Line data Source code
1 : // Copyright 2020 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "context"
9 : "fmt"
10 : "sync/atomic"
11 : "time"
12 :
13 : "github.com/cockroachdb/errors"
14 : "github.com/cockroachdb/pebble/internal/base"
15 : "github.com/cockroachdb/pebble/internal/invariants"
16 : "github.com/cockroachdb/pebble/internal/keyspan"
17 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
18 : "github.com/cockroachdb/pebble/internal/manifest"
19 : )
20 :
21 : // flushable defines the interface for immutable memtables.
22 : type flushable interface {
23 : newIter(o *IterOptions) internalIterator
24 : newFlushIter(o *IterOptions) internalIterator
25 : newRangeDelIter(o *IterOptions) keyspan.FragmentIterator
26 : newRangeKeyIter(o *IterOptions) keyspan.FragmentIterator
27 : containsRangeKeys() bool
28 : // inuseBytes returns the number of inuse bytes by the flushable.
29 : inuseBytes() uint64
30 : // totalBytes returns the total number of bytes allocated by the flushable.
31 : totalBytes() uint64
32 : // readyForFlush returns true when the flushable is ready for flushing. See
33 : // memTable.readyForFlush for one implementation which needs to check whether
34 : // there are any outstanding write references.
35 : readyForFlush() bool
36 : // computePossibleOverlaps determines whether the flushable's keys overlap
37 : // with the bounds of any of the provided bounded items. If an item overlaps
38 : // or might overlap but it's not possible to determine overlap cheaply,
39 : // computePossibleOverlaps invokes the provided function with the object
40 : // that might overlap. computePossibleOverlaps must not perform any I/O and
41 : // implementations should invoke the provided function for items that would
42 : // require I/O to determine overlap.
43 : computePossibleOverlaps(overlaps func(bounded) shouldContinue, bounded ...bounded)
44 : }
45 :
46 : type shouldContinue bool
47 :
48 : const (
49 : continueIteration shouldContinue = true
50 : stopIteration = false
51 : )
52 :
53 : type bounded interface {
54 : UserKeyBounds() base.UserKeyBounds
55 : }
56 :
57 : var _ bounded = (*fileMetadata)(nil)
58 : var _ bounded = KeyRange{}
59 :
60 1 : func sliceAsBounded[B bounded](s []B) []bounded {
61 1 : ret := make([]bounded, len(s))
62 1 : for i := 0; i < len(s); i++ {
63 1 : ret[i] = s[i]
64 1 : }
65 1 : return ret
66 : }
67 :
68 : // flushableEntry wraps a flushable and adds additional metadata and
69 : // functionality that is common to all flushables.
70 : type flushableEntry struct {
71 : flushable
72 : // Channel which is closed when the flushable has been flushed.
73 : flushed chan struct{}
74 : // flushForced indicates whether a flush was forced on this memtable (either
75 : // manual, or due to ingestion). Protected by DB.mu.
76 : flushForced bool
77 : // delayedFlushForcedAt indicates whether a timer has been set to force a
78 : // flush on this memtable at some point in the future. Protected by DB.mu.
79 : // Holds the timestamp of when the flush will be issued.
80 : delayedFlushForcedAt time.Time
81 : // logNum corresponds to the WAL that contains the records present in the
82 : // receiver.
83 : logNum base.DiskFileNum
84 : // logSize is the size in bytes of the associated WAL. Protected by DB.mu.
85 : logSize uint64
86 : // The current logSeqNum at the time the memtable was created. This is
87 : // guaranteed to be less than or equal to any seqnum stored in the memtable.
88 : logSeqNum base.SeqNum
89 : // readerRefs tracks the read references on the flushable. The two sources of
90 : // reader references are DB.mu.mem.queue and readState.memtables. The memory
91 : // reserved by the flushable in the cache is released when the reader refs
92 : // drop to zero. If the flushable is referencing sstables, then the file
93 : // refount is also decreased once the reader refs drops to 0. If the
94 : // flushable is a memTable, when the reader refs drops to zero, the writer
95 : // refs will already be zero because the memtable will have been flushed and
96 : // that only occurs once the writer refs drops to zero.
97 : readerRefs atomic.Int32
98 : // Closure to invoke to release memory accounting.
99 : releaseMemAccounting func()
100 : // unrefFiles, if not nil, should be invoked to decrease the ref count of
101 : // files which are backing the flushable.
102 : unrefFiles func() []*fileBacking
103 : // deleteFnLocked should be called if the caller is holding DB.mu.
104 : deleteFnLocked func(obsolete []*fileBacking)
105 : // deleteFn should be called if the caller is not holding DB.mu.
106 : deleteFn func(obsolete []*fileBacking)
107 : }
108 :
109 1 : func (e *flushableEntry) readerRef() {
110 1 : switch v := e.readerRefs.Add(1); {
111 0 : case v <= 1:
112 0 : panic(fmt.Sprintf("pebble: inconsistent reference count: %d", v))
113 : }
114 : }
115 :
116 : // db.mu must not be held when this is called.
117 1 : func (e *flushableEntry) readerUnref(deleteFiles bool) {
118 1 : e.readerUnrefHelper(deleteFiles, e.deleteFn)
119 1 : }
120 :
121 : // db.mu must be held when this is called.
122 1 : func (e *flushableEntry) readerUnrefLocked(deleteFiles bool) {
123 1 : e.readerUnrefHelper(deleteFiles, e.deleteFnLocked)
124 1 : }
125 :
126 : func (e *flushableEntry) readerUnrefHelper(
127 : deleteFiles bool, deleteFn func(obsolete []*fileBacking),
128 1 : ) {
129 1 : switch v := e.readerRefs.Add(-1); {
130 0 : case v < 0:
131 0 : panic(fmt.Sprintf("pebble: inconsistent reference count: %d", v))
132 1 : case v == 0:
133 1 : if e.releaseMemAccounting == nil {
134 0 : panic("pebble: memtable reservation already released")
135 : }
136 1 : e.releaseMemAccounting()
137 1 : e.releaseMemAccounting = nil
138 1 : if e.unrefFiles != nil {
139 1 : obsolete := e.unrefFiles()
140 1 : e.unrefFiles = nil
141 1 : if deleteFiles {
142 1 : deleteFn(obsolete)
143 1 : }
144 : }
145 : }
146 : }
147 :
148 : type flushableList []*flushableEntry
149 :
150 : // ingestedFlushable is the implementation of the flushable interface for the
151 : // ingesting sstables which are added to the flushable list.
152 : type ingestedFlushable struct {
153 : // files are non-overlapping and ordered (according to their bounds).
154 : files []physicalMeta
155 : comparer *Comparer
156 : newIters tableNewIters
157 : newRangeKeyIters keyspanimpl.TableNewSpanIter
158 :
159 : // Since the level slice is immutable, we construct and set it once. It
160 : // should be safe to read from slice in future reads.
161 : slice manifest.LevelSlice
162 : // hasRangeKeys is set on ingestedFlushable construction.
163 : hasRangeKeys bool
164 : // exciseSpan is populated if an excise operation should be performed during
165 : // flush.
166 : exciseSpan KeyRange
167 : }
168 :
169 : func newIngestedFlushable(
170 : files []*fileMetadata,
171 : comparer *Comparer,
172 : newIters tableNewIters,
173 : newRangeKeyIters keyspanimpl.TableNewSpanIter,
174 : exciseSpan KeyRange,
175 1 : ) *ingestedFlushable {
176 1 : if invariants.Enabled {
177 1 : for i := 1; i < len(files); i++ {
178 1 : prev := files[i-1].UserKeyBounds()
179 1 : this := files[i].UserKeyBounds()
180 1 : if prev.End.IsUpperBoundFor(comparer.Compare, this.Start) {
181 0 : panic(errors.AssertionFailedf("ingested flushable files overlap: %s %s", prev, this))
182 : }
183 : }
184 : }
185 1 : var physicalFiles []physicalMeta
186 1 : var hasRangeKeys bool
187 1 : for _, f := range files {
188 1 : if f.HasRangeKeys {
189 1 : hasRangeKeys = true
190 1 : }
191 1 : physicalFiles = append(physicalFiles, f.PhysicalMeta())
192 : }
193 :
194 1 : ret := &ingestedFlushable{
195 1 : files: physicalFiles,
196 1 : comparer: comparer,
197 1 : newIters: newIters,
198 1 : newRangeKeyIters: newRangeKeyIters,
199 1 : // slice is immutable and can be set once and used many times.
200 1 : slice: manifest.NewLevelSliceKeySorted(comparer.Compare, files),
201 1 : hasRangeKeys: hasRangeKeys,
202 1 : exciseSpan: exciseSpan,
203 1 : }
204 1 :
205 1 : return ret
206 : }
207 :
208 : // TODO(sumeer): ingestedFlushable iters also need to plumb context for
209 : // tracing.
210 :
211 : // newIter is part of the flushable interface.
212 1 : func (s *ingestedFlushable) newIter(o *IterOptions) internalIterator {
213 1 : var opts IterOptions
214 1 : if o != nil {
215 1 : opts = *o
216 1 : }
217 1 : return newLevelIter(
218 1 : context.Background(), opts, s.comparer, s.newIters, s.slice.Iter(), manifest.FlushableIngestsLayer(),
219 1 : internalIterOpts{},
220 1 : )
221 : }
222 :
223 : // newFlushIter is part of the flushable interface.
224 0 : func (s *ingestedFlushable) newFlushIter(*IterOptions) internalIterator {
225 0 : // newFlushIter is only used for writing memtables to disk as sstables.
226 0 : // Since ingested sstables are already present on disk, they don't need to
227 0 : // make use of a flush iter.
228 0 : panic("pebble: not implemented")
229 : }
230 :
231 : func (s *ingestedFlushable) constructRangeDelIter(
232 : ctx context.Context, file *manifest.FileMetadata, _ keyspan.SpanIterOptions,
233 1 : ) (keyspan.FragmentIterator, error) {
234 1 : iters, err := s.newIters(ctx, file, nil, internalIterOpts{}, iterRangeDeletions)
235 1 : if err != nil {
236 0 : return nil, err
237 0 : }
238 1 : return iters.RangeDeletion(), nil
239 : }
240 :
241 : // newRangeDelIter is part of the flushable interface.
242 : // TODO(bananabrick): Using a level iter instead of a keyspan level iter to
243 : // surface range deletes is more efficient.
244 : //
245 : // TODO(sumeer): *IterOptions are being ignored, so the index block load for
246 : // the point iterator in constructRangeDeIter is not tracked.
247 1 : func (s *ingestedFlushable) newRangeDelIter(_ *IterOptions) keyspan.FragmentIterator {
248 1 : return keyspanimpl.NewLevelIter(
249 1 : context.TODO(),
250 1 : keyspan.SpanIterOptions{}, s.comparer.Compare,
251 1 : s.constructRangeDelIter, s.slice.Iter(), manifest.FlushableIngestsLayer(),
252 1 : manifest.KeyTypePoint,
253 1 : )
254 1 : }
255 :
256 : // newRangeKeyIter is part of the flushable interface.
257 1 : func (s *ingestedFlushable) newRangeKeyIter(o *IterOptions) keyspan.FragmentIterator {
258 1 : if !s.containsRangeKeys() {
259 1 : return nil
260 1 : }
261 :
262 1 : return keyspanimpl.NewLevelIter(
263 1 : context.TODO(),
264 1 : keyspan.SpanIterOptions{}, s.comparer.Compare, s.newRangeKeyIters,
265 1 : s.slice.Iter(), manifest.FlushableIngestsLayer(), manifest.KeyTypeRange,
266 1 : )
267 : }
268 :
269 : // containsRangeKeys is part of the flushable interface.
270 1 : func (s *ingestedFlushable) containsRangeKeys() bool {
271 1 : return s.hasRangeKeys
272 1 : }
273 :
274 : // inuseBytes is part of the flushable interface.
275 0 : func (s *ingestedFlushable) inuseBytes() uint64 {
276 0 : // inuseBytes is only used when memtables are flushed to disk as sstables.
277 0 : panic("pebble: not implemented")
278 : }
279 :
280 : // totalBytes is part of the flushable interface.
281 1 : func (s *ingestedFlushable) totalBytes() uint64 {
282 1 : // We don't allocate additional bytes for the ingestedFlushable.
283 1 : return 0
284 1 : }
285 :
286 : // readyForFlush is part of the flushable interface.
287 1 : func (s *ingestedFlushable) readyForFlush() bool {
288 1 : // ingestedFlushable should always be ready to flush. However, note that
289 1 : // memtables before the ingested sstables in the memtable queue must be
290 1 : // flushed before an ingestedFlushable can be flushed. This is because the
291 1 : // ingested sstables need an updated view of the Version to
292 1 : // determine where to place the files in the lsm.
293 1 : return true
294 1 : }
295 :
296 : // computePossibleOverlaps is part of the flushable interface.
297 : func (s *ingestedFlushable) computePossibleOverlaps(
298 : fn func(bounded) shouldContinue, bounded ...bounded,
299 1 : ) {
300 1 : for _, b := range bounded {
301 1 : if s.anyFileOverlaps(b.UserKeyBounds()) {
302 1 : // Some file overlaps in key boundaries. The file doesn't necessarily
303 1 : // contain any keys within the key range, but we would need to perform I/O
304 1 : // to know for sure. The flushable interface dictates that we're not
305 1 : // permitted to perform I/O here, so err towards assuming overlap.
306 1 : if !fn(b) {
307 1 : return
308 1 : }
309 : }
310 : }
311 : }
312 :
313 : // anyFileBoundsOverlap returns true if there is at least a file in s.files with
314 : // bounds that overlap the given bounds.
315 1 : func (s *ingestedFlushable) anyFileOverlaps(bounds base.UserKeyBounds) bool {
316 1 : // Note that s.files are non-overlapping and sorted.
317 1 : for _, f := range s.files {
318 1 : fileBounds := f.UserKeyBounds()
319 1 : if !fileBounds.End.IsUpperBoundFor(s.comparer.Compare, bounds.Start) {
320 1 : // The file ends before the bounds start. Go to the next file.
321 1 : continue
322 : }
323 1 : if !bounds.End.IsUpperBoundFor(s.comparer.Compare, fileBounds.Start) {
324 1 : // The file starts after the bounds end. There is no overlap, and
325 1 : // further files will not overlap either (the files are sorted).
326 1 : return false
327 1 : }
328 : // There is overlap. Note that UserKeyBounds.Overlaps() performs exactly the
329 : // checks above.
330 1 : return true
331 : }
332 1 : return false
333 : }
334 :
335 : // computePossibleOverlapsGenericImpl is an implementation of the flushable
336 : // interface's computePossibleOverlaps function for flushable implementations
337 : // with only in-memory state that do not have special requirements and should
338 : // read through the ordinary flushable iterators.
339 : //
340 : // This function must only be used with implementations that are infallible (eg,
341 : // memtable iterators) and will panic if an error is encountered.
342 : func computePossibleOverlapsGenericImpl[F flushable](
343 : f F, cmp Compare, fn func(bounded) shouldContinue, bounded []bounded,
344 1 : ) {
345 1 : iter := f.newIter(nil)
346 1 : rangeDelIter := f.newRangeDelIter(nil)
347 1 : rangeKeyIter := f.newRangeKeyIter(nil)
348 1 : for _, b := range bounded {
349 1 : overlap, err := determineOverlapAllIters(cmp, b.UserKeyBounds(), iter, rangeDelIter, rangeKeyIter)
350 1 : if invariants.Enabled && err != nil {
351 0 : panic(errors.AssertionFailedf("expected iterator to be infallible: %v", err))
352 : }
353 1 : if overlap {
354 1 : if !fn(b) {
355 1 : break
356 : }
357 : }
358 : }
359 :
360 1 : if iter != nil {
361 1 : if err := iter.Close(); err != nil {
362 0 : // This implementation must be used in circumstances where
363 0 : // reading through the iterator is infallible.
364 0 : panic(err)
365 : }
366 : }
367 1 : if rangeDelIter != nil {
368 1 : rangeDelIter.Close()
369 1 : }
370 1 : if rangeKeyIter != nil {
371 1 : rangeKeyIter.Close()
372 1 : }
373 : }
374 :
375 : // determineOverlapAllIters checks for overlap in a point iterator, range
376 : // deletion iterator and range key iterator.
377 : func determineOverlapAllIters(
378 : cmp base.Compare,
379 : bounds base.UserKeyBounds,
380 : pointIter base.InternalIterator,
381 : rangeDelIter, rangeKeyIter keyspan.FragmentIterator,
382 1 : ) (bool, error) {
383 1 : if pointIter != nil {
384 1 : if pointOverlap, err := determineOverlapPointIterator(cmp, bounds, pointIter); pointOverlap || err != nil {
385 1 : return pointOverlap, err
386 1 : }
387 : }
388 1 : if rangeDelIter != nil {
389 1 : if rangeDelOverlap, err := determineOverlapKeyspanIterator(cmp, bounds, rangeDelIter); rangeDelOverlap || err != nil {
390 1 : return rangeDelOverlap, err
391 1 : }
392 : }
393 1 : if rangeKeyIter != nil {
394 1 : return determineOverlapKeyspanIterator(cmp, bounds, rangeKeyIter)
395 1 : }
396 1 : return false, nil
397 : }
398 :
399 : func determineOverlapPointIterator(
400 : cmp base.Compare, bounds base.UserKeyBounds, iter internalIterator,
401 1 : ) (bool, error) {
402 1 : kv := iter.SeekGE(bounds.Start, base.SeekGEFlagsNone)
403 1 : if kv == nil {
404 1 : return false, iter.Error()
405 1 : }
406 1 : return bounds.End.IsUpperBoundForInternalKey(cmp, kv.K), nil
407 : }
408 :
409 : func determineOverlapKeyspanIterator(
410 : cmp base.Compare, bounds base.UserKeyBounds, iter keyspan.FragmentIterator,
411 1 : ) (bool, error) {
412 1 : // NB: The spans surfaced by the fragment iterator are non-overlapping.
413 1 : span, err := iter.SeekGE(bounds.Start)
414 1 : if err != nil {
415 0 : return false, err
416 0 : }
417 1 : for ; span != nil; span, err = iter.Next() {
418 1 : if !bounds.End.IsUpperBoundFor(cmp, span.Start) {
419 1 : // The span starts after our bounds.
420 1 : return false, nil
421 1 : }
422 1 : if !span.Empty() {
423 1 : return true, nil
424 1 : }
425 : }
426 1 : return false, err
427 : }
|
