tombstone != nil && l.tombstone.VisibleAt(m.snapshot) && m.heap.cmp(l.tombstone.Start, key) <= 0 {
999 1 : // Based on the containment condition tombstone.End > key, so
1000 1 : // the assignment to key results in a monotonically
1001 1 : // non-decreasing key across iterations of this loop.
1002 1 : //
1003 1 : // The adjustment of key here can only move it to a larger key.
1004 1 : // Since the caller of seekGE guaranteed that the original key
1005 1 : // was greater than or equal to m.lower, the new key will
1006 1 : // continue to be greater than or equal to m.lower.
1007 1 : key = l.tombstone.End
1008 1 : }
1009 : }
1010 : }
1011 1 : return m.initMinHeap()
1012 : }
1013 :
1014 0 : func (m *mergingIter) String() string {
1015 0 : return "merging"
1016 0 : }
1017 :
1018 : // SeekGE implements base.InternalIterator.SeekGE. Note that SeekGE only checks
1019 : // the upper bound. It is up to the caller to ensure that key is greater than
1020 : // or equal to the lower bound.
1021 1 : func (m *mergingIter) SeekGE(key []byte, flags base.SeekGEFlags) *base.InternalKV {
1022 1 : m.prefix = nil
1023 1 : m.err = m.seekGE(key, 0 /* start level */, flags)
1024 1 : if m.err != nil {
1025 1 : return nil
1026 1 : }
1027 1 : return m.findNextEntry()
1028 : }
1029 :
1030 : // SeekPrefixGE implements base.InternalIterator.SeekPrefixGE.
1031 1 : func (m *mergingIter) SeekPrefixGE(prefix, key []byte, flags base.SeekGEFlags) *base.InternalKV {
1032 1 : return m.SeekPrefixGEStrict(prefix, key, flags)
1033 1 : }
1034 :
1035 : // SeekPrefixGEStrict implements topLevelIterator.SeekPrefixGEStrict. Note that
1036 : // SeekPrefixGEStrict explicitly checks that the key has a matching prefix.
1037 : func (m *mergingIter) SeekPrefixGEStrict(
1038 : prefix, key []byte, flags base.SeekGEFlags,
1039 1 : ) *base.InternalKV {
1040 1 : m.prefix = prefix
1041 1 : m.err = m.seekGE(key, 0 /* start level */, flags)
1042 1 : if m.err != nil {
1043 1 : return nil
1044 1 : }
1045 :
1046 1 : iterKV := m.findNextEntry()
1047 1 : if invariants.Enabled && iterKV != nil {
1048 1 : if !bytes.Equal(m.prefix, m.split.Prefix(iterKV.K.UserKey)) {
1049 0 : m.logger.Fatalf("mergingIter: prefix violation: returning key %q without prefix %q\n", iterKV, m.prefix)
1050 0 : }
1051 : }
1052 1 : return iterKV
1053 : }
1054 :
1055 : // Seeks levels >= level to < key. Additionally uses range tombstones to extend the seeks.
1056 1 : func (m *mergingIter) seekLT(key []byte, level int, flags base.SeekLTFlags) error {
1057 1 : // See the comment in seekGE regarding using tombstones to adjust the seek
1058 1 : // target per level.
1059 1 : m.prefix = nil
1060 1 : for ; level < len(m.levels); level++ {
1061 1 : if invariants.Enabled && m.upper != nil && m.heap.cmp(key, m.upper) > 0 {
1062 0 : m.logger.Fatalf("mergingIter: upper bound violation: %s > %s\n%s", key, m.upper, debug.Stack())
1063 0 : }
1064 :
1065 1 : l := &m.levels[level]
1066 1 : l.iterKV = l.iter.SeekLT(key, flags)
1067 1 : if l.iterKV == nil {
1068 1 : if err := l.iter.Error(); err != nil {
1069 1 : return err
1070 1 : }
1071 : }
1072 :
1073 : // If this level contains overlapping range tombstones, alter the seek
1074 : // key accordingly. Caveat: If we're performing lazy-combined iteration,
1075 : // we cannot alter the seek key: Range tombstones don't delete range
1076 : // keys, and there might exist live range keys within the range
1077 : // tombstone's span that need to be observed to trigger a switch to
1078 : // combined iteration.
1079 1 : if rangeDelIter := l.rangeDelIter; rangeDelIter != nil &&
1080 1 : (m.combinedIterState == nil || m.combinedIterState.initialized) {
1081 1 : // The level has a range-del iterator. Find the tombstone containing
1082 1 : // the search key.
1083 1 : tomb, err := keyspan.SeekLE(m.heap.cmp, rangeDelIter, key)
1084 1 : if err != nil {
1085 0 : return err
1086 0 : }
1087 1 : l.tombstone = tomb
1088 1 : // Since SeekLT is exclusive on `key` and a tombstone's end key is
1089 1 : // also exclusive, a seek key equal to a tombstone's end key still
1090 1 : // enables the seek optimization (Note this is different than the
1091 1 : // check performed by (*keyspan.Span).Contains).
1092 1 : if l.tombstone != nil && l.tombstone.VisibleAt(m.snapshot) &&
1093 1 : m.heap.cmp(key, l.tombstone.End) <= 0 {
1094 1 : // NB: Based on the containment condition
1095 1 : // tombstone.Start.UserKey <= key, so the assignment to key
1096 1 : // results in a monotonically non-increasing key across
1097 1 : // iterations of this loop.
1098 1 : //
1099 1 : // The adjustment of key here can only move it to a smaller key.
1100 1 : // Since the caller of seekLT guaranteed that the original key
1101 1 : // was less than or equal to m.upper, the new key will continue
1102 1 : // to be less than or equal to m.upper.
1103 1 : key = l.tombstone.Start
1104 1 : }
1105 : }
1106 : }
1107 :
1108 1 : return m.initMaxHeap()
1109 : }
1110 :
1111 : // SeekLT implements base.InternalIterator.SeekLT. Note that SeekLT only checks
1112 : // the lower bound. It is up to the caller to ensure that key is less than the
1113 : // upper bound.
1114 1 : func (m *mergingIter) SeekLT(key []byte, flags base.SeekLTFlags) *base.InternalKV {
1115 1 : m.prefix = nil
1116 1 : m.err = m.seekLT(key, 0 /* start level */, flags)
1117 1 : if m.err != nil {
1118 1 : return nil
1119 1 : }
1120 1 : return m.findPrevEntry()
1121 : }
1122 :
1123 : // First implements base.InternalIterator.First. Note that First only checks
1124 : // the upper bound. It is up to the caller to ensure that key is greater than
1125 : // or equal to the lower bound (e.g. via a call to SeekGE(lower)).
1126 1 : func (m *mergingIter) First() *base.InternalKV {
1127 1 : m.err = nil // clear cached iteration error
1128 1 : m.prefix = nil
1129 1 : m.heap.items = m.heap.items[:0]
1130 1 : for i := range m.levels {
1131 1 : l := &m.levels[i]
1132 1 : l.iterKV = l.iter.First()
1133 1 : if l.iterKV == nil {
1134 1 : if m.err = l.iter.Error(); m.err != nil {
1135 1 : return nil
1136 1 : }
1137 : }
1138 : }
1139 1 : if m.err = m.initMinHeap(); m.err != nil {
1140 0 : return nil
1141 0 : }
1142 1 : return m.findNextEntry()
1143 : }
1144 :
1145 : // Last implements base.InternalIterator.Last. Note that Last only checks the
1146 : // lower bound. It is up to the caller to ensure that key is less than the
1147 : // upper bound (e.g. via a call to SeekLT(upper))
1148 1 : func (m *mergingIter) Last() *base.InternalKV {
1149 1 : m.err = nil // clear cached iteration error
1150 1 : m.prefix = nil
1151 1 : for i := range m.levels {
1152 1 : l := &m.levels[i]
1153 1 : l.iterKV = l.iter.Last()
1154 1 : if l.iterKV == nil {
1155 1 : if m.err = l.iter.Error(); m.err != nil {
1156 1 : return nil
1157 1 : }
1158 : }
1159 : }
1160 1 : if m.err = m.initMaxHeap(); m.err != nil {
1161 0 : return nil
1162 0 : }
1163 1 : return m.findPrevEntry()
1164 : }
1165 :
1166 1 : func (m *mergingIter) Next() *base.InternalKV {
1167 1 : if m.err != nil {
1168 1 : return nil
1169 1 : }
1170 :
1171 1 : if m.dir != 1 {
1172 1 : if m.err = m.switchToMinHeap(); m.err != nil {
1173 1 : return nil
1174 1 : }
1175 1 : return m.findNextEntry()
1176 : }
1177 :
1178 1 : if m.heap.len() == 0 {
1179 1 : return nil
1180 1 : }
1181 :
1182 : // NB: It's okay to call nextEntry directly even during prefix iteration
1183 : // mode. During prefix iteration mode, we rely on the caller to not call
1184 : // Next if the iterator has already advanced beyond the iteration prefix.
1185 : // See the comment above the base.InternalIterator interface.
1186 1 : if m.err = m.nextEntry(m.heap.items[0], nil /* succKey */); m.err != nil {
1187 1 : return nil
1188 1 : }
1189 :
1190 1 : iterKV := m.findNextEntry()
1191 1 : if invariants.Enabled && m.prefix != nil && iterKV != nil {
1192 1 : if !bytes.Equal(m.prefix, m.split.Prefix(iterKV.K.UserKey)) {
1193 0 : m.logger.Fatalf("mergingIter: prefix violation: returning key %q without prefix %q\n", iterKV, m.prefix)
1194 0 : }
1195 : }
1196 1 : return iterKV
1197 : }
1198 :
1199 1 : func (m *mergingIter) NextPrefix(succKey []byte) *base.InternalKV {
1200 1 : if m.dir != 1 {
1201 0 : panic("pebble: cannot switch directions with NextPrefix")
1202 : }
1203 1 : if m.err != nil || m.heap.len() == 0 {
1204 0 : return nil
1205 0 : }
1206 1 : if m.levelsPositioned == nil {
1207 1 : m.levelsPositioned = make([]bool, len(m.levels))
1208 1 : } else {
1209 1 : for i := range m.levelsPositioned {
1210 1 : m.levelsPositioned[i] = false
1211 1 : }
1212 : }
1213 :
1214 : // The heap root necessarily must be positioned at a key < succKey, because
1215 : // NextPrefix was invoked.
1216 1 : root := m.heap.items[0]
1217 1 : if invariants.Enabled && m.heap.cmp((*root).iterKV.K.UserKey, succKey) >= 0 {
1218 0 : m.logger.Fatalf("pebble: invariant violation: NextPrefix(%q) called on merging iterator already positioned at %q",
1219 0 : succKey, (*root).iterKV)
1220 0 : }
1221 : // NB: root is the heap root before we call nextEntry; nextEntry may change
1222 : // the heap root, so we must not `root` to still be the root of the heap, or
1223 : // even to be in the heap if the level's iterator becomes exhausted.
1224 1 : if m.err = m.nextEntry(root, succKey); m.err != nil {
1225 1 : return nil
1226 1 : }
1227 : // We only consider the level to be conclusively positioned at the next
1228 : // prefix if our call to nextEntry did not advance the level onto a range
1229 : // deletion's boundary. Range deletions may have bounds within the prefix
1230 : // that are still surfaced by NextPrefix.
1231 1 : m.levelsPositioned[root.index] = root.iterKV == nil || !root.iterKV.K.IsExclusiveSentinel()
1232 1 :
1233 1 : for m.heap.len() > 0 {
1234 1 : root := m.heap.items[0]
1235 1 : if m.levelsPositioned[root.index] {
1236 1 : // A level we've previously positioned is at the top of the heap, so
1237 1 : // there are no other levels positioned at keys < succKey. We've
1238 1 : // advanced as far as we need to.
1239 1 : break
1240 : }
1241 : // If the current heap root is a sentinel key, we need to skip it.
1242 : // Calling NextPrefix while positioned at a sentinel key is not
1243 : // supported.
1244 1 : if root.iterKV.K.IsExclusiveSentinel() {
1245 1 : if m.err = m.nextEntry(root, nil); m.err != nil {
1246 0 : return nil
1247 0 : }
1248 1 : continue
1249 : }
1250 :
1251 : // Since this level was not the original heap root when NextPrefix was
1252 : // called, we don't know whether this level's current key has the
1253 : // previous prefix or a new one.
1254 1 : if m.heap.cmp(root.iterKV.K.UserKey, succKey) >= 0 {
1255 1 : break
1256 : }
1257 1 : if m.err = m.nextEntry(root, succKey); m.err != nil {
1258 0 : return nil
1259 0 : }
1260 : // We only consider the level to be conclusively positioned at the next
1261 : // prefix if our call to nextEntry did not land onto a range deletion's
1262 : // boundary. Range deletions may have bounds within the prefix that are
1263 : // still surfaced by NextPrefix.
1264 1 : m.levelsPositioned[root.index] = root.iterKV == nil || !root.iterKV.K.IsExclusiveSentinel()
1265 : }
1266 1 : return m.findNextEntry()
1267 : }
1268 :
1269 1 : func (m *mergingIter) Prev() *base.InternalKV {
1270 1 : if m.err != nil {
1271 0 : return nil
1272 0 : }
1273 :
1274 1 : if m.dir != -1 {
1275 1 : if m.prefix != nil {
1276 1 : m.err = errors.New("pebble: unsupported reverse prefix iteration")
1277 1 : return nil
1278 1 : }
1279 1 : if m.err = m.switchToMaxHeap(); m.err != nil {
1280 1 : return nil
1281 1 : }
1282 1 : return m.findPrevEntry()
1283 : }
1284 :
1285 1 : if m.heap.len() == 0 {
1286 1 : return nil
1287 1 : }
1288 1 : if m.err = m.prevEntry(m.heap.items[0]); m.err != nil {
1289 1 : return nil
1290 1 : }
1291 1 : return m.findPrevEntry()
1292 : }
1293 :
1294 1 : func (m *mergingIter) Error() error {
1295 1 : if m.heap.len() == 0 || m.err != nil {
1296 1 : return m.err
1297 1 : }
1298 1 : return m.levels[m.heap.items[0].index].iter.Error()
1299 : }
1300 :
1301 1 : func (m *mergingIter) Close() error {
1302 1 : for i := range m.levels {
1303 1 : iter := m.levels[i].iter
1304 1 : if err := iter.Close(); err != nil && m.err == nil {
1305 0 : m.err = err
1306 0 : }
1307 1 : m.levels[i].setRangeDelIter(nil)
1308 : }
1309 1 : m.levels = nil
1310 1 : m.heap.items = m.heap.items[:0]
1311 1 : return m.err
1312 : }
1313 :
1314 1 : func (m *mergingIter) SetBounds(lower, upper []byte) {
1315 1 : m.prefix = nil
1316 1 : m.lower = lower
1317 1 : m.upper = upper
1318 1 : for i := range m.levels {
1319 1 : m.levels[i].iter.SetBounds(lower, upper)
1320 1 : }
1321 1 : m.heap.clear()
1322 : }
1323 :
1324 1 : func (m *mergingIter) SetContext(ctx context.Context) {
1325 1 : for i := range m.levels {
1326 1 : m.levels[i].iter.SetContext(ctx)
1327 1 : }
1328 : }
1329 :
1330 : // DebugTree is part of the InternalIterator interface.
1331 0 : func (m *mergingIter) DebugTree(tp treeprinter.Node) {
1332 0 : n := tp.Childf("%T(%p)", m, m)
1333 0 : for i := range m.levels {
1334 0 : if iter := m.levels[i].iter; iter != nil {
1335 0 : iter.DebugTree(n)
1336 0 : }
1337 : }
1338 : }
1339 :
1340 0 : func (m *mergingIter) DebugString() string {
1341 0 : var buf bytes.Buffer
1342 0 : sep := ""
1343 0 : for m.heap.len() > 0 {
1344 0 : item := m.heap.pop()
1345 0 : fmt.Fprintf(&buf, "%s%s", sep, item.iterKV.K)
1346 0 : sep = " "
1347 0 : }
1348 0 : var err error
1349 0 : if m.dir == 1 {
1350 0 : err = m.initMinHeap()
1351 0 : } else {
1352 0 : err = m.initMaxHeap()
1353 0 : }
1354 0 : if err != nil {
1355 0 : fmt.Fprintf(&buf, "err=<%s>", err)
1356 0 : }
1357 0 : return buf.String()
1358 : }
1359 :
1360 1 : func (m *mergingIter) ForEachLevelIter(fn func(li *levelIter) bool) {
1361 1 : for _, ml := range m.levels {
1362 1 : if ml.levelIter != nil {
1363 1 : if done := fn(ml.levelIter); done {
1364 1 : break
1365 : }
1366 : }
1367 : }
1368 : }
1369 :
1370 1 : func (m *mergingIter) addItemStats(l *mergingIterLevel) {
1371 1 : m.stats.PointCount++
1372 1 : m.stats.KeyBytes += uint64(len(l.iterKV.K.UserKey))
1373 1 : m.stats.ValueBytes += uint64(len(l.iterKV.V.ValueOrHandle))
1374 1 : }
1375 :
1376 : var _ internalIterator = &mergingIter{}
|
