Line data Source code
1 : // Copyright 2012 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 : "io"
10 : "math"
11 : "sync"
12 : "time"
13 :
14 : "github.com/cockroachdb/pebble/internal/invariants"
15 : "github.com/cockroachdb/pebble/rangekey"
16 : "github.com/cockroachdb/pebble/sstable"
17 : )
18 :
19 : // Snapshot provides a read-only point-in-time view of the DB state.
20 : type Snapshot struct {
21 : // The db the snapshot was created from.
22 : db *DB
23 : seqNum uint64
24 :
25 : // Set if part of an EventuallyFileOnlySnapshot.
26 : efos *EventuallyFileOnlySnapshot
27 :
28 : // The list the snapshot is linked into.
29 : list *snapshotList
30 :
31 : // The next/prev link for the snapshotList doubly-linked list of snapshots.
32 : prev, next *Snapshot
33 : }
34 :
35 : var _ Reader = (*Snapshot)(nil)
36 :
37 : // Get gets the value for the given key. It returns ErrNotFound if the Snapshot
38 : // does not contain the key.
39 : //
40 : // The caller should not modify the contents of the returned slice, but it is
41 : // safe to modify the contents of the argument after Get returns. The returned
42 : // slice will remain valid until the returned Closer is closed. On success, the
43 : // caller MUST call closer.Close() or a memory leak will occur.
44 1 : func (s *Snapshot) Get(key []byte) ([]byte, io.Closer, error) {
45 1 : if s.db == nil {
46 1 : panic(ErrClosed)
47 : }
48 1 : return s.db.getInternal(key, nil /* batch */, s)
49 : }
50 :
51 : // NewIter returns an iterator that is unpositioned (Iterator.Valid() will
52 : // return false). The iterator can be positioned via a call to SeekGE,
53 : // SeekLT, First or Last.
54 1 : func (s *Snapshot) NewIter(o *IterOptions) (*Iterator, error) {
55 1 : return s.NewIterWithContext(context.Background(), o)
56 1 : }
57 :
58 : // NewIterWithContext is like NewIter, and additionally accepts a context for
59 : // tracing.
60 1 : func (s *Snapshot) NewIterWithContext(ctx context.Context, o *IterOptions) (*Iterator, error) {
61 1 : if s.db == nil {
62 1 : panic(ErrClosed)
63 : }
64 1 : return s.db.newIter(ctx, nil /* batch */, newIterOpts{
65 1 : snapshot: snapshotIterOpts{seqNum: s.seqNum},
66 1 : }, o), nil
67 : }
68 :
69 : // ScanInternal scans all internal keys within the specified bounds, truncating
70 : // any rangedels and rangekeys to those bounds. For use when an external user
71 : // needs to be aware of all internal keys that make up a key range.
72 : //
73 : // See comment on db.ScanInternal for the behaviour that can be expected of
74 : // point keys deleted by range dels and keys masked by range keys.
75 : func (s *Snapshot) ScanInternal(
76 : ctx context.Context,
77 : categoryAndQoS sstable.CategoryAndQoS,
78 : lower, upper []byte,
79 : visitPointKey func(key *InternalKey, value LazyValue, iterInfo IteratorLevel) error,
80 : visitRangeDel func(start, end []byte, seqNum uint64) error,
81 : visitRangeKey func(start, end []byte, keys []rangekey.Key) error,
82 : visitSharedFile func(sst *SharedSSTMeta) error,
83 : visitExternalFile func(sst *ExternalFile) error,
84 1 : ) error {
85 1 : if s.db == nil {
86 0 : panic(ErrClosed)
87 : }
88 1 : scanInternalOpts := &scanInternalOptions{
89 1 : CategoryAndQoS: categoryAndQoS,
90 1 : visitPointKey: visitPointKey,
91 1 : visitRangeDel: visitRangeDel,
92 1 : visitRangeKey: visitRangeKey,
93 1 : visitSharedFile: visitSharedFile,
94 1 : visitExternalFile: visitExternalFile,
95 1 : IterOptions: IterOptions{
96 1 : KeyTypes: IterKeyTypePointsAndRanges,
97 1 : LowerBound: lower,
98 1 : UpperBound: upper,
99 1 : },
100 1 : }
101 1 :
102 1 : iter, err := s.db.newInternalIter(ctx, snapshotIterOpts{seqNum: s.seqNum}, scanInternalOpts)
103 1 : if err != nil {
104 0 : return err
105 0 : }
106 1 : defer iter.close()
107 1 :
108 1 : return scanInternalImpl(ctx, lower, upper, iter, scanInternalOpts)
109 : }
110 :
111 : // closeLocked is similar to Close(), except it requires that db.mu be held
112 : // by the caller.
113 1 : func (s *Snapshot) closeLocked() error {
114 1 : s.db.mu.snapshots.remove(s)
115 1 :
116 1 : // If s was the previous earliest snapshot, we might be able to reclaim
117 1 : // disk space by dropping obsolete records that were pinned by s.
118 1 : if e := s.db.mu.snapshots.earliest(); e > s.seqNum {
119 1 : s.db.maybeScheduleCompactionPicker(pickElisionOnly)
120 1 : }
121 1 : s.db = nil
122 1 : return nil
123 : }
124 :
125 : // Close closes the snapshot, releasing its resources. Close must be called.
126 : // Failure to do so will result in a tiny memory leak and a large leak of
127 : // resources on disk due to the entries the snapshot is preventing from being
128 : // deleted.
129 : //
130 : // d.mu must NOT be held by the caller.
131 1 : func (s *Snapshot) Close() error {
132 1 : db := s.db
133 1 : if db == nil {
134 1 : panic(ErrClosed)
135 : }
136 1 : db.mu.Lock()
137 1 : defer db.mu.Unlock()
138 1 : return s.closeLocked()
139 : }
140 :
141 : type snapshotList struct {
142 : root Snapshot
143 : }
144 :
145 1 : func (l *snapshotList) init() {
146 1 : l.root.next = &l.root
147 1 : l.root.prev = &l.root
148 1 : }
149 :
150 1 : func (l *snapshotList) empty() bool {
151 1 : return l.root.next == &l.root
152 1 : }
153 :
154 1 : func (l *snapshotList) count() int {
155 1 : if l.empty() {
156 1 : return 0
157 1 : }
158 0 : var count int
159 0 : for i := l.root.next; i != &l.root; i = i.next {
160 0 : count++
161 0 : }
162 0 : return count
163 : }
164 :
165 1 : func (l *snapshotList) earliest() uint64 {
166 1 : v := uint64(math.MaxUint64)
167 1 : if !l.empty() {
168 1 : v = l.root.next.seqNum
169 1 : }
170 1 : return v
171 : }
172 :
173 1 : func (l *snapshotList) toSlice() []uint64 {
174 1 : if l.empty() {
175 1 : return nil
176 1 : }
177 1 : var results []uint64
178 1 : for i := l.root.next; i != &l.root; i = i.next {
179 1 : results = append(results, i.seqNum)
180 1 : }
181 1 : return results
182 : }
183 :
184 1 : func (l *snapshotList) pushBack(s *Snapshot) {
185 1 : if s.list != nil || s.prev != nil || s.next != nil {
186 0 : panic("pebble: snapshot list is inconsistent")
187 : }
188 1 : s.prev = l.root.prev
189 1 : s.prev.next = s
190 1 : s.next = &l.root
191 1 : s.next.prev = s
192 1 : s.list = l
193 : }
194 :
195 1 : func (l *snapshotList) remove(s *Snapshot) {
196 1 : if s == &l.root {
197 0 : panic("pebble: cannot remove snapshot list root node")
198 : }
199 1 : if s.list != l {
200 0 : panic("pebble: snapshot list is inconsistent")
201 : }
202 1 : s.prev.next = s.next
203 1 : s.next.prev = s.prev
204 1 : s.next = nil // avoid memory leaks
205 1 : s.prev = nil // avoid memory leaks
206 1 : s.list = nil // avoid memory leaks
207 : }
208 :
209 : // EventuallyFileOnlySnapshot (aka EFOS) provides a read-only point-in-time view
210 : // of the database state, similar to Snapshot. An EventuallyFileOnlySnapshot
211 : // induces less write amplification than Snapshot, at the cost of increased space
212 : // amplification. While a Snapshot may increase write amplification across all
213 : // flushes and compactions for the duration of its lifetime, an
214 : // EventuallyFileOnlySnapshot only incurs that cost for flushes/compactions if
215 : // memtables at the time of EFOS instantiation contained keys that the EFOS is
216 : // interested in (i.e. its protectedRanges). In that case, the EFOS prevents
217 : // elision of keys visible to it, similar to a Snapshot, until those memtables
218 : // are flushed, and once that happens, the "EventuallyFileOnlySnapshot"
219 : // transitions to a file-only snapshot state in which it pins zombies sstables
220 : // like an open Iterator would, without pinning any memtables. Callers that can
221 : // tolerate the increased space amplification of pinning zombie sstables until
222 : // the snapshot is closed may prefer EventuallyFileOnlySnapshots for their
223 : // reduced write amplification. Callers that desire the benefits of the file-only
224 : // state that requires no pinning of memtables should call
225 : // `WaitForFileOnlySnapshot()` before relying on the EFOS to keep producing iterators
226 : // with zero write-amp and zero pinning of memtables in memory.
227 : //
228 : // EventuallyFileOnlySnapshots interact with the IngestAndExcise operation in
229 : // subtle ways. The IngestAndExcise can force the transition of an EFOS to a
230 : // file-only snapshot if an excise overlaps with the EFOS bounds.
231 : type EventuallyFileOnlySnapshot struct {
232 : mu struct {
233 : // NB: If both this mutex and db.mu are being grabbed, db.mu should be
234 : // grabbed _before_ grabbing this one.
235 : sync.Mutex
236 :
237 : // Either the snap field is set below, or the version is set at any given
238 : // point of time. If a snapshot is referenced, this is not a file-only
239 : // snapshot yet, and if a version is set (and ref'd) this is a file-only
240 : // snapshot.
241 :
242 : // The wrapped regular snapshot, if not a file-only snapshot yet.
243 : snap *Snapshot
244 : // The wrapped version reference, if a file-only snapshot.
245 : vers *version
246 : }
247 :
248 : // Key ranges to watch for an excise on.
249 : protectedRanges []KeyRange
250 :
251 : // The db the snapshot was created from.
252 : db *DB
253 : seqNum uint64
254 : closed chan struct{}
255 : }
256 :
257 1 : func (d *DB) makeEventuallyFileOnlySnapshot(keyRanges []KeyRange) *EventuallyFileOnlySnapshot {
258 1 : isFileOnly := true
259 1 :
260 1 : d.mu.Lock()
261 1 : defer d.mu.Unlock()
262 1 : seqNum := d.mu.versions.visibleSeqNum.Load()
263 1 : // Check if any of the keyRanges overlap with a memtable.
264 1 : for i := range d.mu.mem.queue {
265 1 : d.mu.mem.queue[i].computePossibleOverlaps(func(bounded) shouldContinue {
266 1 : isFileOnly = false
267 1 : return stopIteration
268 1 : }, sliceAsBounded(keyRanges)...)
269 : }
270 1 : es := &EventuallyFileOnlySnapshot{
271 1 : db: d,
272 1 : seqNum: seqNum,
273 1 : protectedRanges: keyRanges,
274 1 : closed: make(chan struct{}),
275 1 : }
276 1 : if isFileOnly {
277 1 : es.mu.vers = d.mu.versions.currentVersion()
278 1 : es.mu.vers.Ref()
279 1 : } else {
280 1 : s := &Snapshot{
281 1 : db: d,
282 1 : seqNum: seqNum,
283 1 : }
284 1 : s.efos = es
285 1 : es.mu.snap = s
286 1 : d.mu.snapshots.pushBack(s)
287 1 : }
288 1 : return es
289 : }
290 :
291 : // Transitions this EventuallyFileOnlySnapshot to a file-only snapshot. Requires
292 : // earliestUnflushedSeqNum and vers to correspond to the same Version from the
293 : // current or a past acquisition of db.mu. vers must have been Ref()'d before
294 : // that mutex was released, if it was released.
295 : //
296 : // NB: The caller is expected to check for es.excised before making this
297 : // call.
298 : //
299 : // d.mu must be held when calling this method.
300 1 : func (es *EventuallyFileOnlySnapshot) transitionToFileOnlySnapshot(vers *version) error {
301 1 : es.mu.Lock()
302 1 : select {
303 0 : case <-es.closed:
304 0 : vers.UnrefLocked()
305 0 : es.mu.Unlock()
306 0 : return ErrClosed
307 1 : default:
308 : }
309 1 : if es.mu.snap == nil {
310 0 : es.mu.Unlock()
311 0 : panic("pebble: tried to transition an eventually-file-only-snapshot twice")
312 : }
313 : // The caller has already called Ref() on vers.
314 1 : es.mu.vers = vers
315 1 : // NB: The callers should have already done a check of es.excised.
316 1 : oldSnap := es.mu.snap
317 1 : es.mu.snap = nil
318 1 : es.mu.Unlock()
319 1 : return oldSnap.closeLocked()
320 : }
321 :
322 : // hasTransitioned returns true if this EFOS has transitioned to a file-only
323 : // snapshot.
324 1 : func (es *EventuallyFileOnlySnapshot) hasTransitioned() bool {
325 1 : es.mu.Lock()
326 1 : defer es.mu.Unlock()
327 1 : return es.mu.vers != nil
328 1 : }
329 :
330 : // waitForFlush waits for a flush on any memtables that need to be flushed
331 : // before this EFOS can transition to a file-only snapshot. If this EFOS is
332 : // waiting on a flush of the mutable memtable, it forces a rotation within
333 : // `dur` duration. For immutable memtables, it schedules a flush and waits for
334 : // it to finish.
335 1 : func (es *EventuallyFileOnlySnapshot) waitForFlush(ctx context.Context, dur time.Duration) error {
336 1 : es.db.mu.Lock()
337 1 : defer es.db.mu.Unlock()
338 1 :
339 1 : earliestUnflushedSeqNum := es.db.getEarliestUnflushedSeqNumLocked()
340 1 : for earliestUnflushedSeqNum < es.seqNum {
341 1 : select {
342 0 : case <-es.closed:
343 0 : return ErrClosed
344 0 : case <-ctx.Done():
345 0 : return ctx.Err()
346 1 : default:
347 : }
348 : // Check if the current mutable memtable contains keys less than seqNum.
349 : // If so, rotate it.
350 1 : if es.db.mu.mem.mutable.logSeqNum < es.seqNum && dur.Nanoseconds() > 0 {
351 1 : es.db.maybeScheduleDelayedFlush(es.db.mu.mem.mutable, dur)
352 1 : } else {
353 0 : // Find the last memtable that contains seqNums less than es.seqNum,
354 0 : // and force a flush on it.
355 0 : var mem *flushableEntry
356 0 : for i := range es.db.mu.mem.queue {
357 0 : if es.db.mu.mem.queue[i].logSeqNum < es.seqNum {
358 0 : mem = es.db.mu.mem.queue[i]
359 0 : }
360 : }
361 0 : mem.flushForced = true
362 0 : es.db.maybeScheduleFlush()
363 : }
364 1 : es.db.mu.compact.cond.Wait()
365 1 :
366 1 : earliestUnflushedSeqNum = es.db.getEarliestUnflushedSeqNumLocked()
367 : }
368 1 : return nil
369 : }
370 :
371 : // WaitForFileOnlySnapshot blocks the calling goroutine until this snapshot
372 : // has been converted into a file-only snapshot (i.e. all memtables containing
373 : // keys < seqNum are flushed). A duration can be passed in, and if nonzero,
374 : // a delayed flush will be scheduled at that duration if necessary.
375 : //
376 : // Idempotent; can be called multiple times with no side effects.
377 : func (es *EventuallyFileOnlySnapshot) WaitForFileOnlySnapshot(
378 : ctx context.Context, dur time.Duration,
379 1 : ) error {
380 1 : if es.hasTransitioned() {
381 1 : return nil
382 1 : }
383 :
384 1 : if err := es.waitForFlush(ctx, dur); err != nil {
385 0 : return err
386 0 : }
387 :
388 1 : if invariants.Enabled {
389 1 : // Since we aren't returning an error, we _must_ have transitioned to a
390 1 : // file-only snapshot by now.
391 1 : if !es.hasTransitioned() {
392 0 : panic("expected EFOS to have transitioned to file-only snapshot after flush")
393 : }
394 : }
395 1 : return nil
396 : }
397 :
398 : // Close closes the file-only snapshot and releases all referenced resources.
399 : // Not idempotent.
400 1 : func (es *EventuallyFileOnlySnapshot) Close() error {
401 1 : close(es.closed)
402 1 : es.db.mu.Lock()
403 1 : defer es.db.mu.Unlock()
404 1 : es.mu.Lock()
405 1 : defer es.mu.Unlock()
406 1 :
407 1 : if es.mu.snap != nil {
408 1 : if err := es.mu.snap.closeLocked(); err != nil {
409 0 : return err
410 0 : }
411 : }
412 1 : if es.mu.vers != nil {
413 1 : es.mu.vers.UnrefLocked()
414 1 : }
415 1 : return nil
416 : }
417 :
418 : // Get implements the Reader interface.
419 1 : func (es *EventuallyFileOnlySnapshot) Get(key []byte) (value []byte, closer io.Closer, err error) {
420 1 : // TODO(jackson): Use getInternal.
421 1 : iter, err := es.NewIter(nil)
422 1 : if err != nil {
423 0 : return nil, nil, err
424 0 : }
425 1 : if !iter.SeekPrefixGE(key) {
426 1 : if err = firstError(iter.Error(), iter.Close()); err != nil {
427 0 : return nil, nil, err
428 0 : }
429 1 : return nil, nil, ErrNotFound
430 : }
431 0 : if !es.db.equal(iter.Key(), key) {
432 0 : return nil, nil, firstError(iter.Close(), ErrNotFound)
433 0 : }
434 0 : return iter.Value(), iter, nil
435 : }
436 :
437 : // NewIter returns an iterator that is unpositioned (Iterator.Valid() will
438 : // return false). The iterator can be positioned via a call to SeekGE,
439 : // SeekLT, First or Last.
440 1 : func (es *EventuallyFileOnlySnapshot) NewIter(o *IterOptions) (*Iterator, error) {
441 1 : return es.NewIterWithContext(context.Background(), o)
442 1 : }
443 :
444 : // NewIterWithContext is like NewIter, and additionally accepts a context for
445 : // tracing.
446 : func (es *EventuallyFileOnlySnapshot) NewIterWithContext(
447 : ctx context.Context, o *IterOptions,
448 1 : ) (*Iterator, error) {
449 1 : select {
450 0 : case <-es.closed:
451 0 : panic(ErrClosed)
452 1 : default:
453 : }
454 :
455 1 : es.mu.Lock()
456 1 : defer es.mu.Unlock()
457 1 : if es.mu.vers != nil {
458 1 : sOpts := snapshotIterOpts{seqNum: es.seqNum, vers: es.mu.vers}
459 1 : return es.db.newIter(ctx, nil /* batch */, newIterOpts{snapshot: sOpts}, o), nil
460 1 : }
461 :
462 1 : sOpts := snapshotIterOpts{seqNum: es.seqNum}
463 1 : iter := es.db.newIter(ctx, nil /* batch */, newIterOpts{snapshot: sOpts}, o)
464 1 : return iter, nil
465 : }
466 :
467 : // ScanInternal scans all internal keys within the specified bounds, truncating
468 : // any rangedels and rangekeys to those bounds. For use when an external user
469 : // needs to be aware of all internal keys that make up a key range.
470 : //
471 : // See comment on db.ScanInternal for the behaviour that can be expected of
472 : // point keys deleted by range dels and keys masked by range keys.
473 : func (es *EventuallyFileOnlySnapshot) ScanInternal(
474 : ctx context.Context,
475 : categoryAndQoS sstable.CategoryAndQoS,
476 : lower, upper []byte,
477 : visitPointKey func(key *InternalKey, value LazyValue, iterInfo IteratorLevel) error,
478 : visitRangeDel func(start, end []byte, seqNum uint64) error,
479 : visitRangeKey func(start, end []byte, keys []rangekey.Key) error,
480 : visitSharedFile func(sst *SharedSSTMeta) error,
481 : visitExternalFile func(sst *ExternalFile) error,
482 1 : ) error {
483 1 : if es.db == nil {
484 0 : panic(ErrClosed)
485 : }
486 1 : var sOpts snapshotIterOpts
487 1 : opts := &scanInternalOptions{
488 1 : CategoryAndQoS: categoryAndQoS,
489 1 : IterOptions: IterOptions{
490 1 : KeyTypes: IterKeyTypePointsAndRanges,
491 1 : LowerBound: lower,
492 1 : UpperBound: upper,
493 1 : },
494 1 : visitPointKey: visitPointKey,
495 1 : visitRangeDel: visitRangeDel,
496 1 : visitRangeKey: visitRangeKey,
497 1 : visitSharedFile: visitSharedFile,
498 1 : visitExternalFile: visitExternalFile,
499 1 : }
500 1 : es.mu.Lock()
501 1 : if es.mu.vers != nil {
502 1 : sOpts = snapshotIterOpts{
503 1 : seqNum: es.seqNum,
504 1 : vers: es.mu.vers,
505 1 : }
506 1 : } else {
507 1 : sOpts = snapshotIterOpts{
508 1 : seqNum: es.seqNum,
509 1 : }
510 1 : }
511 1 : es.mu.Unlock()
512 1 : iter, err := es.db.newInternalIter(ctx, sOpts, opts)
513 1 : if err != nil {
514 0 : return err
515 0 : }
516 1 : defer iter.close()
517 1 :
518 1 : return scanInternalImpl(ctx, lower, upper, iter, opts)
519 : }
|
