Line data Source code
1 : // Copyright 2019 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 base
6 :
7 : import (
8 : "context"
9 : "fmt"
10 : "time"
11 : )
12 :
13 : // InternalIterator iterates over a DB's key/value pairs in key order. Unlike
14 : // the Iterator interface, the returned keys are InternalKeys composed of the
15 : // user-key, a sequence number and a key kind. In forward iteration, key/value
16 : // pairs for identical user-keys are returned in descending sequence order. In
17 : // reverse iteration, key/value pairs for identical user-keys are returned in
18 : // ascending sequence order.
19 : //
20 : // InternalIterators provide 5 absolute positioning methods and 2 relative
21 : // positioning methods. The absolute positioning methods are:
22 : //
23 : // - SeekGE
24 : // - SeekPrefixGE
25 : // - SeekLT
26 : // - First
27 : // - Last
28 : //
29 : // The relative positioning methods are:
30 : //
31 : // - Next
32 : // - Prev
33 : //
34 : // The relative positioning methods can be used in conjunction with any of the
35 : // absolute positioning methods with one exception: SeekPrefixGE does not
36 : // support reverse iteration via Prev. It is undefined to call relative
37 : // positioning methods without ever calling an absolute positioning method.
38 : //
39 : // InternalIterators can optionally implement a prefix iteration mode. This
40 : // mode is entered by calling SeekPrefixGE and exited by any other absolute
41 : // positioning method (SeekGE, SeekLT, First, Last). When in prefix iteration
42 : // mode, a call to Next will advance to the next key which has the same
43 : // "prefix" as the one supplied to SeekPrefixGE. Note that "prefix" in this
44 : // context is not a strict byte prefix, but defined by byte equality for the
45 : // result of the Comparer.Split method. An InternalIterator is not required to
46 : // support prefix iteration mode, and can implement SeekPrefixGE by forwarding
47 : // to SeekGE. When the iteration prefix is exhausted, it is not valid to call
48 : // Next on an internal iterator that's already returned (nil,nilv) or a key
49 : // beyond the prefix.
50 : //
51 : // Bounds, [lower, upper), can be set on iterators, either using the SetBounds()
52 : // function in the interface, or in implementation specific ways during iterator
53 : // creation. The forward positioning routines (SeekGE, First, and Next) only
54 : // check the upper bound. The reverse positioning routines (SeekLT, Last, and
55 : // Prev) only check the lower bound. It is up to the caller to ensure that the
56 : // forward positioning routines respect the lower bound and the reverse
57 : // positioning routines respect the upper bound (i.e. calling SeekGE instead of
58 : // First if there is a lower bound, and SeekLT instead of Last if there is an
59 : // upper bound). This imposition is done in order to elevate that enforcement to
60 : // the caller (generally pebble.Iterator or pebble.mergingIter) rather than
61 : // having it duplicated in every InternalIterator implementation.
62 : //
63 : // Additionally, the caller needs to ensure that SeekGE/SeekPrefixGE are not
64 : // called with a key > the upper bound, and SeekLT is not called with a key <
65 : // the lower bound. InternalIterator implementations are required to respect
66 : // the iterator bounds, never returning records outside of the bounds with one
67 : // exception: an iterator may generate synthetic RANGEDEL marker records. See
68 : // levelIter.syntheticBoundary for the sole existing example of this behavior.
69 : // Specifically, levelIter can return synthetic keys whose user key is equal to
70 : // the lower/upper bound.
71 : //
72 : // The bounds provided to an internal iterator must remain valid until a
73 : // subsequent call to SetBounds has returned. This requirement exists so that
74 : // iterator implementations may compare old and new bounds to apply low-level
75 : // optimizations. The pebble.Iterator satisfies this requirement by maintaining
76 : // two bound buffers and switching between them.
77 : //
78 : // An iterator must be closed after use, but it is not necessary to read an
79 : // iterator until exhaustion.
80 : //
81 : // An iterator is not goroutine-safe, but it is safe to use multiple iterators
82 : // concurrently, either in separate goroutines or switching between the
83 : // iterators in a single goroutine.
84 : //
85 : // It is also safe to use an iterator concurrently with modifying its
86 : // underlying DB, if that DB permits modification. However, the resultant
87 : // key/value pairs are not guaranteed to be a consistent snapshot of that DB
88 : // at a particular point in time.
89 : //
90 : // InternalIterators accumulate errors encountered during operation, exposing
91 : // them through the Error method. All of the absolute positioning methods
92 : // reset any accumulated error before positioning. Relative positioning
93 : // methods return without advancing if the iterator has accumulated an error.
94 : //
95 : // nilv == shorthand for LazyValue{}, which represents a nil value.
96 : type InternalIterator interface {
97 : // SeekGE moves the iterator to the first key/value pair whose key is greater
98 : // than or equal to the given key. Returns the key and value if the iterator
99 : // is pointing at a valid entry, and (nil, nilv) otherwise. Note that SeekGE
100 : // only checks the upper bound. It is up to the caller to ensure that key
101 : // is greater than or equal to the lower bound.
102 : SeekGE(key []byte, flags SeekGEFlags) (*InternalKey, LazyValue)
103 :
104 : // SeekPrefixGE moves the iterator to the first key/value pair whose key is
105 : // greater than or equal to the given key. Returns the key and value if the
106 : // iterator is pointing at a valid entry, and (nil, nilv) otherwise. Note that
107 : // SeekPrefixGE only checks the upper bound. It is up to the caller to ensure
108 : // that key is greater than or equal to the lower bound.
109 : //
110 : // The prefix argument is used by some InternalIterator implementations (e.g.
111 : // sstable.Reader) to avoid expensive operations. This operation is only
112 : // useful when a user-defined Split function is supplied to the Comparer for
113 : // the DB. The supplied prefix will be the prefix of the given key returned by
114 : // that Split function. If the iterator is able to determine that no key with
115 : // the prefix exists, it can return (nil,nilv). Unlike SeekGE, this is not an
116 : // indication that iteration is exhausted.
117 : //
118 : // Note that the iterator may return keys not matching the prefix. It is up
119 : // to the caller to check if the prefix matches.
120 : //
121 : // Calling SeekPrefixGE places the receiver into prefix iteration mode. Once
122 : // in this mode, reverse iteration may not be supported and will return an
123 : // error. Note that pebble/Iterator.SeekPrefixGE has this same restriction on
124 : // not supporting reverse iteration in prefix iteration mode until a
125 : // different positioning routine (SeekGE, SeekLT, First or Last) switches the
126 : // iterator out of prefix iteration.
127 : SeekPrefixGE(prefix, key []byte, flags SeekGEFlags) (*InternalKey, LazyValue)
128 :
129 : // SeekLT moves the iterator to the last key/value pair whose key is less
130 : // than the given key. Returns the key and value if the iterator is pointing
131 : // at a valid entry, and (nil, nilv) otherwise. Note that SeekLT only checks
132 : // the lower bound. It is up to the caller to ensure that key is less than
133 : // the upper bound.
134 : SeekLT(key []byte, flags SeekLTFlags) (*InternalKey, LazyValue)
135 :
136 : // First moves the iterator the the first key/value pair. Returns the key and
137 : // value if the iterator is pointing at a valid entry, and (nil, nilv)
138 : // otherwise. Note that First only checks the upper bound. It is up to the
139 : // caller to ensure that First() is not called when there is a lower bound,
140 : // and instead call SeekGE(lower).
141 : First() (*InternalKey, LazyValue)
142 :
143 : // Last moves the iterator the the last key/value pair. Returns the key and
144 : // value if the iterator is pointing at a valid entry, and (nil, nilv)
145 : // otherwise. Note that Last only checks the lower bound. It is up to the
146 : // caller to ensure that Last() is not called when there is an upper bound,
147 : // and instead call SeekLT(upper).
148 : Last() (*InternalKey, LazyValue)
149 :
150 : // Next moves the iterator to the next key/value pair. Returns the key and
151 : // value if the iterator is pointing at a valid entry, and (nil, nilv)
152 : // otherwise. Note that Next only checks the upper bound. It is up to the
153 : // caller to ensure that key is greater than or equal to the lower bound.
154 : //
155 : // It is valid to call Next when the iterator is positioned before the first
156 : // key/value pair due to either a prior call to SeekLT or Prev which returned
157 : // (nil, nilv). It is not allowed to call Next when the previous call to SeekGE,
158 : // SeekPrefixGE or Next returned (nil, nilv).
159 : Next() (*InternalKey, LazyValue)
160 :
161 : // NextPrefix moves the iterator to the next key/value pair with a different
162 : // prefix than the key at the current iterator position. Returns the key and
163 : // value if the iterator is pointing at a valid entry, and (nil, nil)
164 : // otherwise. Note that NextPrefix only checks the upper bound. It is up to
165 : // the caller to ensure that key is greater than or equal to the lower
166 : // bound.
167 : //
168 : // NextPrefix is passed the immediate successor to the current prefix key. A
169 : // valid implementation of NextPrefix is to call SeekGE with succKey.
170 : //
171 : // It is not allowed to call NextPrefix when the previous call was a reverse
172 : // positioning operation or a call to a forward positioning method that
173 : // returned (nil, nilv). It is also not allowed to call NextPrefix when the
174 : // iterator is in prefix iteration mode.
175 : NextPrefix(succKey []byte) (*InternalKey, LazyValue)
176 :
177 : // Prev moves the iterator to the previous key/value pair. Returns the key
178 : // and value if the iterator is pointing at a valid entry, and (nil, nilv)
179 : // otherwise. Note that Prev only checks the lower bound. It is up to the
180 : // caller to ensure that key is less than the upper bound.
181 : //
182 : // It is valid to call Prev when the iterator is positioned after the last
183 : // key/value pair due to either a prior call to SeekGE or Next which returned
184 : // (nil, nilv). It is not allowed to call Prev when the previous call to SeekLT
185 : // or Prev returned (nil, nilv).
186 : Prev() (*InternalKey, LazyValue)
187 :
188 : // Error returns any accumulated error. It may not include errors returned
189 : // to the client when calling LazyValue.Value().
190 : Error() error
191 :
192 : // Close closes the iterator and returns any accumulated error. Exhausting
193 : // all the key/value pairs in a table is not considered to be an error.
194 : //
195 : // Once Close is called, the iterator should not be used again. Specific
196 : // implementations may support multiple calls to Close (but no other calls
197 : // after the first Close).
198 : Close() error
199 :
200 : // SetBounds sets the lower and upper bounds for the iterator. Note that the
201 : // result of Next and Prev will be undefined until the iterator has been
202 : // repositioned with SeekGE, SeekPrefixGE, SeekLT, First, or Last.
203 : //
204 : // The bounds provided must remain valid until a subsequent call to
205 : // SetBounds has returned. This requirement exists so that iterator
206 : // implementations may compare old and new bounds to apply low-level
207 : // optimizations.
208 : SetBounds(lower, upper []byte)
209 :
210 : // SetContext replaces the context provided at iterator creation, or the
211 : // last one provided by SetContext.
212 : SetContext(ctx context.Context)
213 :
214 : fmt.Stringer
215 : }
216 :
217 : // TopLevelIterator extends InternalIterator to include an additional absolute
218 : // positioning method, SeekPrefixGEStrict.
219 : type TopLevelIterator interface {
220 : InternalIterator
221 :
222 : // SeekPrefixGEStrict extends InternalIterator.SeekPrefixGE with a guarantee
223 : // that the iterator only returns keys matching the prefix.
224 : SeekPrefixGEStrict(prefix, key []byte, flags SeekGEFlags) (*InternalKey, LazyValue)
225 : }
226 :
227 : // SeekGEFlags holds flags that may configure the behavior of a forward seek.
228 : // Not all flags are relevant to all iterators.
229 : type SeekGEFlags uint8
230 :
231 : const (
232 : seekGEFlagTrySeekUsingNext uint8 = iota
233 : seekGEFlagRelativeSeek
234 : seekGEFlagBatchJustRefreshed
235 : )
236 :
237 : // SeekGEFlagsNone is the default value of SeekGEFlags, with all flags disabled.
238 : const SeekGEFlagsNone = SeekGEFlags(0)
239 :
240 : // TrySeekUsingNext indicates whether a performance optimization was enabled
241 : // by a caller, indicating the caller has not done any action to move this
242 : // iterator beyond the first key that would be found if this iterator were to
243 : // honestly do the intended seek. For example, say the caller did a
244 : // SeekGE(k1...), followed by SeekGE(k2...) where k1 <= k2, without any
245 : // intermediate positioning calls. The caller can safely specify true for this
246 : // parameter in the second call. As another example, say the caller did do one
247 : // call to Next between the two Seek calls, and k1 < k2. Again, the caller can
248 : // safely specify a true value for this parameter. Note that a false value is
249 : // always safe. The callee is free to ignore the true value if its
250 : // implementation does not permit this optimization.
251 : //
252 : // We make the caller do this determination since a string comparison of k1, k2
253 : // is not necessarily cheap, and there may be many iterators in the iterator
254 : // stack. Doing it once at the root of the iterator stack is cheaper.
255 : //
256 : // This optimization could also be applied to SeekLT (where it would be
257 : // trySeekUsingPrev). We currently only do it for SeekPrefixGE and SeekGE
258 : // because this is where this optimization helps the performance of CockroachDB.
259 : // The SeekLT cases in CockroachDB are typically accompanied with bounds that
260 : // change between seek calls, and is optimized inside certain iterator
261 : // implementations, like singleLevelIterator, without any extra parameter
262 : // passing (though the same amortization of string comparisons could be done to
263 : // improve that optimization, by making the root of the iterator stack do it).
264 1 : func (s SeekGEFlags) TrySeekUsingNext() bool { return (s & (1 << seekGEFlagTrySeekUsingNext)) != 0 }
265 :
266 : // RelativeSeek is set when in the course of a forward positioning operation, a
267 : // higher-level iterator seeks a lower-level iterator to a larger key than the
268 : // one at the current iterator position.
269 : //
270 : // Concretely, this occurs when the merging iterator observes a range deletion
271 : // covering the key at a level's current position, and the merging iterator
272 : // seeks the level to the range deletion's end key. During lazy-combined
273 : // iteration, this flag signals to the level iterator that the seek is NOT an
274 : // absolute-positioning operation from the perspective of the pebble.Iterator,
275 : // and the level iterator must look for range keys in tables between the current
276 : // iterator position and the new seeked position.
277 1 : func (s SeekGEFlags) RelativeSeek() bool { return (s & (1 << seekGEFlagRelativeSeek)) != 0 }
278 :
279 : // BatchJustRefreshed is set by Seek[Prefix]GE when an iterator's view of an
280 : // indexed batch was just refreshed. It serves as a signal to the batch iterator
281 : // to ignore the TrySeekUsingNext optimization, because the external knowledge
282 : // imparted by the TrySeekUsingNext flag does not apply to the batch iterator's
283 : // position. See (pebble.Iterator).batchJustRefreshed.
284 1 : func (s SeekGEFlags) BatchJustRefreshed() bool { return (s & (1 << seekGEFlagBatchJustRefreshed)) != 0 }
285 :
286 : // EnableTrySeekUsingNext returns the provided flags with the
287 : // try-seek-using-next optimization enabled. See TrySeekUsingNext for an
288 : // explanation of this optimization.
289 1 : func (s SeekGEFlags) EnableTrySeekUsingNext() SeekGEFlags {
290 1 : return s | (1 << seekGEFlagTrySeekUsingNext)
291 1 : }
292 :
293 : // DisableTrySeekUsingNext returns the provided flags with the
294 : // try-seek-using-next optimization disabled.
295 1 : func (s SeekGEFlags) DisableTrySeekUsingNext() SeekGEFlags {
296 1 : return s &^ (1 << seekGEFlagTrySeekUsingNext)
297 1 : }
298 :
299 : // EnableRelativeSeek returns the provided flags with the relative-seek flag
300 : // enabled. See RelativeSeek for an explanation of this flag's use.
301 1 : func (s SeekGEFlags) EnableRelativeSeek() SeekGEFlags {
302 1 : return s | (1 << seekGEFlagRelativeSeek)
303 1 : }
304 :
305 : // DisableRelativeSeek returns the provided flags with the relative-seek flag
306 : // disabled.
307 1 : func (s SeekGEFlags) DisableRelativeSeek() SeekGEFlags {
308 1 : return s &^ (1 << seekGEFlagRelativeSeek)
309 1 : }
310 :
311 : // EnableBatchJustRefreshed returns the provided flags with the
312 : // batch-just-refreshed bit set. See BatchJustRefreshed for an explanation of
313 : // this flag.
314 1 : func (s SeekGEFlags) EnableBatchJustRefreshed() SeekGEFlags {
315 1 : return s | (1 << seekGEFlagBatchJustRefreshed)
316 1 : }
317 :
318 : // DisableBatchJustRefreshed returns the provided flags with the
319 : // batch-just-refreshed bit unset.
320 1 : func (s SeekGEFlags) DisableBatchJustRefreshed() SeekGEFlags {
321 1 : return s &^ (1 << seekGEFlagBatchJustRefreshed)
322 1 : }
323 :
324 : // SeekLTFlags holds flags that may configure the behavior of a reverse seek.
325 : // Not all flags are relevant to all iterators.
326 : type SeekLTFlags uint8
327 :
328 : const (
329 : seekLTFlagRelativeSeek uint8 = iota
330 : )
331 :
332 : // SeekLTFlagsNone is the default value of SeekLTFlags, with all flags disabled.
333 : const SeekLTFlagsNone = SeekLTFlags(0)
334 :
335 : // RelativeSeek is set when in the course of a reverse positioning operation, a
336 : // higher-level iterator seeks a lower-level iterator to a smaller key than the
337 : // one at the current iterator position.
338 : //
339 : // Concretely, this occurs when the merging iterator observes a range deletion
340 : // covering the key at a level's current position, and the merging iterator
341 : // seeks the level to the range deletion's start key. During lazy-combined
342 : // iteration, this flag signals to the level iterator that the seek is NOT an
343 : // absolute-positioning operation from the perspective of the pebble.Iterator,
344 : // and the level iterator must look for range keys in tables between the current
345 : // iterator position and the new seeked position.
346 1 : func (s SeekLTFlags) RelativeSeek() bool { return s&(1<<seekLTFlagRelativeSeek) != 0 }
347 :
348 : // EnableRelativeSeek returns the provided flags with the relative-seek flag
349 : // enabled. See RelativeSeek for an explanation of this flag's use.
350 1 : func (s SeekLTFlags) EnableRelativeSeek() SeekLTFlags {
351 1 : return s | (1 << seekLTFlagRelativeSeek)
352 1 : }
353 :
354 : // DisableRelativeSeek returns the provided flags with the relative-seek flag
355 : // disabled.
356 1 : func (s SeekLTFlags) DisableRelativeSeek() SeekLTFlags {
357 1 : return s &^ (1 << seekLTFlagRelativeSeek)
358 1 : }
359 :
360 : // InternalIteratorStats contains miscellaneous stats produced by
361 : // InternalIterators that are part of the InternalIterator tree. Not every
362 : // field is relevant for an InternalIterator implementation. The field values
363 : // are aggregated as one goes up the InternalIterator tree.
364 : type InternalIteratorStats struct {
365 : // Bytes in the loaded blocks. If the block was compressed, this is the
366 : // compressed bytes. Currently, only the index blocks, data blocks
367 : // containing points, and filter blocks are included.
368 : BlockBytes uint64
369 : // Subset of BlockBytes that were in the block cache.
370 : BlockBytesInCache uint64
371 : // BlockReadDuration accumulates the duration spent fetching blocks
372 : // due to block cache misses.
373 : // TODO(sumeer): this currently excludes the time spent in Reader creation,
374 : // and in reading the rangedel and rangekey blocks. Fix that.
375 : BlockReadDuration time.Duration
376 : // The following can repeatedly count the same points if they are iterated
377 : // over multiple times. Additionally, they may count a point twice when
378 : // switching directions. The latter could be improved if needed.
379 :
380 : // Bytes in keys that were iterated over. Currently, only point keys are
381 : // included.
382 : KeyBytes uint64
383 : // Bytes in values that were iterated over. Currently, only point values are
384 : // included. For separated values, this is the size of the handle.
385 : ValueBytes uint64
386 : // The count of points iterated over.
387 : PointCount uint64
388 : // Points that were iterated over that were covered by range tombstones. It
389 : // can be useful for discovering instances of
390 : // https://github.com/cockroachdb/pebble/issues/1070.
391 : PointsCoveredByRangeTombstones uint64
392 :
393 : // Stats related to points in value blocks encountered during iteration.
394 : // These are useful to understand outliers, since typical user facing
395 : // iteration should tend to only look at the latest point, and hence have
396 : // the following stats close to 0.
397 : SeparatedPointValue struct {
398 : // Count is a count of points that were in value blocks. This is not a
399 : // subset of PointCount: PointCount is produced by mergingIter and if
400 : // positioned once, and successful in returning a point, will have a
401 : // PointCount of 1, regardless of how many sstables (and memtables etc.)
402 : // in the heap got positioned. The count here includes every sstable
403 : // iterator that got positioned in the heap.
404 : Count uint64
405 : // ValueBytes represent the total byte length of the values (in value
406 : // blocks) of the points corresponding to Count.
407 : ValueBytes uint64
408 : // ValueBytesFetched is the total byte length of the values (in value
409 : // blocks) that were retrieved.
410 : ValueBytesFetched uint64
411 : }
412 : }
413 :
414 : // Merge merges the stats in from into the given stats.
415 1 : func (s *InternalIteratorStats) Merge(from InternalIteratorStats) {
416 1 : s.BlockBytes += from.BlockBytes
417 1 : s.BlockBytesInCache += from.BlockBytesInCache
418 1 : s.BlockReadDuration += from.BlockReadDuration
419 1 : s.KeyBytes += from.KeyBytes
420 1 : s.ValueBytes += from.ValueBytes
421 1 : s.PointCount += from.PointCount
422 1 : s.PointsCoveredByRangeTombstones += from.PointsCoveredByRangeTombstones
423 1 : s.SeparatedPointValue.Count += from.SeparatedPointValue.Count
424 1 : s.SeparatedPointValue.ValueBytes += from.SeparatedPointValue.ValueBytes
425 1 : s.SeparatedPointValue.ValueBytesFetched += from.SeparatedPointValue.ValueBytesFetched
426 1 : }
|