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 provides an ordered key/value store.
6 : package pebble // import "github.com/cockroachdb/pebble"
7 :
8 : import (
9 : "context"
10 : "fmt"
11 : "io"
12 : "strconv"
13 : "sync"
14 : "sync/atomic"
15 : "time"
16 :
17 : "github.com/cockroachdb/errors"
18 : "github.com/cockroachdb/pebble/internal/arenaskl"
19 : "github.com/cockroachdb/pebble/internal/base"
20 : "github.com/cockroachdb/pebble/internal/invalidating"
21 : "github.com/cockroachdb/pebble/internal/invariants"
22 : "github.com/cockroachdb/pebble/internal/keyspan"
23 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
24 : "github.com/cockroachdb/pebble/internal/manifest"
25 : "github.com/cockroachdb/pebble/internal/manual"
26 : "github.com/cockroachdb/pebble/objstorage"
27 : "github.com/cockroachdb/pebble/objstorage/remote"
28 : "github.com/cockroachdb/pebble/rangekey"
29 : "github.com/cockroachdb/pebble/record"
30 : "github.com/cockroachdb/pebble/sstable"
31 : "github.com/cockroachdb/pebble/vfs"
32 : "github.com/cockroachdb/pebble/vfs/atomicfs"
33 : "github.com/cockroachdb/pebble/wal"
34 : "github.com/cockroachdb/tokenbucket"
35 : "github.com/prometheus/client_golang/prometheus"
36 : )
37 :
38 : const (
39 : // minTableCacheSize is the minimum size of the table cache, for a single db.
40 : minTableCacheSize = 64
41 :
42 : // numNonTableCacheFiles is an approximation for the number of files
43 : // that we don't use for table caches, for a given db.
44 : numNonTableCacheFiles = 10
45 : )
46 :
47 : var (
48 : // ErrNotFound is returned when a get operation does not find the requested
49 : // key.
50 : ErrNotFound = base.ErrNotFound
51 : // ErrClosed is panicked when an operation is performed on a closed snapshot or
52 : // DB. Use errors.Is(err, ErrClosed) to check for this error.
53 : ErrClosed = errors.New("pebble: closed")
54 : // ErrReadOnly is returned when a write operation is performed on a read-only
55 : // database.
56 : ErrReadOnly = errors.New("pebble: read-only")
57 : // errNoSplit indicates that the user is trying to perform a range key
58 : // operation but the configured Comparer does not provide a Split
59 : // implementation.
60 : errNoSplit = errors.New("pebble: Comparer.Split required for range key operations")
61 : )
62 :
63 : // Reader is a readable key/value store.
64 : //
65 : // It is safe to call Get and NewIter from concurrent goroutines.
66 : type Reader interface {
67 : // Get gets the value for the given key. It returns ErrNotFound if the DB
68 : // does not contain the key.
69 : //
70 : // The caller should not modify the contents of the returned slice, but it is
71 : // safe to modify the contents of the argument after Get returns. The
72 : // returned slice will remain valid until the returned Closer is closed. On
73 : // success, the caller MUST call closer.Close() or a memory leak will occur.
74 : Get(key []byte) (value []byte, closer io.Closer, err error)
75 :
76 : // NewIter returns an iterator that is unpositioned (Iterator.Valid() will
77 : // return false). The iterator can be positioned via a call to SeekGE,
78 : // SeekLT, First or Last.
79 : NewIter(o *IterOptions) (*Iterator, error)
80 :
81 : // NewIterWithContext is like NewIter, and additionally accepts a context
82 : // for tracing.
83 : NewIterWithContext(ctx context.Context, o *IterOptions) (*Iterator, error)
84 :
85 : // Close closes the Reader. It may or may not close any underlying io.Reader
86 : // or io.Writer, depending on how the DB was created.
87 : //
88 : // It is not safe to close a DB until all outstanding iterators are closed.
89 : // It is valid to call Close multiple times. Other methods should not be
90 : // called after the DB has been closed.
91 : Close() error
92 : }
93 :
94 : // Writer is a writable key/value store.
95 : //
96 : // Goroutine safety is dependent on the specific implementation.
97 : type Writer interface {
98 : // Apply the operations contained in the batch to the DB.
99 : //
100 : // It is safe to modify the contents of the arguments after Apply returns.
101 : Apply(batch *Batch, o *WriteOptions) error
102 :
103 : // Delete deletes the value for the given key. Deletes are blind all will
104 : // succeed even if the given key does not exist.
105 : //
106 : // It is safe to modify the contents of the arguments after Delete returns.
107 : Delete(key []byte, o *WriteOptions) error
108 :
109 : // DeleteSized behaves identically to Delete, but takes an additional
110 : // argument indicating the size of the value being deleted. DeleteSized
111 : // should be preferred when the caller has the expectation that there exists
112 : // a single internal KV pair for the key (eg, the key has not been
113 : // overwritten recently), and the caller knows the size of its value.
114 : //
115 : // DeleteSized will record the value size within the tombstone and use it to
116 : // inform compaction-picking heuristics which strive to reduce space
117 : // amplification in the LSM. This "calling your shot" mechanic allows the
118 : // storage engine to more accurately estimate and reduce space
119 : // amplification.
120 : //
121 : // It is safe to modify the contents of the arguments after DeleteSized
122 : // returns.
123 : DeleteSized(key []byte, valueSize uint32, _ *WriteOptions) error
124 :
125 : // SingleDelete is similar to Delete in that it deletes the value for the given key. Like Delete,
126 : // it is a blind operation that will succeed even if the given key does not exist.
127 : //
128 : // WARNING: Undefined (non-deterministic) behavior will result if a key is overwritten and
129 : // then deleted using SingleDelete. The record may appear deleted immediately, but be
130 : // resurrected at a later time after compactions have been performed. Or the record may
131 : // be deleted permanently. A Delete operation lays down a "tombstone" which shadows all
132 : // previous versions of a key. The SingleDelete operation is akin to "anti-matter" and will
133 : // only delete the most recently written version for a key. These different semantics allow
134 : // the DB to avoid propagating a SingleDelete operation during a compaction as soon as the
135 : // corresponding Set operation is encountered. These semantics require extreme care to handle
136 : // properly. Only use if you have a workload where the performance gain is critical and you
137 : // can guarantee that a record is written once and then deleted once.
138 : //
139 : // SingleDelete is internally transformed into a Delete if the most recent record for a key is either
140 : // a Merge or Delete record.
141 : //
142 : // It is safe to modify the contents of the arguments after SingleDelete returns.
143 : SingleDelete(key []byte, o *WriteOptions) error
144 :
145 : // DeleteRange deletes all of the point keys (and values) in the range
146 : // [start,end) (inclusive on start, exclusive on end). DeleteRange does NOT
147 : // delete overlapping range keys (eg, keys set via RangeKeySet).
148 : //
149 : // It is safe to modify the contents of the arguments after DeleteRange
150 : // returns.
151 : DeleteRange(start, end []byte, o *WriteOptions) error
152 :
153 : // LogData adds the specified to the batch. The data will be written to the
154 : // WAL, but not added to memtables or sstables. Log data is never indexed,
155 : // which makes it useful for testing WAL performance.
156 : //
157 : // It is safe to modify the contents of the argument after LogData returns.
158 : LogData(data []byte, opts *WriteOptions) error
159 :
160 : // Merge merges the value for the given key. The details of the merge are
161 : // dependent upon the configured merge operation.
162 : //
163 : // It is safe to modify the contents of the arguments after Merge returns.
164 : Merge(key, value []byte, o *WriteOptions) error
165 :
166 : // Set sets the value for the given key. It overwrites any previous value
167 : // for that key; a DB is not a multi-map.
168 : //
169 : // It is safe to modify the contents of the arguments after Set returns.
170 : Set(key, value []byte, o *WriteOptions) error
171 :
172 : // RangeKeySet sets a range key mapping the key range [start, end) at the MVCC
173 : // timestamp suffix to value. The suffix is optional. If any portion of the key
174 : // range [start, end) is already set by a range key with the same suffix value,
175 : // RangeKeySet overrides it.
176 : //
177 : // It is safe to modify the contents of the arguments after RangeKeySet returns.
178 : RangeKeySet(start, end, suffix, value []byte, opts *WriteOptions) error
179 :
180 : // RangeKeyUnset removes a range key mapping the key range [start, end) at the
181 : // MVCC timestamp suffix. The suffix may be omitted to remove an unsuffixed
182 : // range key. RangeKeyUnset only removes portions of range keys that fall within
183 : // the [start, end) key span, and only range keys with suffixes that exactly
184 : // match the unset suffix.
185 : //
186 : // It is safe to modify the contents of the arguments after RangeKeyUnset
187 : // returns.
188 : RangeKeyUnset(start, end, suffix []byte, opts *WriteOptions) error
189 :
190 : // RangeKeyDelete deletes all of the range keys in the range [start,end)
191 : // (inclusive on start, exclusive on end). It does not delete point keys (for
192 : // that use DeleteRange). RangeKeyDelete removes all range keys within the
193 : // bounds, including those with or without suffixes.
194 : //
195 : // It is safe to modify the contents of the arguments after RangeKeyDelete
196 : // returns.
197 : RangeKeyDelete(start, end []byte, opts *WriteOptions) error
198 : }
199 :
200 : // CPUWorkHandle represents a handle used by the CPUWorkPermissionGranter API.
201 : type CPUWorkHandle interface {
202 : // Permitted indicates whether Pebble can use additional CPU resources.
203 : Permitted() bool
204 : }
205 :
206 : // CPUWorkPermissionGranter is used to request permission to opportunistically
207 : // use additional CPUs to speed up internal background work.
208 : type CPUWorkPermissionGranter interface {
209 : // GetPermission returns a handle regardless of whether permission is granted
210 : // or not. In the latter case, the handle is only useful for recording
211 : // the CPU time actually spent on this calling goroutine.
212 : GetPermission(time.Duration) CPUWorkHandle
213 : // CPUWorkDone must be called regardless of whether CPUWorkHandle.Permitted
214 : // returns true or false.
215 : CPUWorkDone(CPUWorkHandle)
216 : }
217 :
218 : // Use a default implementation for the CPU work granter to avoid excessive nil
219 : // checks in the code.
220 : type defaultCPUWorkHandle struct{}
221 :
222 1 : func (d defaultCPUWorkHandle) Permitted() bool {
223 1 : return false
224 1 : }
225 :
226 : type defaultCPUWorkGranter struct{}
227 :
228 1 : func (d defaultCPUWorkGranter) GetPermission(_ time.Duration) CPUWorkHandle {
229 1 : return defaultCPUWorkHandle{}
230 1 : }
231 :
232 1 : func (d defaultCPUWorkGranter) CPUWorkDone(_ CPUWorkHandle) {}
233 :
234 : // DB provides a concurrent, persistent ordered key/value store.
235 : //
236 : // A DB's basic operations (Get, Set, Delete) should be self-explanatory. Get
237 : // and Delete will return ErrNotFound if the requested key is not in the store.
238 : // Callers are free to ignore this error.
239 : //
240 : // A DB also allows for iterating over the key/value pairs in key order. If d
241 : // is a DB, the code below prints all key/value pairs whose keys are 'greater
242 : // than or equal to' k:
243 : //
244 : // iter := d.NewIter(readOptions)
245 : // for iter.SeekGE(k); iter.Valid(); iter.Next() {
246 : // fmt.Printf("key=%q value=%q\n", iter.Key(), iter.Value())
247 : // }
248 : // return iter.Close()
249 : //
250 : // The Options struct holds the optional parameters for the DB, including a
251 : // Comparer to define a 'less than' relationship over keys. It is always valid
252 : // to pass a nil *Options, which means to use the default parameter values. Any
253 : // zero field of a non-nil *Options also means to use the default value for
254 : // that parameter. Thus, the code below uses a custom Comparer, but the default
255 : // values for every other parameter:
256 : //
257 : // db := pebble.Open(&Options{
258 : // Comparer: myComparer,
259 : // })
260 : type DB struct {
261 : // The count and size of referenced memtables. This includes memtables
262 : // present in DB.mu.mem.queue, as well as memtables that have been flushed
263 : // but are still referenced by an inuse readState, as well as up to one
264 : // memTable waiting to be reused and stored in d.memTableRecycle.
265 : memTableCount atomic.Int64
266 : memTableReserved atomic.Int64 // number of bytes reserved in the cache for memtables
267 : // memTableRecycle holds a pointer to an obsolete memtable. The next
268 : // memtable allocation will reuse this memtable if it has not already been
269 : // recycled.
270 : memTableRecycle atomic.Pointer[memTable]
271 :
272 : // The logical size of the current WAL.
273 : logSize atomic.Uint64
274 :
275 : // The number of bytes available on disk.
276 : diskAvailBytes atomic.Uint64
277 :
278 : cacheID uint64
279 : dirname string
280 : opts *Options
281 : cmp Compare
282 : equal Equal
283 : merge Merge
284 : split Split
285 : abbreviatedKey AbbreviatedKey
286 : // The threshold for determining when a batch is "large" and will skip being
287 : // inserted into a memtable.
288 : largeBatchThreshold uint64
289 : // The current OPTIONS file number.
290 : optionsFileNum base.DiskFileNum
291 : // The on-disk size of the current OPTIONS file.
292 : optionsFileSize uint64
293 :
294 : // objProvider is used to access and manage SSTs.
295 : objProvider objstorage.Provider
296 :
297 : fileLock *Lock
298 : dataDir vfs.File
299 :
300 : tableCache *tableCacheContainer
301 : newIters tableNewIters
302 : tableNewRangeKeyIter keyspanimpl.TableNewSpanIter
303 :
304 : commit *commitPipeline
305 :
306 : // readState provides access to the state needed for reading without needing
307 : // to acquire DB.mu.
308 : readState struct {
309 : sync.RWMutex
310 : val *readState
311 : }
312 :
313 : closed *atomic.Value
314 : closedCh chan struct{}
315 :
316 : cleanupManager *cleanupManager
317 :
318 : // During an iterator close, we may asynchronously schedule read compactions.
319 : // We want to wait for those goroutines to finish, before closing the DB.
320 : // compactionShedulers.Wait() should not be called while the DB.mu is held.
321 : compactionSchedulers sync.WaitGroup
322 :
323 : // The main mutex protecting internal DB state. This mutex encompasses many
324 : // fields because those fields need to be accessed and updated atomically. In
325 : // particular, the current version, log.*, mem.*, and snapshot list need to
326 : // be accessed and updated atomically during compaction.
327 : //
328 : // Care is taken to avoid holding DB.mu during IO operations. Accomplishing
329 : // this sometimes requires releasing DB.mu in a method that was called with
330 : // it held. See versionSet.logAndApply() and DB.makeRoomForWrite() for
331 : // examples. This is a common pattern, so be careful about expectations that
332 : // DB.mu will be held continuously across a set of calls.
333 : mu struct {
334 : sync.Mutex
335 :
336 : formatVers struct {
337 : // vers is the database's current format major version.
338 : // Backwards-incompatible features are gated behind new
339 : // format major versions and not enabled until a database's
340 : // version is ratcheted upwards.
341 : //
342 : // Although this is under the `mu` prefix, readers may read vers
343 : // atomically without holding d.mu. Writers must only write to this
344 : // value through finalizeFormatVersUpgrade which requires d.mu is
345 : // held.
346 : vers atomic.Uint64
347 : // marker is the atomic marker for the format major version.
348 : // When a database's version is ratcheted upwards, the
349 : // marker is moved in order to atomically record the new
350 : // version.
351 : marker *atomicfs.Marker
352 : // ratcheting when set to true indicates that the database is
353 : // currently in the process of ratcheting the format major version
354 : // to vers + 1. As a part of ratcheting the format major version,
355 : // migrations may drop and re-acquire the mutex.
356 : ratcheting bool
357 : }
358 :
359 : // The ID of the next job. Job IDs are passed to event listener
360 : // notifications and act as a mechanism for tying together the events and
361 : // log messages for a single job such as a flush, compaction, or file
362 : // ingestion. Job IDs are not serialized to disk or used for correctness.
363 : nextJobID JobID
364 :
365 : // The collection of immutable versions and state about the log and visible
366 : // sequence numbers. Use the pointer here to ensure the atomic fields in
367 : // version set are aligned properly.
368 : versions *versionSet
369 :
370 : log struct {
371 : // manager is not protected by mu, but calls to Create must be
372 : // serialized, and happen after the previous writer is closed.
373 : manager wal.Manager
374 : // The number of input bytes to the log. This is the raw size of the
375 : // batches written to the WAL, without the overhead of the record
376 : // envelopes.
377 : bytesIn uint64
378 : // The Writer is protected by commitPipeline.mu. This allows log writes
379 : // to be performed without holding DB.mu, but requires both
380 : // commitPipeline.mu and DB.mu to be held when rotating the WAL/memtable
381 : // (i.e. makeRoomForWrite). Can be nil.
382 : writer wal.Writer
383 : metrics struct {
384 : // fsyncLatency has its own internal synchronization, and is not
385 : // protected by mu.
386 : fsyncLatency prometheus.Histogram
387 : // Updated whenever a wal.Writer is closed.
388 : record.LogWriterMetrics
389 : }
390 : }
391 :
392 : mem struct {
393 : // The current mutable memTable.
394 : mutable *memTable
395 : // Queue of flushables (the mutable memtable is at end). Elements are
396 : // added to the end of the slice and removed from the beginning. Once an
397 : // index is set it is never modified making a fixed slice immutable and
398 : // safe for concurrent reads.
399 : queue flushableList
400 : // nextSize is the size of the next memtable. The memtable size starts at
401 : // min(256KB,Options.MemTableSize) and doubles each time a new memtable
402 : // is allocated up to Options.MemTableSize. This reduces the memory
403 : // footprint of memtables when lots of DB instances are used concurrently
404 : // in test environments.
405 : nextSize uint64
406 : }
407 :
408 : compact struct {
409 : // Condition variable used to signal when a flush or compaction has
410 : // completed. Used by the write-stall mechanism to wait for the stall
411 : // condition to clear. See DB.makeRoomForWrite().
412 : cond sync.Cond
413 : // True when a flush is in progress.
414 : flushing bool
415 : // The number of ongoing non-download compactions.
416 : compactingCount int
417 : // The number of download compactions.
418 : downloadingCount int
419 : // The list of deletion hints, suggesting ranges for delete-only
420 : // compactions.
421 : deletionHints []deleteCompactionHint
422 : // The list of manual compactions. The next manual compaction to perform
423 : // is at the start of the list. New entries are added to the end.
424 : manual []*manualCompaction
425 : // downloads is the list of suggested download tasks. The next download to
426 : // perform is at the start of the list. New entries are added to the end.
427 : downloads []*downloadSpan
428 : // inProgress is the set of in-progress flushes and compactions.
429 : // It's used in the calculation of some metrics and to initialize L0
430 : // sublevels' state. Some of the compactions contained within this
431 : // map may have already committed an edit to the version but are
432 : // lingering performing cleanup, like deleting obsolete files.
433 : inProgress map[*compaction]struct{}
434 :
435 : // rescheduleReadCompaction indicates to an iterator that a read compaction
436 : // should be scheduled.
437 : rescheduleReadCompaction bool
438 :
439 : // readCompactions is a readCompactionQueue which keeps track of the
440 : // compactions which we might have to perform.
441 : readCompactions readCompactionQueue
442 :
443 : // The cumulative duration of all completed compactions since Open.
444 : // Does not include flushes.
445 : duration time.Duration
446 : // Flush throughput metric.
447 : flushWriteThroughput ThroughputMetric
448 : // The idle start time for the flush "loop", i.e., when the flushing
449 : // bool above transitions to false.
450 : noOngoingFlushStartTime time.Time
451 : }
452 :
453 : // Non-zero when file cleaning is disabled. The disabled count acts as a
454 : // reference count to prohibit file cleaning. See
455 : // DB.{disable,Enable}FileDeletions().
456 : disableFileDeletions int
457 :
458 : snapshots struct {
459 : // The list of active snapshots.
460 : snapshotList
461 :
462 : // The cumulative count and size of snapshot-pinned keys written to
463 : // sstables.
464 : cumulativePinnedCount uint64
465 : cumulativePinnedSize uint64
466 : }
467 :
468 : tableStats struct {
469 : // Condition variable used to signal the completion of a
470 : // job to collect table stats.
471 : cond sync.Cond
472 : // True when a stat collection operation is in progress.
473 : loading bool
474 : // True if stat collection has loaded statistics for all tables
475 : // other than those listed explicitly in pending. This flag starts
476 : // as false when a database is opened and flips to true once stat
477 : // collection has caught up.
478 : loadedInitial bool
479 : // A slice of files for which stats have not been computed.
480 : // Compactions, ingests, flushes append files to be processed. An
481 : // active stat collection goroutine clears the list and processes
482 : // them.
483 : pending []manifest.NewFileEntry
484 : }
485 :
486 : tableValidation struct {
487 : // cond is a condition variable used to signal the completion of a
488 : // job to validate one or more sstables.
489 : cond sync.Cond
490 : // pending is a slice of metadata for sstables waiting to be
491 : // validated. Only physical sstables should be added to the pending
492 : // queue.
493 : pending []newFileEntry
494 : // validating is set to true when validation is running.
495 : validating bool
496 : }
497 : }
498 :
499 : // Normally equal to time.Now() but may be overridden in tests.
500 : timeNow func() time.Time
501 : // the time at database Open; may be used to compute metrics like effective
502 : // compaction concurrency
503 : openedAt time.Time
504 : }
505 :
506 : var _ Reader = (*DB)(nil)
507 : var _ Writer = (*DB)(nil)
508 :
509 : // TestOnlyWaitForCleaning MUST only be used in tests.
510 1 : func (d *DB) TestOnlyWaitForCleaning() {
511 1 : d.cleanupManager.Wait()
512 1 : }
513 :
514 : // Get gets the value for the given key. It returns ErrNotFound if the DB does
515 : // not contain the key.
516 : //
517 : // The caller should not modify the contents of the returned slice, but it is
518 : // safe to modify the contents of the argument after Get returns. The returned
519 : // slice will remain valid until the returned Closer is closed. On success, the
520 : // caller MUST call closer.Close() or a memory leak will occur.
521 1 : func (d *DB) Get(key []byte) ([]byte, io.Closer, error) {
522 1 : return d.getInternal(key, nil /* batch */, nil /* snapshot */)
523 1 : }
524 :
525 : type getIterAlloc struct {
526 : dbi Iterator
527 : keyBuf []byte
528 : get getIter
529 : }
530 :
531 : var getIterAllocPool = sync.Pool{
532 1 : New: func() interface{} {
533 1 : return &getIterAlloc{}
534 1 : },
535 : }
536 :
537 1 : func (d *DB) getInternal(key []byte, b *Batch, s *Snapshot) ([]byte, io.Closer, error) {
538 1 : if err := d.closed.Load(); err != nil {
539 1 : panic(err)
540 : }
541 :
542 : // Grab and reference the current readState. This prevents the underlying
543 : // files in the associated version from being deleted if there is a current
544 : // compaction. The readState is unref'd by Iterator.Close().
545 1 : readState := d.loadReadState()
546 1 :
547 1 : // Determine the seqnum to read at after grabbing the read state (current and
548 1 : // memtables) above.
549 1 : var seqNum uint64
550 1 : if s != nil {
551 1 : seqNum = s.seqNum
552 1 : } else {
553 1 : seqNum = d.mu.versions.visibleSeqNum.Load()
554 1 : }
555 :
556 1 : buf := getIterAllocPool.Get().(*getIterAlloc)
557 1 :
558 1 : get := &buf.get
559 1 : *get = getIter{
560 1 : logger: d.opts.Logger,
561 1 : comparer: d.opts.Comparer,
562 1 : newIters: d.newIters,
563 1 : snapshot: seqNum,
564 1 : key: key,
565 1 : batch: b,
566 1 : mem: readState.memtables,
567 1 : l0: readState.current.L0SublevelFiles,
568 1 : version: readState.current,
569 1 : }
570 1 :
571 1 : // Strip off memtables which cannot possibly contain the seqNum being read
572 1 : // at.
573 1 : for len(get.mem) > 0 {
574 1 : n := len(get.mem)
575 1 : if logSeqNum := get.mem[n-1].logSeqNum; logSeqNum < seqNum {
576 1 : break
577 : }
578 1 : get.mem = get.mem[:n-1]
579 : }
580 :
581 1 : i := &buf.dbi
582 1 : pointIter := get
583 1 : *i = Iterator{
584 1 : ctx: context.Background(),
585 1 : getIterAlloc: buf,
586 1 : iter: pointIter,
587 1 : pointIter: pointIter,
588 1 : merge: d.merge,
589 1 : comparer: *d.opts.Comparer,
590 1 : readState: readState,
591 1 : keyBuf: buf.keyBuf,
592 1 : }
593 1 :
594 1 : if !i.First() {
595 1 : err := i.Close()
596 1 : if err != nil {
597 1 : return nil, nil, err
598 1 : }
599 1 : return nil, nil, ErrNotFound
600 : }
601 1 : return i.Value(), i, nil
602 : }
603 :
604 : // Set sets the value for the given key. It overwrites any previous value
605 : // for that key; a DB is not a multi-map.
606 : //
607 : // It is safe to modify the contents of the arguments after Set returns.
608 1 : func (d *DB) Set(key, value []byte, opts *WriteOptions) error {
609 1 : b := newBatch(d)
610 1 : _ = b.Set(key, value, opts)
611 1 : if err := d.Apply(b, opts); err != nil {
612 1 : return err
613 1 : }
614 : // Only release the batch on success.
615 1 : return b.Close()
616 : }
617 :
618 : // Delete deletes the value for the given key. Deletes are blind all will
619 : // succeed even if the given key does not exist.
620 : //
621 : // It is safe to modify the contents of the arguments after Delete returns.
622 1 : func (d *DB) Delete(key []byte, opts *WriteOptions) error {
623 1 : b := newBatch(d)
624 1 : _ = b.Delete(key, opts)
625 1 : if err := d.Apply(b, opts); err != nil {
626 1 : return err
627 1 : }
628 : // Only release the batch on success.
629 1 : return b.Close()
630 : }
631 :
632 : // DeleteSized behaves identically to Delete, but takes an additional
633 : // argument indicating the size of the value being deleted. DeleteSized
634 : // should be preferred when the caller has the expectation that there exists
635 : // a single internal KV pair for the key (eg, the key has not been
636 : // overwritten recently), and the caller knows the size of its value.
637 : //
638 : // DeleteSized will record the value size within the tombstone and use it to
639 : // inform compaction-picking heuristics which strive to reduce space
640 : // amplification in the LSM. This "calling your shot" mechanic allows the
641 : // storage engine to more accurately estimate and reduce space amplification.
642 : //
643 : // It is safe to modify the contents of the arguments after DeleteSized
644 : // returns.
645 1 : func (d *DB) DeleteSized(key []byte, valueSize uint32, opts *WriteOptions) error {
646 1 : b := newBatch(d)
647 1 : _ = b.DeleteSized(key, valueSize, opts)
648 1 : if err := d.Apply(b, opts); err != nil {
649 0 : return err
650 0 : }
651 : // Only release the batch on success.
652 1 : return b.Close()
653 : }
654 :
655 : // SingleDelete adds an action to the batch that single deletes the entry for key.
656 : // See Writer.SingleDelete for more details on the semantics of SingleDelete.
657 : //
658 : // It is safe to modify the contents of the arguments after SingleDelete returns.
659 1 : func (d *DB) SingleDelete(key []byte, opts *WriteOptions) error {
660 1 : b := newBatch(d)
661 1 : _ = b.SingleDelete(key, opts)
662 1 : if err := d.Apply(b, opts); err != nil {
663 0 : return err
664 0 : }
665 : // Only release the batch on success.
666 1 : return b.Close()
667 : }
668 :
669 : // DeleteRange deletes all of the keys (and values) in the range [start,end)
670 : // (inclusive on start, exclusive on end).
671 : //
672 : // It is safe to modify the contents of the arguments after DeleteRange
673 : // returns.
674 1 : func (d *DB) DeleteRange(start, end []byte, opts *WriteOptions) error {
675 1 : b := newBatch(d)
676 1 : _ = b.DeleteRange(start, end, opts)
677 1 : if err := d.Apply(b, opts); err != nil {
678 1 : return err
679 1 : }
680 : // Only release the batch on success.
681 1 : return b.Close()
682 : }
683 :
684 : // Merge adds an action to the DB that merges the value at key with the new
685 : // value. The details of the merge are dependent upon the configured merge
686 : // operator.
687 : //
688 : // It is safe to modify the contents of the arguments after Merge returns.
689 1 : func (d *DB) Merge(key, value []byte, opts *WriteOptions) error {
690 1 : b := newBatch(d)
691 1 : _ = b.Merge(key, value, opts)
692 1 : if err := d.Apply(b, opts); err != nil {
693 1 : return err
694 1 : }
695 : // Only release the batch on success.
696 1 : return b.Close()
697 : }
698 :
699 : // LogData adds the specified to the batch. The data will be written to the
700 : // WAL, but not added to memtables or sstables. Log data is never indexed,
701 : // which makes it useful for testing WAL performance.
702 : //
703 : // It is safe to modify the contents of the argument after LogData returns.
704 1 : func (d *DB) LogData(data []byte, opts *WriteOptions) error {
705 1 : b := newBatch(d)
706 1 : _ = b.LogData(data, opts)
707 1 : if err := d.Apply(b, opts); err != nil {
708 1 : return err
709 1 : }
710 : // Only release the batch on success.
711 1 : return b.Close()
712 : }
713 :
714 : // RangeKeySet sets a range key mapping the key range [start, end) at the MVCC
715 : // timestamp suffix to value. The suffix is optional. If any portion of the key
716 : // range [start, end) is already set by a range key with the same suffix value,
717 : // RangeKeySet overrides it.
718 : //
719 : // It is safe to modify the contents of the arguments after RangeKeySet returns.
720 1 : func (d *DB) RangeKeySet(start, end, suffix, value []byte, opts *WriteOptions) error {
721 1 : b := newBatch(d)
722 1 : _ = b.RangeKeySet(start, end, suffix, value, opts)
723 1 : if err := d.Apply(b, opts); err != nil {
724 0 : return err
725 0 : }
726 : // Only release the batch on success.
727 1 : return b.Close()
728 : }
729 :
730 : // RangeKeyUnset removes a range key mapping the key range [start, end) at the
731 : // MVCC timestamp suffix. The suffix may be omitted to remove an unsuffixed
732 : // range key. RangeKeyUnset only removes portions of range keys that fall within
733 : // the [start, end) key span, and only range keys with suffixes that exactly
734 : // match the unset suffix.
735 : //
736 : // It is safe to modify the contents of the arguments after RangeKeyUnset
737 : // returns.
738 1 : func (d *DB) RangeKeyUnset(start, end, suffix []byte, opts *WriteOptions) error {
739 1 : b := newBatch(d)
740 1 : _ = b.RangeKeyUnset(start, end, suffix, opts)
741 1 : if err := d.Apply(b, opts); err != nil {
742 0 : return err
743 0 : }
744 : // Only release the batch on success.
745 1 : return b.Close()
746 : }
747 :
748 : // RangeKeyDelete deletes all of the range keys in the range [start,end)
749 : // (inclusive on start, exclusive on end). It does not delete point keys (for
750 : // that use DeleteRange). RangeKeyDelete removes all range keys within the
751 : // bounds, including those with or without suffixes.
752 : //
753 : // It is safe to modify the contents of the arguments after RangeKeyDelete
754 : // returns.
755 1 : func (d *DB) RangeKeyDelete(start, end []byte, opts *WriteOptions) error {
756 1 : b := newBatch(d)
757 1 : _ = b.RangeKeyDelete(start, end, opts)
758 1 : if err := d.Apply(b, opts); err != nil {
759 0 : return err
760 0 : }
761 : // Only release the batch on success.
762 1 : return b.Close()
763 : }
764 :
765 : // Apply the operations contained in the batch to the DB. If the batch is large
766 : // the contents of the batch may be retained by the database. If that occurs
767 : // the batch contents will be cleared preventing the caller from attempting to
768 : // reuse them.
769 : //
770 : // It is safe to modify the contents of the arguments after Apply returns.
771 : //
772 : // Apply returns ErrInvalidBatch if the provided batch is invalid in any way.
773 1 : func (d *DB) Apply(batch *Batch, opts *WriteOptions) error {
774 1 : return d.applyInternal(batch, opts, false)
775 1 : }
776 :
777 : // ApplyNoSyncWait must only be used when opts.Sync is true and the caller
778 : // does not want to wait for the WAL fsync to happen. The method will return
779 : // once the mutation is applied to the memtable and is visible (note that a
780 : // mutation is visible before the WAL sync even in the wait case, so we have
781 : // not weakened the durability semantics). The caller must call Batch.SyncWait
782 : // to wait for the WAL fsync. The caller must not Close the batch without
783 : // first calling Batch.SyncWait.
784 : //
785 : // RECOMMENDATION: Prefer using Apply unless you really understand why you
786 : // need ApplyNoSyncWait.
787 : // EXPERIMENTAL: API/feature subject to change. Do not yet use outside
788 : // CockroachDB.
789 1 : func (d *DB) ApplyNoSyncWait(batch *Batch, opts *WriteOptions) error {
790 1 : if !opts.Sync {
791 0 : return errors.Errorf("cannot request asynchonous apply when WriteOptions.Sync is false")
792 0 : }
793 1 : return d.applyInternal(batch, opts, true)
794 : }
795 :
796 : // REQUIRES: noSyncWait => opts.Sync
797 1 : func (d *DB) applyInternal(batch *Batch, opts *WriteOptions, noSyncWait bool) error {
798 1 : if err := d.closed.Load(); err != nil {
799 1 : panic(err)
800 : }
801 1 : if batch.committing {
802 0 : panic("pebble: batch already committing")
803 : }
804 1 : if batch.applied.Load() {
805 0 : panic("pebble: batch already applied")
806 : }
807 1 : if d.opts.ReadOnly {
808 1 : return ErrReadOnly
809 1 : }
810 1 : if batch.db != nil && batch.db != d {
811 1 : panic(fmt.Sprintf("pebble: batch db mismatch: %p != %p", batch.db, d))
812 : }
813 :
814 1 : sync := opts.GetSync()
815 1 : if sync && d.opts.DisableWAL {
816 0 : return errors.New("pebble: WAL disabled")
817 0 : }
818 :
819 1 : if fmv := d.FormatMajorVersion(); fmv < batch.minimumFormatMajorVersion {
820 0 : panic(fmt.Sprintf(
821 0 : "pebble: batch requires at least format major version %d (current: %d)",
822 0 : batch.minimumFormatMajorVersion, fmv,
823 0 : ))
824 : }
825 :
826 1 : if batch.countRangeKeys > 0 {
827 1 : if d.split == nil {
828 0 : return errNoSplit
829 0 : }
830 : }
831 1 : batch.committing = true
832 1 :
833 1 : if batch.db == nil {
834 1 : if err := batch.refreshMemTableSize(); err != nil {
835 0 : return err
836 0 : }
837 : }
838 1 : if batch.memTableSize >= d.largeBatchThreshold {
839 1 : var err error
840 1 : batch.flushable, err = newFlushableBatch(batch, d.opts.Comparer)
841 1 : if err != nil {
842 0 : return err
843 0 : }
844 : }
845 1 : if err := d.commit.Commit(batch, sync, noSyncWait); err != nil {
846 0 : // There isn't much we can do on an error here. The commit pipeline will be
847 0 : // horked at this point.
848 0 : d.opts.Logger.Fatalf("pebble: fatal commit error: %v", err)
849 0 : }
850 : // If this is a large batch, we need to clear the batch contents as the
851 : // flushable batch may still be present in the flushables queue.
852 : //
853 : // TODO(peter): Currently large batches are written to the WAL. We could
854 : // skip the WAL write and instead wait for the large batch to be flushed to
855 : // an sstable. For a 100 MB batch, this might actually be faster. For a 1
856 : // GB batch this is almost certainly faster.
857 1 : if batch.flushable != nil {
858 1 : batch.data = nil
859 1 : }
860 1 : return nil
861 : }
862 :
863 1 : func (d *DB) commitApply(b *Batch, mem *memTable) error {
864 1 : if b.flushable != nil {
865 1 : // This is a large batch which was already added to the immutable queue.
866 1 : return nil
867 1 : }
868 1 : err := mem.apply(b, b.SeqNum())
869 1 : if err != nil {
870 0 : return err
871 0 : }
872 :
873 : // If the batch contains range tombstones and the database is configured
874 : // to flush range deletions, schedule a delayed flush so that disk space
875 : // may be reclaimed without additional writes or an explicit flush.
876 1 : if b.countRangeDels > 0 && d.opts.FlushDelayDeleteRange > 0 {
877 1 : d.mu.Lock()
878 1 : d.maybeScheduleDelayedFlush(mem, d.opts.FlushDelayDeleteRange)
879 1 : d.mu.Unlock()
880 1 : }
881 :
882 : // If the batch contains range keys and the database is configured to flush
883 : // range keys, schedule a delayed flush so that the range keys are cleared
884 : // from the memtable.
885 1 : if b.countRangeKeys > 0 && d.opts.FlushDelayRangeKey > 0 {
886 1 : d.mu.Lock()
887 1 : d.maybeScheduleDelayedFlush(mem, d.opts.FlushDelayRangeKey)
888 1 : d.mu.Unlock()
889 1 : }
890 :
891 1 : if mem.writerUnref() {
892 1 : d.mu.Lock()
893 1 : d.maybeScheduleFlush()
894 1 : d.mu.Unlock()
895 1 : }
896 1 : return nil
897 : }
898 :
899 1 : func (d *DB) commitWrite(b *Batch, syncWG *sync.WaitGroup, syncErr *error) (*memTable, error) {
900 1 : var size int64
901 1 : repr := b.Repr()
902 1 :
903 1 : if b.flushable != nil {
904 1 : // We have a large batch. Such batches are special in that they don't get
905 1 : // added to the memtable, and are instead inserted into the queue of
906 1 : // memtables. The call to makeRoomForWrite with this batch will force the
907 1 : // current memtable to be flushed. We want the large batch to be part of
908 1 : // the same log, so we add it to the WAL here, rather than after the call
909 1 : // to makeRoomForWrite().
910 1 : //
911 1 : // Set the sequence number since it was not set to the correct value earlier
912 1 : // (see comment in newFlushableBatch()).
913 1 : b.flushable.setSeqNum(b.SeqNum())
914 1 : if !d.opts.DisableWAL {
915 1 : var err error
916 1 : size, err = d.mu.log.writer.WriteRecord(repr, wal.SyncOptions{Done: syncWG, Err: syncErr}, b.refData)
917 1 : if err != nil {
918 0 : panic(err)
919 : }
920 : }
921 : }
922 :
923 1 : d.mu.Lock()
924 1 :
925 1 : var err error
926 1 : if !b.ingestedSSTBatch {
927 1 : // Batches which contain keys of kind InternalKeyKindIngestSST will
928 1 : // never be applied to the memtable, so we don't need to make room for
929 1 : // write. For the other cases, switch out the memtable if there was not
930 1 : // enough room to store the batch.
931 1 : err = d.makeRoomForWrite(b)
932 1 : }
933 :
934 1 : if err == nil && !d.opts.DisableWAL {
935 1 : d.mu.log.bytesIn += uint64(len(repr))
936 1 : }
937 :
938 : // Grab a reference to the memtable while holding DB.mu. Note that for
939 : // non-flushable batches (b.flushable == nil) makeRoomForWrite() added a
940 : // reference to the memtable which will prevent it from being flushed until
941 : // we unreference it. This reference is dropped in DB.commitApply().
942 1 : mem := d.mu.mem.mutable
943 1 :
944 1 : d.mu.Unlock()
945 1 : if err != nil {
946 0 : return nil, err
947 0 : }
948 :
949 1 : if d.opts.DisableWAL {
950 1 : return mem, nil
951 1 : }
952 :
953 1 : if b.flushable == nil {
954 1 : size, err = d.mu.log.writer.WriteRecord(repr, wal.SyncOptions{Done: syncWG, Err: syncErr}, b.refData)
955 1 : if err != nil {
956 0 : panic(err)
957 : }
958 : }
959 :
960 1 : d.logSize.Store(uint64(size))
961 1 : return mem, err
962 : }
963 :
964 : type iterAlloc struct {
965 : dbi Iterator
966 : keyBuf []byte
967 : boundsBuf [2][]byte
968 : prefixOrFullSeekKey []byte
969 : merging mergingIter
970 : mlevels [3 + numLevels]mergingIterLevel
971 : levels [3 + numLevels]levelIter
972 : levelsPositioned [3 + numLevels]bool
973 : }
974 :
975 : var iterAllocPool = sync.Pool{
976 1 : New: func() interface{} {
977 1 : return &iterAlloc{}
978 1 : },
979 : }
980 :
981 : // snapshotIterOpts denotes snapshot-related iterator options when calling
982 : // newIter. These are the possible cases for a snapshotIterOpts:
983 : // - No snapshot: All fields are zero values.
984 : // - Classic snapshot: Only `seqNum` is set. The latest readState will be used
985 : // and the specified seqNum will be used as the snapshot seqNum.
986 : // - EventuallyFileOnlySnapshot (EFOS) behaving as a classic snapshot. Only
987 : // the `seqNum` is set. The latest readState will be used
988 : // and the specified seqNum will be used as the snapshot seqNum.
989 : // - EFOS in file-only state: Only `seqNum` and `vers` are set. All the
990 : // relevant SSTs are referenced by the *version.
991 : // - EFOS that has been excised but is in alwaysCreateIters mode (tests only).
992 : // Only `seqNum` and `readState` are set.
993 : type snapshotIterOpts struct {
994 : seqNum uint64
995 : vers *version
996 : readState *readState
997 : }
998 :
999 : type batchIterOpts struct {
1000 : batchOnly bool
1001 : }
1002 : type newIterOpts struct {
1003 : snapshot snapshotIterOpts
1004 : batch batchIterOpts
1005 : }
1006 :
1007 : // newIter constructs a new iterator, merging in batch iterators as an extra
1008 : // level.
1009 : func (d *DB) newIter(
1010 : ctx context.Context, batch *Batch, internalOpts newIterOpts, o *IterOptions,
1011 1 : ) *Iterator {
1012 1 : if internalOpts.batch.batchOnly {
1013 1 : if batch == nil {
1014 0 : panic("batchOnly is true, but batch is nil")
1015 : }
1016 1 : if internalOpts.snapshot.vers != nil {
1017 0 : panic("batchOnly is true, but snapshotIterOpts is initialized")
1018 : }
1019 : }
1020 1 : if err := d.closed.Load(); err != nil {
1021 1 : panic(err)
1022 : }
1023 1 : seqNum := internalOpts.snapshot.seqNum
1024 1 : if o != nil && o.RangeKeyMasking.Suffix != nil && o.KeyTypes != IterKeyTypePointsAndRanges {
1025 0 : panic("pebble: range key masking requires IterKeyTypePointsAndRanges")
1026 : }
1027 1 : if (batch != nil || seqNum != 0) && (o != nil && o.OnlyReadGuaranteedDurable) {
1028 1 : // We could add support for OnlyReadGuaranteedDurable on snapshots if
1029 1 : // there was a need: this would require checking that the sequence number
1030 1 : // of the snapshot has been flushed, by comparing with
1031 1 : // DB.mem.queue[0].logSeqNum.
1032 1 : panic("OnlyReadGuaranteedDurable is not supported for batches or snapshots")
1033 : }
1034 1 : var readState *readState
1035 1 : var newIters tableNewIters
1036 1 : var newIterRangeKey keyspanimpl.TableNewSpanIter
1037 1 : if !internalOpts.batch.batchOnly {
1038 1 : // Grab and reference the current readState. This prevents the underlying
1039 1 : // files in the associated version from being deleted if there is a current
1040 1 : // compaction. The readState is unref'd by Iterator.Close().
1041 1 : if internalOpts.snapshot.vers == nil {
1042 1 : if internalOpts.snapshot.readState != nil {
1043 0 : readState = internalOpts.snapshot.readState
1044 0 : readState.ref()
1045 1 : } else {
1046 1 : // NB: loadReadState() calls readState.ref().
1047 1 : readState = d.loadReadState()
1048 1 : }
1049 1 : } else {
1050 1 : // vers != nil
1051 1 : internalOpts.snapshot.vers.Ref()
1052 1 : }
1053 :
1054 : // Determine the seqnum to read at after grabbing the read state (current and
1055 : // memtables) above.
1056 1 : if seqNum == 0 {
1057 1 : seqNum = d.mu.versions.visibleSeqNum.Load()
1058 1 : }
1059 1 : newIters = d.newIters
1060 1 : newIterRangeKey = d.tableNewRangeKeyIter
1061 : }
1062 :
1063 : // Bundle various structures under a single umbrella in order to allocate
1064 : // them together.
1065 1 : buf := iterAllocPool.Get().(*iterAlloc)
1066 1 : dbi := &buf.dbi
1067 1 : *dbi = Iterator{
1068 1 : ctx: ctx,
1069 1 : alloc: buf,
1070 1 : merge: d.merge,
1071 1 : comparer: *d.opts.Comparer,
1072 1 : readState: readState,
1073 1 : version: internalOpts.snapshot.vers,
1074 1 : keyBuf: buf.keyBuf,
1075 1 : prefixOrFullSeekKey: buf.prefixOrFullSeekKey,
1076 1 : boundsBuf: buf.boundsBuf,
1077 1 : batch: batch,
1078 1 : newIters: newIters,
1079 1 : newIterRangeKey: newIterRangeKey,
1080 1 : seqNum: seqNum,
1081 1 : batchOnlyIter: internalOpts.batch.batchOnly,
1082 1 : }
1083 1 : if o != nil {
1084 1 : dbi.opts = *o
1085 1 : dbi.processBounds(o.LowerBound, o.UpperBound)
1086 1 : }
1087 1 : dbi.opts.logger = d.opts.Logger
1088 1 : if d.opts.private.disableLazyCombinedIteration {
1089 0 : dbi.opts.disableLazyCombinedIteration = true
1090 0 : }
1091 1 : if batch != nil {
1092 1 : dbi.batchSeqNum = dbi.batch.nextSeqNum()
1093 1 : }
1094 1 : return finishInitializingIter(ctx, buf)
1095 : }
1096 :
1097 : // finishInitializingIter is a helper for doing the non-trivial initialization
1098 : // of an Iterator. It's invoked to perform the initial initialization of an
1099 : // Iterator during NewIter or Clone, and to perform reinitialization due to a
1100 : // change in IterOptions by a call to Iterator.SetOptions.
1101 1 : func finishInitializingIter(ctx context.Context, buf *iterAlloc) *Iterator {
1102 1 : // Short-hand.
1103 1 : dbi := &buf.dbi
1104 1 : var memtables flushableList
1105 1 : if dbi.readState != nil {
1106 1 : memtables = dbi.readState.memtables
1107 1 : }
1108 1 : if dbi.opts.OnlyReadGuaranteedDurable {
1109 1 : memtables = nil
1110 1 : } else {
1111 1 : // We only need to read from memtables which contain sequence numbers older
1112 1 : // than seqNum. Trim off newer memtables.
1113 1 : for i := len(memtables) - 1; i >= 0; i-- {
1114 1 : if logSeqNum := memtables[i].logSeqNum; logSeqNum < dbi.seqNum {
1115 1 : break
1116 : }
1117 1 : memtables = memtables[:i]
1118 : }
1119 : }
1120 :
1121 1 : if dbi.opts.pointKeys() {
1122 1 : // Construct the point iterator, initializing dbi.pointIter to point to
1123 1 : // dbi.merging. If this is called during a SetOptions call and this
1124 1 : // Iterator has already initialized dbi.merging, constructPointIter is a
1125 1 : // noop and an initialized pointIter already exists in dbi.pointIter.
1126 1 : dbi.constructPointIter(ctx, memtables, buf)
1127 1 : dbi.iter = dbi.pointIter
1128 1 : } else {
1129 1 : dbi.iter = emptyIter
1130 1 : }
1131 :
1132 1 : if dbi.opts.rangeKeys() {
1133 1 : dbi.rangeKeyMasking.init(dbi, dbi.comparer.Compare, dbi.comparer.Split)
1134 1 :
1135 1 : // When iterating over both point and range keys, don't create the
1136 1 : // range-key iterator stack immediately if we can avoid it. This
1137 1 : // optimization takes advantage of the expected sparseness of range
1138 1 : // keys, and configures the point-key iterator to dynamically switch to
1139 1 : // combined iteration when it observes a file containing range keys.
1140 1 : //
1141 1 : // Lazy combined iteration is not possible if a batch or a memtable
1142 1 : // contains any range keys.
1143 1 : useLazyCombinedIteration := dbi.rangeKey == nil &&
1144 1 : dbi.opts.KeyTypes == IterKeyTypePointsAndRanges &&
1145 1 : (dbi.batch == nil || dbi.batch.countRangeKeys == 0) &&
1146 1 : !dbi.opts.disableLazyCombinedIteration
1147 1 : if useLazyCombinedIteration {
1148 1 : // The user requested combined iteration, and there's no indexed
1149 1 : // batch currently containing range keys that would prevent lazy
1150 1 : // combined iteration. Check the memtables to see if they contain
1151 1 : // any range keys.
1152 1 : for i := range memtables {
1153 1 : if memtables[i].containsRangeKeys() {
1154 1 : useLazyCombinedIteration = false
1155 1 : break
1156 : }
1157 : }
1158 : }
1159 :
1160 1 : if useLazyCombinedIteration {
1161 1 : dbi.lazyCombinedIter = lazyCombinedIter{
1162 1 : parent: dbi,
1163 1 : pointIter: dbi.pointIter,
1164 1 : combinedIterState: combinedIterState{
1165 1 : initialized: false,
1166 1 : },
1167 1 : }
1168 1 : dbi.iter = &dbi.lazyCombinedIter
1169 1 : dbi.iter = invalidating.MaybeWrapIfInvariants(dbi.iter)
1170 1 : } else {
1171 1 : dbi.lazyCombinedIter.combinedIterState = combinedIterState{
1172 1 : initialized: true,
1173 1 : }
1174 1 : if dbi.rangeKey == nil {
1175 1 : dbi.rangeKey = iterRangeKeyStateAllocPool.Get().(*iteratorRangeKeyState)
1176 1 : dbi.rangeKey.init(dbi.comparer.Compare, dbi.comparer.Split, &dbi.opts)
1177 1 : dbi.constructRangeKeyIter()
1178 1 : } else {
1179 1 : dbi.rangeKey.iterConfig.SetBounds(dbi.opts.LowerBound, dbi.opts.UpperBound)
1180 1 : }
1181 :
1182 : // Wrap the point iterator (currently dbi.iter) with an interleaving
1183 : // iterator that interleaves range keys pulled from
1184 : // dbi.rangeKey.rangeKeyIter.
1185 : //
1186 : // NB: The interleaving iterator is always reinitialized, even if
1187 : // dbi already had an initialized range key iterator, in case the point
1188 : // iterator changed or the range key masking suffix changed.
1189 1 : dbi.rangeKey.iiter.Init(&dbi.comparer, dbi.iter, dbi.rangeKey.rangeKeyIter,
1190 1 : keyspan.InterleavingIterOpts{
1191 1 : Mask: &dbi.rangeKeyMasking,
1192 1 : LowerBound: dbi.opts.LowerBound,
1193 1 : UpperBound: dbi.opts.UpperBound,
1194 1 : })
1195 1 : dbi.iter = &dbi.rangeKey.iiter
1196 : }
1197 1 : } else {
1198 1 : // !dbi.opts.rangeKeys()
1199 1 : //
1200 1 : // Reset the combined iterator state. The initialized=true ensures the
1201 1 : // iterator doesn't unnecessarily try to switch to combined iteration.
1202 1 : dbi.lazyCombinedIter.combinedIterState = combinedIterState{initialized: true}
1203 1 : }
1204 1 : return dbi
1205 : }
1206 :
1207 : // ScanInternal scans all internal keys within the specified bounds, truncating
1208 : // any rangedels and rangekeys to those bounds if they span past them. For use
1209 : // when an external user needs to be aware of all internal keys that make up a
1210 : // key range.
1211 : //
1212 : // Keys deleted by range deletions must not be returned or exposed by this
1213 : // method, while the range deletion deleting that key must be exposed using
1214 : // visitRangeDel. Keys that would be masked by range key masking (if an
1215 : // appropriate prefix were set) should be exposed, alongside the range key
1216 : // that would have masked it. This method also collapses all point keys into
1217 : // one InternalKey; so only one internal key at most per user key is returned
1218 : // to visitPointKey.
1219 : //
1220 : // If visitSharedFile is not nil, ScanInternal iterates in skip-shared iteration
1221 : // mode. In this iteration mode, sstables in levels L5 and L6 are skipped, and
1222 : // their metadatas truncated to [lower, upper) and passed into visitSharedFile.
1223 : // ErrInvalidSkipSharedIteration is returned if visitSharedFile is not nil and an
1224 : // sstable in L5 or L6 is found that is not in shared storage according to
1225 : // provider.IsShared, or an sstable in those levels contains a newer key than the
1226 : // snapshot sequence number (only applicable for snapshot.ScanInternal). Examples
1227 : // of when this could happen could be if Pebble started writing sstables before a
1228 : // creator ID was set (as creator IDs are necessary to enable shared storage)
1229 : // resulting in some lower level SSTs being on non-shared storage. Skip-shared
1230 : // iteration is invalid in those cases.
1231 : func (d *DB) ScanInternal(
1232 : ctx context.Context,
1233 : categoryAndQoS sstable.CategoryAndQoS,
1234 : lower, upper []byte,
1235 : visitPointKey func(key *InternalKey, value LazyValue, iterInfo IteratorLevel) error,
1236 : visitRangeDel func(start, end []byte, seqNum uint64) error,
1237 : visitRangeKey func(start, end []byte, keys []rangekey.Key) error,
1238 : visitSharedFile func(sst *SharedSSTMeta) error,
1239 : visitExternalFile func(sst *ExternalFile) error,
1240 1 : ) error {
1241 1 : scanInternalOpts := &scanInternalOptions{
1242 1 : CategoryAndQoS: categoryAndQoS,
1243 1 : visitPointKey: visitPointKey,
1244 1 : visitRangeDel: visitRangeDel,
1245 1 : visitRangeKey: visitRangeKey,
1246 1 : visitSharedFile: visitSharedFile,
1247 1 : visitExternalFile: visitExternalFile,
1248 1 : IterOptions: IterOptions{
1249 1 : KeyTypes: IterKeyTypePointsAndRanges,
1250 1 : LowerBound: lower,
1251 1 : UpperBound: upper,
1252 1 : },
1253 1 : }
1254 1 : iter, err := d.newInternalIter(ctx, snapshotIterOpts{} /* snapshot */, scanInternalOpts)
1255 1 : if err != nil {
1256 0 : return err
1257 0 : }
1258 1 : defer iter.close()
1259 1 : return scanInternalImpl(ctx, lower, upper, iter, scanInternalOpts)
1260 : }
1261 :
1262 : // newInternalIter constructs and returns a new scanInternalIterator on this db.
1263 : // If o.skipSharedLevels is true, levels below sharedLevelsStart are *not* added
1264 : // to the internal iterator.
1265 : //
1266 : // TODO(bilal): This method has a lot of similarities with db.newIter as well as
1267 : // finishInitializingIter. Both pairs of methods should be refactored to reduce
1268 : // this duplication.
1269 : func (d *DB) newInternalIter(
1270 : ctx context.Context, sOpts snapshotIterOpts, o *scanInternalOptions,
1271 1 : ) (*scanInternalIterator, error) {
1272 1 : if err := d.closed.Load(); err != nil {
1273 0 : panic(err)
1274 : }
1275 : // Grab and reference the current readState. This prevents the underlying
1276 : // files in the associated version from being deleted if there is a current
1277 : // compaction. The readState is unref'd by Iterator.Close().
1278 1 : var readState *readState
1279 1 : if sOpts.vers == nil {
1280 1 : if sOpts.readState != nil {
1281 0 : readState = sOpts.readState
1282 0 : readState.ref()
1283 1 : } else {
1284 1 : readState = d.loadReadState()
1285 1 : }
1286 : }
1287 1 : if sOpts.vers != nil {
1288 1 : sOpts.vers.Ref()
1289 1 : }
1290 :
1291 : // Determine the seqnum to read at after grabbing the read state (current and
1292 : // memtables) above.
1293 1 : seqNum := sOpts.seqNum
1294 1 : if seqNum == 0 {
1295 1 : seqNum = d.mu.versions.visibleSeqNum.Load()
1296 1 : }
1297 :
1298 : // Bundle various structures under a single umbrella in order to allocate
1299 : // them together.
1300 1 : buf := iterAllocPool.Get().(*iterAlloc)
1301 1 : dbi := &scanInternalIterator{
1302 1 : ctx: ctx,
1303 1 : db: d,
1304 1 : comparer: d.opts.Comparer,
1305 1 : merge: d.opts.Merger.Merge,
1306 1 : readState: readState,
1307 1 : version: sOpts.vers,
1308 1 : alloc: buf,
1309 1 : newIters: d.newIters,
1310 1 : newIterRangeKey: d.tableNewRangeKeyIter,
1311 1 : seqNum: seqNum,
1312 1 : mergingIter: &buf.merging,
1313 1 : }
1314 1 : dbi.opts = *o
1315 1 : dbi.opts.logger = d.opts.Logger
1316 1 : if d.opts.private.disableLazyCombinedIteration {
1317 0 : dbi.opts.disableLazyCombinedIteration = true
1318 0 : }
1319 1 : return finishInitializingInternalIter(buf, dbi)
1320 : }
1321 :
1322 : func finishInitializingInternalIter(
1323 : buf *iterAlloc, i *scanInternalIterator,
1324 1 : ) (*scanInternalIterator, error) {
1325 1 : // Short-hand.
1326 1 : var memtables flushableList
1327 1 : if i.readState != nil {
1328 1 : memtables = i.readState.memtables
1329 1 : }
1330 : // We only need to read from memtables which contain sequence numbers older
1331 : // than seqNum. Trim off newer memtables.
1332 1 : for j := len(memtables) - 1; j >= 0; j-- {
1333 1 : if logSeqNum := memtables[j].logSeqNum; logSeqNum < i.seqNum {
1334 1 : break
1335 : }
1336 1 : memtables = memtables[:j]
1337 : }
1338 1 : i.initializeBoundBufs(i.opts.LowerBound, i.opts.UpperBound)
1339 1 :
1340 1 : if err := i.constructPointIter(i.opts.CategoryAndQoS, memtables, buf); err != nil {
1341 0 : return nil, err
1342 0 : }
1343 :
1344 : // For internal iterators, we skip the lazy combined iteration optimization
1345 : // entirely, and create the range key iterator stack directly.
1346 1 : i.rangeKey = iterRangeKeyStateAllocPool.Get().(*iteratorRangeKeyState)
1347 1 : i.rangeKey.init(i.comparer.Compare, i.comparer.Split, &i.opts.IterOptions)
1348 1 : if err := i.constructRangeKeyIter(); err != nil {
1349 0 : return nil, err
1350 0 : }
1351 :
1352 : // Wrap the point iterator (currently i.iter) with an interleaving
1353 : // iterator that interleaves range keys pulled from
1354 : // i.rangeKey.rangeKeyIter.
1355 1 : i.rangeKey.iiter.Init(i.comparer, i.iter, i.rangeKey.rangeKeyIter,
1356 1 : keyspan.InterleavingIterOpts{
1357 1 : LowerBound: i.opts.LowerBound,
1358 1 : UpperBound: i.opts.UpperBound,
1359 1 : })
1360 1 : i.iter = &i.rangeKey.iiter
1361 1 :
1362 1 : return i, nil
1363 : }
1364 :
1365 : func (i *Iterator) constructPointIter(
1366 : ctx context.Context, memtables flushableList, buf *iterAlloc,
1367 1 : ) {
1368 1 : if i.pointIter != nil {
1369 1 : // Already have one.
1370 1 : return
1371 1 : }
1372 1 : internalOpts := internalIterOpts{stats: &i.stats.InternalStats}
1373 1 : if i.opts.RangeKeyMasking.Filter != nil {
1374 1 : internalOpts.boundLimitedFilter = &i.rangeKeyMasking
1375 1 : }
1376 :
1377 : // Merging levels and levels from iterAlloc.
1378 1 : mlevels := buf.mlevels[:0]
1379 1 : levels := buf.levels[:0]
1380 1 :
1381 1 : // We compute the number of levels needed ahead of time and reallocate a slice if
1382 1 : // the array from the iterAlloc isn't large enough. Doing this allocation once
1383 1 : // should improve the performance.
1384 1 : numMergingLevels := 0
1385 1 : numLevelIters := 0
1386 1 : if i.batch != nil {
1387 1 : numMergingLevels++
1388 1 : }
1389 :
1390 1 : var current *version
1391 1 : if !i.batchOnlyIter {
1392 1 : numMergingLevels += len(memtables)
1393 1 :
1394 1 : current = i.version
1395 1 : if current == nil {
1396 1 : current = i.readState.current
1397 1 : }
1398 1 : numMergingLevels += len(current.L0SublevelFiles)
1399 1 : numLevelIters += len(current.L0SublevelFiles)
1400 1 : for level := 1; level < len(current.Levels); level++ {
1401 1 : if current.Levels[level].Empty() {
1402 1 : continue
1403 : }
1404 1 : numMergingLevels++
1405 1 : numLevelIters++
1406 : }
1407 : }
1408 :
1409 1 : if numMergingLevels > cap(mlevels) {
1410 1 : mlevels = make([]mergingIterLevel, 0, numMergingLevels)
1411 1 : }
1412 1 : if numLevelIters > cap(levels) {
1413 1 : levels = make([]levelIter, 0, numLevelIters)
1414 1 : }
1415 :
1416 : // Top-level is the batch, if any.
1417 1 : if i.batch != nil {
1418 1 : if i.batch.index == nil {
1419 0 : // This isn't an indexed batch. We shouldn't have gotten this far.
1420 0 : panic(errors.AssertionFailedf("creating an iterator over an unindexed batch"))
1421 1 : } else {
1422 1 : i.batch.initInternalIter(&i.opts, &i.batchPointIter)
1423 1 : i.batch.initRangeDelIter(&i.opts, &i.batchRangeDelIter, i.batchSeqNum)
1424 1 : // Only include the batch's rangedel iterator if it's non-empty.
1425 1 : // This requires some subtle logic in the case a rangedel is later
1426 1 : // written to the batch and the view of the batch is refreshed
1427 1 : // during a call to SetOptions—in this case, we need to reconstruct
1428 1 : // the point iterator to add the batch rangedel iterator.
1429 1 : var rangeDelIter keyspan.FragmentIterator
1430 1 : if i.batchRangeDelIter.Count() > 0 {
1431 1 : rangeDelIter = &i.batchRangeDelIter
1432 1 : }
1433 1 : mlevels = append(mlevels, mergingIterLevel{
1434 1 : iter: &i.batchPointIter,
1435 1 : rangeDelIter: rangeDelIter,
1436 1 : })
1437 : }
1438 : }
1439 :
1440 1 : if !i.batchOnlyIter {
1441 1 : // Next are the memtables.
1442 1 : for j := len(memtables) - 1; j >= 0; j-- {
1443 1 : mem := memtables[j]
1444 1 : mlevels = append(mlevels, mergingIterLevel{
1445 1 : iter: mem.newIter(&i.opts),
1446 1 : rangeDelIter: mem.newRangeDelIter(&i.opts),
1447 1 : })
1448 1 : }
1449 :
1450 : // Next are the file levels: L0 sub-levels followed by lower levels.
1451 1 : mlevelsIndex := len(mlevels)
1452 1 : levelsIndex := len(levels)
1453 1 : mlevels = mlevels[:numMergingLevels]
1454 1 : levels = levels[:numLevelIters]
1455 1 : i.opts.snapshotForHideObsoletePoints = buf.dbi.seqNum
1456 1 : addLevelIterForFiles := func(files manifest.LevelIterator, level manifest.Level) {
1457 1 : li := &levels[levelsIndex]
1458 1 :
1459 1 : li.init(ctx, i.opts, &i.comparer, i.newIters, files, level, internalOpts)
1460 1 : li.initRangeDel(&mlevels[mlevelsIndex].rangeDelIter)
1461 1 : li.initBoundaryContext(&mlevels[mlevelsIndex].levelIterBoundaryContext)
1462 1 : li.initCombinedIterState(&i.lazyCombinedIter.combinedIterState)
1463 1 : mlevels[mlevelsIndex].levelIter = li
1464 1 : mlevels[mlevelsIndex].iter = invalidating.MaybeWrapIfInvariants(li)
1465 1 :
1466 1 : levelsIndex++
1467 1 : mlevelsIndex++
1468 1 : }
1469 :
1470 : // Add level iterators for the L0 sublevels, iterating from newest to
1471 : // oldest.
1472 1 : for i := len(current.L0SublevelFiles) - 1; i >= 0; i-- {
1473 1 : addLevelIterForFiles(current.L0SublevelFiles[i].Iter(), manifest.L0Sublevel(i))
1474 1 : }
1475 :
1476 : // Add level iterators for the non-empty non-L0 levels.
1477 1 : for level := 1; level < len(current.Levels); level++ {
1478 1 : if current.Levels[level].Empty() {
1479 1 : continue
1480 : }
1481 1 : addLevelIterForFiles(current.Levels[level].Iter(), manifest.Level(level))
1482 : }
1483 : }
1484 1 : buf.merging.init(&i.opts, &i.stats.InternalStats, i.comparer.Compare, i.comparer.Split, mlevels...)
1485 1 : if len(mlevels) <= cap(buf.levelsPositioned) {
1486 1 : buf.merging.levelsPositioned = buf.levelsPositioned[:len(mlevels)]
1487 1 : }
1488 1 : buf.merging.snapshot = i.seqNum
1489 1 : buf.merging.batchSnapshot = i.batchSeqNum
1490 1 : buf.merging.combinedIterState = &i.lazyCombinedIter.combinedIterState
1491 1 : i.pointIter = invalidating.MaybeWrapIfInvariants(&buf.merging).(topLevelIterator)
1492 1 : i.merging = &buf.merging
1493 : }
1494 :
1495 : // NewBatch returns a new empty write-only batch. Any reads on the batch will
1496 : // return an error. If the batch is committed it will be applied to the DB.
1497 1 : func (d *DB) NewBatch() *Batch {
1498 1 : return newBatch(d)
1499 1 : }
1500 :
1501 : // NewBatchWithSize is mostly identical to NewBatch, but it will allocate the
1502 : // the specified memory space for the internal slice in advance.
1503 0 : func (d *DB) NewBatchWithSize(size int) *Batch {
1504 0 : return newBatchWithSize(d, size)
1505 0 : }
1506 :
1507 : // NewIndexedBatch returns a new empty read-write batch. Any reads on the batch
1508 : // will read from both the batch and the DB. If the batch is committed it will
1509 : // be applied to the DB. An indexed batch is slower that a non-indexed batch
1510 : // for insert operations. If you do not need to perform reads on the batch, use
1511 : // NewBatch instead.
1512 1 : func (d *DB) NewIndexedBatch() *Batch {
1513 1 : return newIndexedBatch(d, d.opts.Comparer)
1514 1 : }
1515 :
1516 : // NewIndexedBatchWithSize is mostly identical to NewIndexedBatch, but it will
1517 : // allocate the the specified memory space for the internal slice in advance.
1518 0 : func (d *DB) NewIndexedBatchWithSize(size int) *Batch {
1519 0 : return newIndexedBatchWithSize(d, d.opts.Comparer, size)
1520 0 : }
1521 :
1522 : // NewIter returns an iterator that is unpositioned (Iterator.Valid() will
1523 : // return false). The iterator can be positioned via a call to SeekGE, SeekLT,
1524 : // First or Last. The iterator provides a point-in-time view of the current DB
1525 : // state. This view is maintained by preventing file deletions and preventing
1526 : // memtables referenced by the iterator from being deleted. Using an iterator
1527 : // to maintain a long-lived point-in-time view of the DB state can lead to an
1528 : // apparent memory and disk usage leak. Use snapshots (see NewSnapshot) for
1529 : // point-in-time snapshots which avoids these problems.
1530 1 : func (d *DB) NewIter(o *IterOptions) (*Iterator, error) {
1531 1 : return d.NewIterWithContext(context.Background(), o)
1532 1 : }
1533 :
1534 : // NewIterWithContext is like NewIter, and additionally accepts a context for
1535 : // tracing.
1536 1 : func (d *DB) NewIterWithContext(ctx context.Context, o *IterOptions) (*Iterator, error) {
1537 1 : return d.newIter(ctx, nil /* batch */, newIterOpts{}, o), nil
1538 1 : }
1539 :
1540 : // NewSnapshot returns a point-in-time view of the current DB state. Iterators
1541 : // created with this handle will all observe a stable snapshot of the current
1542 : // DB state. The caller must call Snapshot.Close() when the snapshot is no
1543 : // longer needed. Snapshots are not persisted across DB restarts (close ->
1544 : // open). Unlike the implicit snapshot maintained by an iterator, a snapshot
1545 : // will not prevent memtables from being released or sstables from being
1546 : // deleted. Instead, a snapshot prevents deletion of sequence numbers
1547 : // referenced by the snapshot.
1548 1 : func (d *DB) NewSnapshot() *Snapshot {
1549 1 : if err := d.closed.Load(); err != nil {
1550 1 : panic(err)
1551 : }
1552 :
1553 1 : d.mu.Lock()
1554 1 : s := &Snapshot{
1555 1 : db: d,
1556 1 : seqNum: d.mu.versions.visibleSeqNum.Load(),
1557 1 : }
1558 1 : d.mu.snapshots.pushBack(s)
1559 1 : d.mu.Unlock()
1560 1 : return s
1561 : }
1562 :
1563 : // NewEventuallyFileOnlySnapshot returns a point-in-time view of the current DB
1564 : // state, similar to NewSnapshot, but with consistency constrained to the
1565 : // provided set of key ranges. See the comment at EventuallyFileOnlySnapshot for
1566 : // its semantics.
1567 1 : func (d *DB) NewEventuallyFileOnlySnapshot(keyRanges []KeyRange) *EventuallyFileOnlySnapshot {
1568 1 : if err := d.closed.Load(); err != nil {
1569 0 : panic(err)
1570 : }
1571 1 : for i := range keyRanges {
1572 1 : if i > 0 && d.cmp(keyRanges[i-1].End, keyRanges[i].Start) > 0 {
1573 0 : panic("pebble: key ranges for eventually-file-only-snapshot not in order")
1574 : }
1575 : }
1576 1 : return d.makeEventuallyFileOnlySnapshot(keyRanges)
1577 : }
1578 :
1579 : // Close closes the DB.
1580 : //
1581 : // It is not safe to close a DB until all outstanding iterators are closed
1582 : // or to call Close concurrently with any other DB method. It is not valid
1583 : // to call any of a DB's methods after the DB has been closed.
1584 1 : func (d *DB) Close() error {
1585 1 : // Lock the commit pipeline for the duration of Close. This prevents a race
1586 1 : // with makeRoomForWrite. Rotating the WAL in makeRoomForWrite requires
1587 1 : // dropping d.mu several times for I/O. If Close only holds d.mu, an
1588 1 : // in-progress WAL rotation may re-acquire d.mu only once the database is
1589 1 : // closed.
1590 1 : //
1591 1 : // Additionally, locking the commit pipeline makes it more likely that
1592 1 : // (illegal) concurrent writes will observe d.closed.Load() != nil, creating
1593 1 : // more understable panics if the database is improperly used concurrently
1594 1 : // during Close.
1595 1 : d.commit.mu.Lock()
1596 1 : defer d.commit.mu.Unlock()
1597 1 : d.mu.Lock()
1598 1 : defer d.mu.Unlock()
1599 1 : if err := d.closed.Load(); err != nil {
1600 1 : panic(err)
1601 : }
1602 :
1603 : // Clear the finalizer that is used to check that an unreferenced DB has been
1604 : // closed. We're closing the DB here, so the check performed by that
1605 : // finalizer isn't necessary.
1606 : //
1607 : // Note: this is a no-op if invariants are disabled or race is enabled.
1608 1 : invariants.SetFinalizer(d.closed, nil)
1609 1 :
1610 1 : d.closed.Store(errors.WithStack(ErrClosed))
1611 1 : close(d.closedCh)
1612 1 :
1613 1 : defer d.opts.Cache.Unref()
1614 1 :
1615 1 : for d.mu.compact.compactingCount > 0 || d.mu.compact.downloadingCount > 0 || d.mu.compact.flushing {
1616 1 : d.mu.compact.cond.Wait()
1617 1 : }
1618 1 : for d.mu.tableStats.loading {
1619 1 : d.mu.tableStats.cond.Wait()
1620 1 : }
1621 1 : for d.mu.tableValidation.validating {
1622 0 : d.mu.tableValidation.cond.Wait()
1623 0 : }
1624 :
1625 1 : var err error
1626 1 : if n := len(d.mu.compact.inProgress); n > 0 {
1627 1 : err = errors.Errorf("pebble: %d unexpected in-progress compactions", errors.Safe(n))
1628 1 : }
1629 1 : err = firstError(err, d.mu.formatVers.marker.Close())
1630 1 : err = firstError(err, d.tableCache.close())
1631 1 : if !d.opts.ReadOnly {
1632 1 : if d.mu.log.writer != nil {
1633 1 : _, err2 := d.mu.log.writer.Close()
1634 1 : err = firstError(err, err2)
1635 1 : }
1636 1 : } else if d.mu.log.writer != nil {
1637 0 : panic("pebble: log-writer should be nil in read-only mode")
1638 : }
1639 1 : err = firstError(err, d.mu.log.manager.Close())
1640 1 : err = firstError(err, d.fileLock.Close())
1641 1 :
1642 1 : // Note that versionSet.close() only closes the MANIFEST. The versions list
1643 1 : // is still valid for the checks below.
1644 1 : err = firstError(err, d.mu.versions.close())
1645 1 :
1646 1 : err = firstError(err, d.dataDir.Close())
1647 1 :
1648 1 : d.readState.val.unrefLocked()
1649 1 :
1650 1 : current := d.mu.versions.currentVersion()
1651 1 : for v := d.mu.versions.versions.Front(); true; v = v.Next() {
1652 1 : refs := v.Refs()
1653 1 : if v == current {
1654 1 : if refs != 1 {
1655 1 : err = firstError(err, errors.Errorf("leaked iterators: current\n%s", v))
1656 1 : }
1657 1 : break
1658 : }
1659 0 : if refs != 0 {
1660 0 : err = firstError(err, errors.Errorf("leaked iterators:\n%s", v))
1661 0 : }
1662 : }
1663 :
1664 1 : for _, mem := range d.mu.mem.queue {
1665 1 : // Usually, we'd want to delete the files returned by readerUnref. But
1666 1 : // in this case, even if we're unreferencing the flushables, the
1667 1 : // flushables aren't obsolete. They will be reconstructed during WAL
1668 1 : // replay.
1669 1 : mem.readerUnrefLocked(false)
1670 1 : }
1671 : // If there's an unused, recycled memtable, we need to release its memory.
1672 1 : if obsoleteMemTable := d.memTableRecycle.Swap(nil); obsoleteMemTable != nil {
1673 1 : d.freeMemTable(obsoleteMemTable)
1674 1 : }
1675 1 : if reserved := d.memTableReserved.Load(); reserved != 0 {
1676 1 : err = firstError(err, errors.Errorf("leaked memtable reservation: %d", errors.Safe(reserved)))
1677 1 : }
1678 :
1679 : // Since we called d.readState.val.unrefLocked() above, we are expected to
1680 : // manually schedule deletion of obsolete files.
1681 1 : if len(d.mu.versions.obsoleteTables) > 0 {
1682 1 : d.deleteObsoleteFiles(d.newJobIDLocked())
1683 1 : }
1684 :
1685 1 : d.mu.Unlock()
1686 1 : d.compactionSchedulers.Wait()
1687 1 :
1688 1 : // Wait for all cleaning jobs to finish.
1689 1 : d.cleanupManager.Close()
1690 1 :
1691 1 : // Sanity check metrics.
1692 1 : if invariants.Enabled {
1693 1 : m := d.Metrics()
1694 1 : if m.Compact.NumInProgress > 0 || m.Compact.InProgressBytes > 0 {
1695 0 : d.mu.Lock()
1696 0 : panic(fmt.Sprintf("invalid metrics on close:\n%s", m))
1697 : }
1698 : }
1699 :
1700 1 : d.mu.Lock()
1701 1 :
1702 1 : // As a sanity check, ensure that there are no zombie tables. A non-zero count
1703 1 : // hints at a reference count leak.
1704 1 : if ztbls := len(d.mu.versions.zombieTables); ztbls > 0 {
1705 0 : err = firstError(err, errors.Errorf("non-zero zombie file count: %d", ztbls))
1706 0 : }
1707 :
1708 1 : err = firstError(err, d.objProvider.Close())
1709 1 :
1710 1 : // If the options include a closer to 'close' the filesystem, close it.
1711 1 : if d.opts.private.fsCloser != nil {
1712 1 : d.opts.private.fsCloser.Close()
1713 1 : }
1714 :
1715 : // Return an error if the user failed to close all open snapshots.
1716 1 : if v := d.mu.snapshots.count(); v > 0 {
1717 0 : err = firstError(err, errors.Errorf("leaked snapshots: %d open snapshots on DB %p", v, d))
1718 0 : }
1719 :
1720 1 : return err
1721 : }
1722 :
1723 : // Compact the specified range of keys in the database.
1724 1 : func (d *DB) Compact(start, end []byte, parallelize bool) error {
1725 1 : if err := d.closed.Load(); err != nil {
1726 1 : panic(err)
1727 : }
1728 1 : if d.opts.ReadOnly {
1729 1 : return ErrReadOnly
1730 1 : }
1731 1 : if d.cmp(start, end) >= 0 {
1732 1 : return errors.Errorf("Compact start %s is not less than end %s",
1733 1 : d.opts.Comparer.FormatKey(start), d.opts.Comparer.FormatKey(end))
1734 1 : }
1735 :
1736 1 : d.mu.Lock()
1737 1 : maxLevelWithFiles := 1
1738 1 : cur := d.mu.versions.currentVersion()
1739 1 : for level := 0; level < numLevels; level++ {
1740 1 : overlaps := cur.Overlaps(level, base.UserKeyBoundsInclusive(start, end))
1741 1 : if !overlaps.Empty() {
1742 1 : maxLevelWithFiles = level + 1
1743 1 : }
1744 : }
1745 :
1746 : // Determine if any memtable overlaps with the compaction range. We wait for
1747 : // any such overlap to flush (initiating a flush if necessary).
1748 1 : mem, err := func() (*flushableEntry, error) {
1749 1 : // Check to see if any files overlap with any of the memtables. The queue
1750 1 : // is ordered from oldest to newest with the mutable memtable being the
1751 1 : // last element in the slice. We want to wait for the newest table that
1752 1 : // overlaps.
1753 1 : for i := len(d.mu.mem.queue) - 1; i >= 0; i-- {
1754 1 : mem := d.mu.mem.queue[i]
1755 1 : var anyOverlaps bool
1756 1 : mem.computePossibleOverlaps(func(b bounded) shouldContinue {
1757 1 : anyOverlaps = true
1758 1 : return stopIteration
1759 1 : }, KeyRange{Start: start, End: end})
1760 1 : if !anyOverlaps {
1761 1 : continue
1762 : }
1763 1 : var err error
1764 1 : if mem.flushable == d.mu.mem.mutable {
1765 1 : // We have to hold both commitPipeline.mu and DB.mu when calling
1766 1 : // makeRoomForWrite(). Lock order requirements elsewhere force us to
1767 1 : // unlock DB.mu in order to grab commitPipeline.mu first.
1768 1 : d.mu.Unlock()
1769 1 : d.commit.mu.Lock()
1770 1 : d.mu.Lock()
1771 1 : defer d.commit.mu.Unlock()
1772 1 : if mem.flushable == d.mu.mem.mutable {
1773 1 : // Only flush if the active memtable is unchanged.
1774 1 : err = d.makeRoomForWrite(nil)
1775 1 : }
1776 : }
1777 1 : mem.flushForced = true
1778 1 : d.maybeScheduleFlush()
1779 1 : return mem, err
1780 : }
1781 1 : return nil, nil
1782 : }()
1783 :
1784 1 : d.mu.Unlock()
1785 1 :
1786 1 : if err != nil {
1787 0 : return err
1788 0 : }
1789 1 : if mem != nil {
1790 1 : <-mem.flushed
1791 1 : }
1792 :
1793 1 : for level := 0; level < maxLevelWithFiles; {
1794 1 : for {
1795 1 : if err := d.manualCompact(
1796 1 : start, end, level, parallelize); err != nil {
1797 1 : if errors.Is(err, ErrCancelledCompaction) {
1798 1 : continue
1799 : }
1800 1 : return err
1801 : }
1802 1 : break
1803 : }
1804 1 : level++
1805 1 : if level == numLevels-1 {
1806 1 : // A manual compaction of the bottommost level occurred.
1807 1 : // There is no next level to try and compact.
1808 1 : break
1809 : }
1810 : }
1811 1 : return nil
1812 : }
1813 :
1814 1 : func (d *DB) manualCompact(start, end []byte, level int, parallelize bool) error {
1815 1 : d.mu.Lock()
1816 1 : curr := d.mu.versions.currentVersion()
1817 1 : files := curr.Overlaps(level, base.UserKeyBoundsInclusive(start, end))
1818 1 : if files.Empty() {
1819 1 : d.mu.Unlock()
1820 1 : return nil
1821 1 : }
1822 :
1823 1 : var compactions []*manualCompaction
1824 1 : if parallelize {
1825 1 : compactions = append(compactions, d.splitManualCompaction(start, end, level)...)
1826 1 : } else {
1827 1 : compactions = append(compactions, &manualCompaction{
1828 1 : level: level,
1829 1 : done: make(chan error, 1),
1830 1 : start: start,
1831 1 : end: end,
1832 1 : })
1833 1 : }
1834 1 : d.mu.compact.manual = append(d.mu.compact.manual, compactions...)
1835 1 : d.maybeScheduleCompaction()
1836 1 : d.mu.Unlock()
1837 1 :
1838 1 : // Each of the channels is guaranteed to be eventually sent to once. After a
1839 1 : // compaction is possibly picked in d.maybeScheduleCompaction(), either the
1840 1 : // compaction is dropped, executed after being scheduled, or retried later.
1841 1 : // Assuming eventual progress when a compaction is retried, all outcomes send
1842 1 : // a value to the done channel. Since the channels are buffered, it is not
1843 1 : // necessary to read from each channel, and so we can exit early in the event
1844 1 : // of an error.
1845 1 : for _, compaction := range compactions {
1846 1 : if err := <-compaction.done; err != nil {
1847 1 : return err
1848 1 : }
1849 : }
1850 1 : return nil
1851 : }
1852 :
1853 : // splitManualCompaction splits a manual compaction over [start,end] on level
1854 : // such that the resulting compactions have no key overlap.
1855 : func (d *DB) splitManualCompaction(
1856 : start, end []byte, level int,
1857 1 : ) (splitCompactions []*manualCompaction) {
1858 1 : curr := d.mu.versions.currentVersion()
1859 1 : endLevel := level + 1
1860 1 : baseLevel := d.mu.versions.picker.getBaseLevel()
1861 1 : if level == 0 {
1862 1 : endLevel = baseLevel
1863 1 : }
1864 1 : keyRanges := curr.CalculateInuseKeyRanges(level, endLevel, start, end)
1865 1 : for _, keyRange := range keyRanges {
1866 1 : splitCompactions = append(splitCompactions, &manualCompaction{
1867 1 : level: level,
1868 1 : done: make(chan error, 1),
1869 1 : start: keyRange.Start,
1870 1 : end: keyRange.End,
1871 1 : split: true,
1872 1 : })
1873 1 : }
1874 1 : return splitCompactions
1875 : }
1876 :
1877 : // DownloadSpan is a key range passed to the Download method.
1878 : type DownloadSpan struct {
1879 : StartKey []byte
1880 : // EndKey is exclusive.
1881 : EndKey []byte
1882 : // ViaBackingFileDownload, if true, indicates the span should be downloaded by
1883 : // downloading any remote backing files byte-for-byte and replacing them with
1884 : // the downloaded local files, while otherwise leaving the virtual SSTables
1885 : // as-is. If false, a "normal" rewriting compaction of the span, that iterates
1886 : // the keys and produces a new SSTable, is used instead. Downloading raw files
1887 : // can be faster when the whole file is being downloaded, as it avoids some
1888 : // cpu-intensive steps involved in iteration and new file construction such as
1889 : // compression, however it can also be wasteful when only a small portion of a
1890 : // larger backing file is being used by a virtual file. Additionally, if the
1891 : // virtual file has expensive read-time transformations, such as prefix
1892 : // replacement, rewriting once can persist the result of these for future use
1893 : // while copying only the backing file will obligate future reads to continue
1894 : // to compute such transforms.
1895 : ViaBackingFileDownload bool
1896 : }
1897 :
1898 1 : func (d *DB) downloadSpan(ctx context.Context, span DownloadSpan) error {
1899 1 : dSpan := &downloadSpan{
1900 1 : start: span.StartKey,
1901 1 : end: span.EndKey,
1902 1 : // Protected by d.mu.
1903 1 : doneChans: make([]chan error, 1),
1904 1 : kind: compactionKindRewrite,
1905 1 : }
1906 1 : if span.ViaBackingFileDownload {
1907 1 : dSpan.kind = compactionKindCopy
1908 1 : }
1909 1 : dSpan.doneChans[0] = make(chan error, 1)
1910 1 : doneChan := dSpan.doneChans[0]
1911 1 : compactionIdx := 0
1912 1 :
1913 1 : func() {
1914 1 : d.mu.Lock()
1915 1 : defer d.mu.Unlock()
1916 1 :
1917 1 : d.mu.compact.downloads = append(d.mu.compact.downloads, dSpan)
1918 1 : d.maybeScheduleCompaction()
1919 1 : }()
1920 :
1921 : // Requires d.mu to be held.
1922 1 : noExternalFilesInSpan := func() (noExternalFiles bool) {
1923 1 : vers := d.mu.versions.currentVersion()
1924 1 :
1925 1 : for i := 0; i < len(vers.Levels); i++ {
1926 1 : if vers.Levels[i].Empty() {
1927 1 : continue
1928 : }
1929 1 : overlap := vers.Overlaps(i, base.UserKeyBoundsEndExclusive(span.StartKey, span.EndKey))
1930 1 : foundExternalFile := false
1931 1 : overlap.Each(func(metadata *manifest.FileMetadata) {
1932 1 : objMeta, err := d.objProvider.Lookup(fileTypeTable, metadata.FileBacking.DiskFileNum)
1933 1 : if err != nil {
1934 0 : return
1935 0 : }
1936 1 : if objMeta.IsExternal() {
1937 1 : foundExternalFile = true
1938 1 : }
1939 : })
1940 1 : if foundExternalFile {
1941 1 : return false
1942 1 : }
1943 : }
1944 1 : return true
1945 : }
1946 :
1947 : // Requires d.mu to be held.
1948 1 : removeUsFromList := func() {
1949 1 : // Check where we are in d.mu.compact.downloads. Remove us from the
1950 1 : // list.
1951 1 : for i := range d.mu.compact.downloads {
1952 0 : if d.mu.compact.downloads[i] != dSpan {
1953 0 : continue
1954 : }
1955 0 : copy(d.mu.compact.downloads[i:], d.mu.compact.downloads[i+1:])
1956 0 : d.mu.compact.downloads = d.mu.compact.downloads[:len(d.mu.compact.downloads)-1]
1957 0 : break
1958 : }
1959 : }
1960 :
1961 1 : d.mu.Lock()
1962 1 : defer d.mu.Unlock()
1963 1 :
1964 1 : for {
1965 1 : // Non-blocking select.
1966 1 : select {
1967 0 : case <-ctx.Done():
1968 0 : removeUsFromList()
1969 0 : return ctx.Err()
1970 :
1971 1 : case err := <-doneChan:
1972 1 : if err != nil {
1973 0 : removeUsFromList()
1974 0 : return err
1975 0 : }
1976 : // Grab the next doneCh to wait on.
1977 1 : compactionIdx++
1978 1 : doneChan = dSpan.doneChans[compactionIdx]
1979 1 : continue
1980 :
1981 1 : default:
1982 : }
1983 :
1984 : // It's possible to have downloaded all files without getting notifications
1985 : // on all doneChans. This is expected if there are a significant amount of
1986 : // overlapping writes that schedule regular, non-download compactions.
1987 : //
1988 : // TODO(radu): if the span overlaps many files, checking this after every
1989 : // compaction completes might be expensive.
1990 1 : if noExternalFilesInSpan() {
1991 1 : removeUsFromList()
1992 1 : return nil
1993 1 : }
1994 :
1995 1 : d.maybeScheduleCompaction()
1996 1 :
1997 1 : // If no compactions/download are running (and none were scheduled above),
1998 1 : // we cannot wait on the condition.
1999 1 : if d.mu.compact.compactingCount == 0 && d.mu.compact.downloadingCount == 0 {
2000 0 : if d.closed.Load() != nil {
2001 0 : return nil
2002 0 : }
2003 : // It is possible that the last compaction just updated the version but
2004 : // did not yet update the compacting state on the external file(s). Sleep
2005 : // for a bit and try again.
2006 : // TODO(radu): investigate this case further.
2007 0 : d.mu.Unlock()
2008 0 : time.Sleep(100 * time.Microsecond)
2009 0 : d.mu.Lock()
2010 0 : continue
2011 : }
2012 :
2013 : // Wait for a compaction/download to complete and recheck.
2014 1 : d.mu.compact.cond.Wait()
2015 : }
2016 : }
2017 :
2018 : // Download ensures that the LSM does not use any external sstables for the
2019 : // given key ranges. It does so by performing appropriate compactions so that
2020 : // all external data becomes available locally.
2021 : //
2022 : // Note that calling this method does not imply that all other compactions stop;
2023 : // it simply informs Pebble of a list of spans for which external data should be
2024 : // downloaded with high priority.
2025 : //
2026 : // The method returns once no external sstasbles overlap the given spans, the
2027 : // context is canceled, the db is closed, or an error is hit.
2028 : //
2029 : // TODO(radu): consider passing a priority/impact knob to express how important
2030 : // the download is (versus live traffic performance, LSM health).
2031 1 : func (d *DB) Download(ctx context.Context, spans []DownloadSpan) error {
2032 1 : ctx, cancel := context.WithCancel(ctx)
2033 1 : defer cancel()
2034 1 : if err := d.closed.Load(); err != nil {
2035 0 : panic(err)
2036 : }
2037 1 : if d.opts.ReadOnly {
2038 0 : return ErrReadOnly
2039 0 : }
2040 1 : for i := range spans {
2041 1 : if err := ctx.Err(); err != nil {
2042 0 : return err
2043 0 : }
2044 1 : if err := d.downloadSpan(ctx, spans[i]); err != nil {
2045 0 : return err
2046 0 : }
2047 : }
2048 1 : return nil
2049 : }
2050 :
2051 : // Flush the memtable to stable storage.
2052 1 : func (d *DB) Flush() error {
2053 1 : flushDone, err := d.AsyncFlush()
2054 1 : if err != nil {
2055 1 : return err
2056 1 : }
2057 1 : <-flushDone
2058 1 : return nil
2059 : }
2060 :
2061 : // AsyncFlush asynchronously flushes the memtable to stable storage.
2062 : //
2063 : // If no error is returned, the caller can receive from the returned channel in
2064 : // order to wait for the flush to complete.
2065 1 : func (d *DB) AsyncFlush() (<-chan struct{}, error) {
2066 1 : if err := d.closed.Load(); err != nil {
2067 1 : panic(err)
2068 : }
2069 1 : if d.opts.ReadOnly {
2070 1 : return nil, ErrReadOnly
2071 1 : }
2072 :
2073 1 : d.commit.mu.Lock()
2074 1 : defer d.commit.mu.Unlock()
2075 1 : d.mu.Lock()
2076 1 : defer d.mu.Unlock()
2077 1 : flushed := d.mu.mem.queue[len(d.mu.mem.queue)-1].flushed
2078 1 : err := d.makeRoomForWrite(nil)
2079 1 : if err != nil {
2080 0 : return nil, err
2081 0 : }
2082 1 : return flushed, nil
2083 : }
2084 :
2085 : // Metrics returns metrics about the database.
2086 1 : func (d *DB) Metrics() *Metrics {
2087 1 : metrics := &Metrics{}
2088 1 : walStats := d.mu.log.manager.Stats()
2089 1 :
2090 1 : d.mu.Lock()
2091 1 : vers := d.mu.versions.currentVersion()
2092 1 : *metrics = d.mu.versions.metrics
2093 1 : metrics.Compact.EstimatedDebt = d.mu.versions.picker.estimatedCompactionDebt(0)
2094 1 : metrics.Compact.InProgressBytes = d.mu.versions.atomicInProgressBytes.Load()
2095 1 : // TODO(radu): split this to separate the download compactions.
2096 1 : metrics.Compact.NumInProgress = int64(d.mu.compact.compactingCount + d.mu.compact.downloadingCount)
2097 1 : metrics.Compact.MarkedFiles = vers.Stats.MarkedForCompaction
2098 1 : metrics.Compact.Duration = d.mu.compact.duration
2099 1 : for c := range d.mu.compact.inProgress {
2100 1 : if c.kind != compactionKindFlush {
2101 1 : metrics.Compact.Duration += d.timeNow().Sub(c.beganAt)
2102 1 : }
2103 : }
2104 :
2105 1 : for _, m := range d.mu.mem.queue {
2106 1 : metrics.MemTable.Size += m.totalBytes()
2107 1 : }
2108 1 : metrics.Snapshots.Count = d.mu.snapshots.count()
2109 1 : if metrics.Snapshots.Count > 0 {
2110 0 : metrics.Snapshots.EarliestSeqNum = d.mu.snapshots.earliest()
2111 0 : }
2112 1 : metrics.Snapshots.PinnedKeys = d.mu.snapshots.cumulativePinnedCount
2113 1 : metrics.Snapshots.PinnedSize = d.mu.snapshots.cumulativePinnedSize
2114 1 : metrics.MemTable.Count = int64(len(d.mu.mem.queue))
2115 1 : metrics.MemTable.ZombieCount = d.memTableCount.Load() - metrics.MemTable.Count
2116 1 : metrics.MemTable.ZombieSize = uint64(d.memTableReserved.Load()) - metrics.MemTable.Size
2117 1 : metrics.WAL.ObsoleteFiles = int64(walStats.ObsoleteFileCount)
2118 1 : metrics.WAL.ObsoletePhysicalSize = walStats.ObsoleteFileSize
2119 1 : metrics.WAL.Files = int64(walStats.LiveFileCount)
2120 1 : // The current WAL's size (d.logSize) is the logical size, which may be less
2121 1 : // than the WAL's physical size if it was recycled. walStats.LiveFileSize
2122 1 : // includes the physical size of all live WALs, but for the current WAL it
2123 1 : // reflects the physical size when it was opened. So it is possible that
2124 1 : // d.atomic.logSize has exceeded that physical size. We allow for this
2125 1 : // anomaly.
2126 1 : metrics.WAL.PhysicalSize = walStats.LiveFileSize
2127 1 : metrics.WAL.BytesIn = d.mu.log.bytesIn // protected by d.mu
2128 1 : metrics.WAL.Size = d.logSize.Load()
2129 1 : for i, n := 0, len(d.mu.mem.queue)-1; i < n; i++ {
2130 1 : metrics.WAL.Size += d.mu.mem.queue[i].logSize
2131 1 : }
2132 1 : metrics.WAL.BytesWritten = metrics.Levels[0].BytesIn + metrics.WAL.Size
2133 1 : metrics.WAL.Failover = walStats.Failover
2134 1 :
2135 1 : if p := d.mu.versions.picker; p != nil {
2136 1 : compactions := d.getInProgressCompactionInfoLocked(nil)
2137 1 : for level, score := range p.getScores(compactions) {
2138 1 : metrics.Levels[level].Score = score
2139 1 : }
2140 : }
2141 1 : metrics.Table.ZombieCount = int64(len(d.mu.versions.zombieTables))
2142 1 : for _, info := range d.mu.versions.zombieTables {
2143 1 : metrics.Table.ZombieSize += info.FileSize
2144 1 : if info.isLocal {
2145 1 : metrics.Table.Local.ZombieSize += info.FileSize
2146 1 : }
2147 : }
2148 1 : metrics.private.optionsFileSize = d.optionsFileSize
2149 1 :
2150 1 : // TODO(jackson): Consider making these metrics optional.
2151 1 : metrics.Keys.RangeKeySetsCount = countRangeKeySetFragments(vers)
2152 1 : metrics.Keys.TombstoneCount = countTombstones(vers)
2153 1 :
2154 1 : d.mu.versions.logLock()
2155 1 : metrics.private.manifestFileSize = uint64(d.mu.versions.manifest.Size())
2156 1 : backingCount, backingTotalSize := d.mu.versions.virtualBackings.Stats()
2157 1 : metrics.Table.BackingTableCount = uint64(backingCount)
2158 1 : metrics.Table.BackingTableSize = backingTotalSize
2159 1 : d.mu.versions.logUnlock()
2160 1 :
2161 1 : metrics.LogWriter.FsyncLatency = d.mu.log.metrics.fsyncLatency
2162 1 : if err := metrics.LogWriter.Merge(&d.mu.log.metrics.LogWriterMetrics); err != nil {
2163 0 : d.opts.Logger.Errorf("metrics error: %s", err)
2164 0 : }
2165 1 : metrics.Flush.WriteThroughput = d.mu.compact.flushWriteThroughput
2166 1 : if d.mu.compact.flushing {
2167 1 : metrics.Flush.NumInProgress = 1
2168 1 : }
2169 1 : for i := 0; i < numLevels; i++ {
2170 1 : metrics.Levels[i].Additional.ValueBlocksSize = valueBlocksSizeForLevel(vers, i)
2171 1 : }
2172 :
2173 1 : d.mu.Unlock()
2174 1 :
2175 1 : metrics.BlockCache = d.opts.Cache.Metrics()
2176 1 : metrics.TableCache, metrics.Filter = d.tableCache.metrics()
2177 1 : metrics.TableIters = int64(d.tableCache.iterCount())
2178 1 : metrics.CategoryStats = d.tableCache.dbOpts.sstStatsCollector.GetStats()
2179 1 :
2180 1 : metrics.SecondaryCacheMetrics = d.objProvider.Metrics()
2181 1 :
2182 1 : metrics.Uptime = d.timeNow().Sub(d.openedAt)
2183 1 :
2184 1 : return metrics
2185 : }
2186 :
2187 : // sstablesOptions hold the optional parameters to retrieve TableInfo for all sstables.
2188 : type sstablesOptions struct {
2189 : // set to true will return the sstable properties in TableInfo
2190 : withProperties bool
2191 :
2192 : // if set, return sstables that overlap the key range (end-exclusive)
2193 : start []byte
2194 : end []byte
2195 :
2196 : withApproximateSpanBytes bool
2197 : }
2198 :
2199 : // SSTablesOption set optional parameter used by `DB.SSTables`.
2200 : type SSTablesOption func(*sstablesOptions)
2201 :
2202 : // WithProperties enable return sstable properties in each TableInfo.
2203 : //
2204 : // NOTE: if most of the sstable properties need to be read from disk,
2205 : // this options may make method `SSTables` quite slow.
2206 1 : func WithProperties() SSTablesOption {
2207 1 : return func(opt *sstablesOptions) {
2208 1 : opt.withProperties = true
2209 1 : }
2210 : }
2211 :
2212 : // WithKeyRangeFilter ensures returned sstables overlap start and end (end-exclusive)
2213 : // if start and end are both nil these properties have no effect.
2214 1 : func WithKeyRangeFilter(start, end []byte) SSTablesOption {
2215 1 : return func(opt *sstablesOptions) {
2216 1 : opt.end = end
2217 1 : opt.start = start
2218 1 : }
2219 : }
2220 :
2221 : // WithApproximateSpanBytes enables capturing the approximate number of bytes that
2222 : // overlap the provided key span for each sstable.
2223 : // NOTE: this option can only be used with WithKeyRangeFilter and WithProperties
2224 : // provided.
2225 1 : func WithApproximateSpanBytes() SSTablesOption {
2226 1 : return func(opt *sstablesOptions) {
2227 1 : opt.withApproximateSpanBytes = true
2228 1 : }
2229 : }
2230 :
2231 : // BackingType denotes the type of storage backing a given sstable.
2232 : type BackingType int
2233 :
2234 : const (
2235 : // BackingTypeLocal denotes an sstable stored on local disk according to the
2236 : // objprovider. This file is completely owned by us.
2237 : BackingTypeLocal BackingType = iota
2238 : // BackingTypeShared denotes an sstable stored on shared storage, created
2239 : // by this Pebble instance and possibly shared by other Pebble instances.
2240 : // These types of files have lifecycle managed by Pebble.
2241 : BackingTypeShared
2242 : // BackingTypeSharedForeign denotes an sstable stored on shared storage,
2243 : // created by a Pebble instance other than this one. These types of files have
2244 : // lifecycle managed by Pebble.
2245 : BackingTypeSharedForeign
2246 : // BackingTypeExternal denotes an sstable stored on external storage,
2247 : // not owned by any Pebble instance and with no refcounting/cleanup methods
2248 : // or lifecycle management. An example of an external file is a file restored
2249 : // from a backup.
2250 : BackingTypeExternal
2251 : )
2252 :
2253 : // SSTableInfo export manifest.TableInfo with sstable.Properties alongside
2254 : // other file backing info.
2255 : type SSTableInfo struct {
2256 : manifest.TableInfo
2257 : // Virtual indicates whether the sstable is virtual.
2258 : Virtual bool
2259 : // BackingSSTNum is the disk file number associated with the backing sstable.
2260 : // If Virtual is false, BackingSSTNum == PhysicalTableDiskFileNum(FileNum).
2261 : BackingSSTNum base.DiskFileNum
2262 : // BackingType is the type of storage backing this sstable.
2263 : BackingType BackingType
2264 : // Locator is the remote.Locator backing this sstable, if the backing type is
2265 : // not BackingTypeLocal.
2266 : Locator remote.Locator
2267 :
2268 : // Properties is the sstable properties of this table. If Virtual is true,
2269 : // then the Properties are associated with the backing sst.
2270 : Properties *sstable.Properties
2271 : }
2272 :
2273 : // SSTables retrieves the current sstables. The returned slice is indexed by
2274 : // level and each level is indexed by the position of the sstable within the
2275 : // level. Note that this information may be out of date due to concurrent
2276 : // flushes and compactions.
2277 1 : func (d *DB) SSTables(opts ...SSTablesOption) ([][]SSTableInfo, error) {
2278 1 : opt := &sstablesOptions{}
2279 1 : for _, fn := range opts {
2280 1 : fn(opt)
2281 1 : }
2282 :
2283 1 : if opt.withApproximateSpanBytes && !opt.withProperties {
2284 1 : return nil, errors.Errorf("Cannot use WithApproximateSpanBytes without WithProperties option.")
2285 1 : }
2286 1 : if opt.withApproximateSpanBytes && (opt.start == nil || opt.end == nil) {
2287 1 : return nil, errors.Errorf("Cannot use WithApproximateSpanBytes without WithKeyRangeFilter option.")
2288 1 : }
2289 :
2290 : // Grab and reference the current readState.
2291 1 : readState := d.loadReadState()
2292 1 : defer readState.unref()
2293 1 :
2294 1 : // TODO(peter): This is somewhat expensive, especially on a large
2295 1 : // database. It might be worthwhile to unify TableInfo and FileMetadata and
2296 1 : // then we could simply return current.Files. Note that RocksDB is doing
2297 1 : // something similar to the current code, so perhaps it isn't too bad.
2298 1 : srcLevels := readState.current.Levels
2299 1 : var totalTables int
2300 1 : for i := range srcLevels {
2301 1 : totalTables += srcLevels[i].Len()
2302 1 : }
2303 :
2304 1 : destTables := make([]SSTableInfo, totalTables)
2305 1 : destLevels := make([][]SSTableInfo, len(srcLevels))
2306 1 : for i := range destLevels {
2307 1 : iter := srcLevels[i].Iter()
2308 1 : j := 0
2309 1 : for m := iter.First(); m != nil; m = iter.Next() {
2310 1 : if opt.start != nil && opt.end != nil {
2311 1 : b := base.UserKeyBoundsEndExclusive(opt.start, opt.end)
2312 1 : if !m.Overlaps(d.opts.Comparer.Compare, &b) {
2313 1 : continue
2314 : }
2315 : }
2316 1 : destTables[j] = SSTableInfo{TableInfo: m.TableInfo()}
2317 1 : if opt.withProperties {
2318 1 : p, err := d.tableCache.getTableProperties(
2319 1 : m,
2320 1 : )
2321 1 : if err != nil {
2322 0 : return nil, err
2323 0 : }
2324 1 : destTables[j].Properties = p
2325 : }
2326 1 : destTables[j].Virtual = m.Virtual
2327 1 : destTables[j].BackingSSTNum = m.FileBacking.DiskFileNum
2328 1 : objMeta, err := d.objProvider.Lookup(fileTypeTable, m.FileBacking.DiskFileNum)
2329 1 : if err != nil {
2330 0 : return nil, err
2331 0 : }
2332 1 : if objMeta.IsRemote() {
2333 0 : if objMeta.IsShared() {
2334 0 : if d.objProvider.IsSharedForeign(objMeta) {
2335 0 : destTables[j].BackingType = BackingTypeSharedForeign
2336 0 : } else {
2337 0 : destTables[j].BackingType = BackingTypeShared
2338 0 : }
2339 0 : } else {
2340 0 : destTables[j].BackingType = BackingTypeExternal
2341 0 : }
2342 0 : destTables[j].Locator = objMeta.Remote.Locator
2343 1 : } else {
2344 1 : destTables[j].BackingType = BackingTypeLocal
2345 1 : }
2346 :
2347 1 : if opt.withApproximateSpanBytes {
2348 1 : var spanBytes uint64
2349 1 : if m.ContainedWithinSpan(d.opts.Comparer.Compare, opt.start, opt.end) {
2350 0 : spanBytes = m.Size
2351 1 : } else {
2352 1 : size, err := d.tableCache.estimateSize(m, opt.start, opt.end)
2353 1 : if err != nil {
2354 0 : return nil, err
2355 0 : }
2356 1 : spanBytes = size
2357 : }
2358 1 : propertiesCopy := *destTables[j].Properties
2359 1 :
2360 1 : // Deep copy user properties so approximate span bytes can be added.
2361 1 : propertiesCopy.UserProperties = make(map[string]string, len(destTables[j].Properties.UserProperties)+1)
2362 1 : for k, v := range destTables[j].Properties.UserProperties {
2363 0 : propertiesCopy.UserProperties[k] = v
2364 0 : }
2365 1 : propertiesCopy.UserProperties["approximate-span-bytes"] = strconv.FormatUint(spanBytes, 10)
2366 1 : destTables[j].Properties = &propertiesCopy
2367 : }
2368 1 : j++
2369 : }
2370 1 : destLevels[i] = destTables[:j]
2371 1 : destTables = destTables[j:]
2372 : }
2373 :
2374 1 : return destLevels, nil
2375 : }
2376 :
2377 : // EstimateDiskUsage returns the estimated filesystem space used in bytes for
2378 : // storing the range `[start, end]`. The estimation is computed as follows:
2379 : //
2380 : // - For sstables fully contained in the range the whole file size is included.
2381 : // - For sstables partially contained in the range the overlapping data block sizes
2382 : // are included. Even if a data block partially overlaps, or we cannot determine
2383 : // overlap due to abbreviated index keys, the full data block size is included in
2384 : // the estimation. Note that unlike fully contained sstables, none of the
2385 : // meta-block space is counted for partially overlapped files.
2386 : // - For virtual sstables, we use the overlap between start, end and the virtual
2387 : // sstable bounds to determine disk usage.
2388 : // - There may also exist WAL entries for unflushed keys in this range. This
2389 : // estimation currently excludes space used for the range in the WAL.
2390 1 : func (d *DB) EstimateDiskUsage(start, end []byte) (uint64, error) {
2391 1 : bytes, _, _, err := d.EstimateDiskUsageByBackingType(start, end)
2392 1 : return bytes, err
2393 1 : }
2394 :
2395 : // EstimateDiskUsageByBackingType is like EstimateDiskUsage but additionally
2396 : // returns the subsets of that size in remote ane external files.
2397 : func (d *DB) EstimateDiskUsageByBackingType(
2398 : start, end []byte,
2399 1 : ) (totalSize, remoteSize, externalSize uint64, _ error) {
2400 1 : if err := d.closed.Load(); err != nil {
2401 0 : panic(err)
2402 : }
2403 1 : if d.opts.Comparer.Compare(start, end) > 0 {
2404 0 : return 0, 0, 0, errors.New("invalid key-range specified (start > end)")
2405 0 : }
2406 :
2407 : // Grab and reference the current readState. This prevents the underlying
2408 : // files in the associated version from being deleted if there is a concurrent
2409 : // compaction.
2410 1 : readState := d.loadReadState()
2411 1 : defer readState.unref()
2412 1 :
2413 1 : for level, files := range readState.current.Levels {
2414 1 : iter := files.Iter()
2415 1 : if level > 0 {
2416 1 : // We can only use `Overlaps` to restrict `files` at L1+ since at L0 it
2417 1 : // expands the range iteratively until it has found a set of files that
2418 1 : // do not overlap any other L0 files outside that set.
2419 1 : overlaps := readState.current.Overlaps(level, base.UserKeyBoundsInclusive(start, end))
2420 1 : iter = overlaps.Iter()
2421 1 : }
2422 1 : for file := iter.First(); file != nil; file = iter.Next() {
2423 1 : if d.opts.Comparer.Compare(start, file.Smallest.UserKey) <= 0 &&
2424 1 : d.opts.Comparer.Compare(file.Largest.UserKey, end) <= 0 {
2425 1 : // The range fully contains the file, so skip looking it up in
2426 1 : // table cache/looking at its indexes, and add the full file size.
2427 1 : meta, err := d.objProvider.Lookup(fileTypeTable, file.FileBacking.DiskFileNum)
2428 1 : if err != nil {
2429 0 : return 0, 0, 0, err
2430 0 : }
2431 1 : if meta.IsRemote() {
2432 0 : remoteSize += file.Size
2433 0 : if meta.Remote.CleanupMethod == objstorage.SharedNoCleanup {
2434 0 : externalSize += file.Size
2435 0 : }
2436 : }
2437 1 : totalSize += file.Size
2438 1 : } else if d.opts.Comparer.Compare(file.Smallest.UserKey, end) <= 0 &&
2439 1 : d.opts.Comparer.Compare(start, file.Largest.UserKey) <= 0 {
2440 1 : var size uint64
2441 1 : var err error
2442 1 : if file.Virtual {
2443 0 : err = d.tableCache.withVirtualReader(
2444 0 : file.VirtualMeta(),
2445 0 : func(r sstable.VirtualReader) (err error) {
2446 0 : size, err = r.EstimateDiskUsage(start, end)
2447 0 : return err
2448 0 : },
2449 : )
2450 1 : } else {
2451 1 : err = d.tableCache.withReader(
2452 1 : file.PhysicalMeta(),
2453 1 : func(r *sstable.Reader) (err error) {
2454 1 : size, err = r.EstimateDiskUsage(start, end)
2455 1 : return err
2456 1 : },
2457 : )
2458 : }
2459 1 : if err != nil {
2460 0 : return 0, 0, 0, err
2461 0 : }
2462 1 : meta, err := d.objProvider.Lookup(fileTypeTable, file.FileBacking.DiskFileNum)
2463 1 : if err != nil {
2464 0 : return 0, 0, 0, err
2465 0 : }
2466 1 : if meta.IsRemote() {
2467 0 : remoteSize += size
2468 0 : if meta.Remote.CleanupMethod == objstorage.SharedNoCleanup {
2469 0 : externalSize += size
2470 0 : }
2471 : }
2472 1 : totalSize += size
2473 : }
2474 : }
2475 : }
2476 1 : return totalSize, remoteSize, externalSize, nil
2477 : }
2478 :
2479 1 : func (d *DB) walPreallocateSize() int {
2480 1 : // Set the WAL preallocate size to 110% of the memtable size. Note that there
2481 1 : // is a bit of apples and oranges in units here as the memtabls size
2482 1 : // corresponds to the memory usage of the memtable while the WAL size is the
2483 1 : // size of the batches (plus overhead) stored in the WAL.
2484 1 : //
2485 1 : // TODO(peter): 110% of the memtable size is quite hefty for a block
2486 1 : // size. This logic is taken from GetWalPreallocateBlockSize in
2487 1 : // RocksDB. Could a smaller preallocation block size be used?
2488 1 : size := d.opts.MemTableSize
2489 1 : size = (size / 10) + size
2490 1 : return int(size)
2491 1 : }
2492 :
2493 1 : func (d *DB) newMemTable(logNum base.DiskFileNum, logSeqNum uint64) (*memTable, *flushableEntry) {
2494 1 : size := d.mu.mem.nextSize
2495 1 : if d.mu.mem.nextSize < d.opts.MemTableSize {
2496 1 : d.mu.mem.nextSize *= 2
2497 1 : if d.mu.mem.nextSize > d.opts.MemTableSize {
2498 1 : d.mu.mem.nextSize = d.opts.MemTableSize
2499 1 : }
2500 : }
2501 :
2502 1 : memtblOpts := memTableOptions{
2503 1 : Options: d.opts,
2504 1 : logSeqNum: logSeqNum,
2505 1 : }
2506 1 :
2507 1 : // Before attempting to allocate a new memtable, check if there's one
2508 1 : // available for recycling in memTableRecycle. Large contiguous allocations
2509 1 : // can be costly as fragmentation makes it more difficult to find a large
2510 1 : // contiguous free space. We've observed 64MB allocations taking 10ms+.
2511 1 : //
2512 1 : // To reduce these costly allocations, up to 1 obsolete memtable is stashed
2513 1 : // in `d.memTableRecycle` to allow a future memtable rotation to reuse
2514 1 : // existing memory.
2515 1 : var mem *memTable
2516 1 : mem = d.memTableRecycle.Swap(nil)
2517 1 : if mem != nil && uint64(len(mem.arenaBuf)) != size {
2518 1 : d.freeMemTable(mem)
2519 1 : mem = nil
2520 1 : }
2521 1 : if mem != nil {
2522 1 : // Carry through the existing buffer and memory reservation.
2523 1 : memtblOpts.arenaBuf = mem.arenaBuf
2524 1 : memtblOpts.releaseAccountingReservation = mem.releaseAccountingReservation
2525 1 : } else {
2526 1 : mem = new(memTable)
2527 1 : memtblOpts.arenaBuf = manual.New(int(size))
2528 1 : memtblOpts.releaseAccountingReservation = d.opts.Cache.Reserve(int(size))
2529 1 : d.memTableCount.Add(1)
2530 1 : d.memTableReserved.Add(int64(size))
2531 1 :
2532 1 : // Note: this is a no-op if invariants are disabled or race is enabled.
2533 1 : invariants.SetFinalizer(mem, checkMemTable)
2534 1 : }
2535 1 : mem.init(memtblOpts)
2536 1 :
2537 1 : entry := d.newFlushableEntry(mem, logNum, logSeqNum)
2538 1 : entry.releaseMemAccounting = func() {
2539 1 : // If the user leaks iterators, we may be releasing the memtable after
2540 1 : // the DB is already closed. In this case, we want to just release the
2541 1 : // memory because DB.Close won't come along to free it for us.
2542 1 : if err := d.closed.Load(); err != nil {
2543 1 : d.freeMemTable(mem)
2544 1 : return
2545 1 : }
2546 :
2547 : // The next memtable allocation might be able to reuse this memtable.
2548 : // Stash it on d.memTableRecycle.
2549 1 : if unusedMem := d.memTableRecycle.Swap(mem); unusedMem != nil {
2550 1 : // There was already a memtable waiting to be recycled. We're now
2551 1 : // responsible for freeing it.
2552 1 : d.freeMemTable(unusedMem)
2553 1 : }
2554 : }
2555 1 : return mem, entry
2556 : }
2557 :
2558 1 : func (d *DB) freeMemTable(m *memTable) {
2559 1 : d.memTableCount.Add(-1)
2560 1 : d.memTableReserved.Add(-int64(len(m.arenaBuf)))
2561 1 : m.free()
2562 1 : }
2563 :
2564 : func (d *DB) newFlushableEntry(
2565 : f flushable, logNum base.DiskFileNum, logSeqNum uint64,
2566 1 : ) *flushableEntry {
2567 1 : fe := &flushableEntry{
2568 1 : flushable: f,
2569 1 : flushed: make(chan struct{}),
2570 1 : logNum: logNum,
2571 1 : logSeqNum: logSeqNum,
2572 1 : deleteFn: d.mu.versions.addObsolete,
2573 1 : deleteFnLocked: d.mu.versions.addObsoleteLocked,
2574 1 : }
2575 1 : fe.readerRefs.Store(1)
2576 1 : return fe
2577 1 : }
2578 :
2579 : // makeRoomForWrite ensures that the memtable has room to hold the contents of
2580 : // Batch. It reserves the space in the memtable and adds a reference to the
2581 : // memtable. The caller must later ensure that the memtable is unreferenced. If
2582 : // the memtable is full, or a nil Batch is provided, the current memtable is
2583 : // rotated (marked as immutable) and a new mutable memtable is allocated. This
2584 : // memtable rotation also causes a log rotation.
2585 : //
2586 : // Both DB.mu and commitPipeline.mu must be held by the caller. Note that DB.mu
2587 : // may be released and reacquired.
2588 1 : func (d *DB) makeRoomForWrite(b *Batch) error {
2589 1 : if b != nil && b.ingestedSSTBatch {
2590 0 : panic("pebble: invalid function call")
2591 : }
2592 :
2593 1 : force := b == nil || b.flushable != nil
2594 1 : stalled := false
2595 1 : for {
2596 1 : if b != nil && b.flushable == nil {
2597 1 : err := d.mu.mem.mutable.prepare(b)
2598 1 : if err != arenaskl.ErrArenaFull {
2599 1 : if stalled {
2600 1 : d.opts.EventListener.WriteStallEnd()
2601 1 : }
2602 1 : return err
2603 : }
2604 1 : } else if !force {
2605 1 : if stalled {
2606 1 : d.opts.EventListener.WriteStallEnd()
2607 1 : }
2608 1 : return nil
2609 : }
2610 : // force || err == ErrArenaFull, so we need to rotate the current memtable.
2611 1 : {
2612 1 : var size uint64
2613 1 : for i := range d.mu.mem.queue {
2614 1 : size += d.mu.mem.queue[i].totalBytes()
2615 1 : }
2616 : // If ElevateWriteStallThresholdForFailover is true, we give an
2617 : // unlimited memory budget for memtables. This is simpler than trying to
2618 : // configure an explicit value, given that memory resources can vary.
2619 : // When using WAL failover in CockroachDB, an OOM risk is worth
2620 : // tolerating for workloads that have a strict latency SLO. Also, an
2621 : // unlimited budget here does not mean that the disk stall in the
2622 : // primary will go unnoticed until the OOM -- CockroachDB is monitoring
2623 : // disk stalls, and we expect it to fail the node after ~60s if the
2624 : // primary is stalled.
2625 1 : if size >= uint64(d.opts.MemTableStopWritesThreshold)*d.opts.MemTableSize &&
2626 1 : !d.mu.log.manager.ElevateWriteStallThresholdForFailover() {
2627 1 : // We have filled up the current memtable, but already queued memtables
2628 1 : // are still flushing, so we wait.
2629 1 : if !stalled {
2630 1 : stalled = true
2631 1 : d.opts.EventListener.WriteStallBegin(WriteStallBeginInfo{
2632 1 : Reason: "memtable count limit reached",
2633 1 : })
2634 1 : }
2635 1 : now := time.Now()
2636 1 : d.mu.compact.cond.Wait()
2637 1 : if b != nil {
2638 1 : b.commitStats.MemTableWriteStallDuration += time.Since(now)
2639 1 : }
2640 1 : continue
2641 : }
2642 : }
2643 1 : l0ReadAmp := d.mu.versions.currentVersion().L0Sublevels.ReadAmplification()
2644 1 : if l0ReadAmp >= d.opts.L0StopWritesThreshold {
2645 1 : // There are too many level-0 files, so we wait.
2646 1 : if !stalled {
2647 1 : stalled = true
2648 1 : d.opts.EventListener.WriteStallBegin(WriteStallBeginInfo{
2649 1 : Reason: "L0 file count limit exceeded",
2650 1 : })
2651 1 : }
2652 1 : now := time.Now()
2653 1 : d.mu.compact.cond.Wait()
2654 1 : if b != nil {
2655 1 : b.commitStats.L0ReadAmpWriteStallDuration += time.Since(now)
2656 1 : }
2657 1 : continue
2658 : }
2659 :
2660 1 : var newLogNum base.DiskFileNum
2661 1 : var prevLogSize uint64
2662 1 : if !d.opts.DisableWAL {
2663 1 : now := time.Now()
2664 1 : newLogNum, prevLogSize = d.recycleWAL()
2665 1 : if b != nil {
2666 1 : b.commitStats.WALRotationDuration += time.Since(now)
2667 1 : }
2668 : }
2669 :
2670 1 : immMem := d.mu.mem.mutable
2671 1 : imm := d.mu.mem.queue[len(d.mu.mem.queue)-1]
2672 1 : imm.logSize = prevLogSize
2673 1 : imm.flushForced = imm.flushForced || (b == nil)
2674 1 :
2675 1 : // If we are manually flushing and we used less than half of the bytes in
2676 1 : // the memtable, don't increase the size for the next memtable. This
2677 1 : // reduces memtable memory pressure when an application is frequently
2678 1 : // manually flushing.
2679 1 : if (b == nil) && uint64(immMem.availBytes()) > immMem.totalBytes()/2 {
2680 1 : d.mu.mem.nextSize = immMem.totalBytes()
2681 1 : }
2682 :
2683 1 : if b != nil && b.flushable != nil {
2684 1 : // The batch is too large to fit in the memtable so add it directly to
2685 1 : // the immutable queue. The flushable batch is associated with the same
2686 1 : // log as the immutable memtable, but logically occurs after it in
2687 1 : // seqnum space. We ensure while flushing that the flushable batch
2688 1 : // is flushed along with the previous memtable in the flushable
2689 1 : // queue. See the top level comment in DB.flush1 to learn how this
2690 1 : // is ensured.
2691 1 : //
2692 1 : // See DB.commitWrite for the special handling of log writes for large
2693 1 : // batches. In particular, the large batch has already written to
2694 1 : // imm.logNum.
2695 1 : entry := d.newFlushableEntry(b.flushable, imm.logNum, b.SeqNum())
2696 1 : // The large batch is by definition large. Reserve space from the cache
2697 1 : // for it until it is flushed.
2698 1 : entry.releaseMemAccounting = d.opts.Cache.Reserve(int(b.flushable.totalBytes()))
2699 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
2700 1 : }
2701 :
2702 1 : var logSeqNum uint64
2703 1 : if b != nil {
2704 1 : logSeqNum = b.SeqNum()
2705 1 : if b.flushable != nil {
2706 1 : logSeqNum += uint64(b.Count())
2707 1 : }
2708 1 : } else {
2709 1 : logSeqNum = d.mu.versions.logSeqNum.Load()
2710 1 : }
2711 1 : d.rotateMemtable(newLogNum, logSeqNum, immMem)
2712 1 : force = false
2713 : }
2714 : }
2715 :
2716 : // Both DB.mu and commitPipeline.mu must be held by the caller.
2717 1 : func (d *DB) rotateMemtable(newLogNum base.DiskFileNum, logSeqNum uint64, prev *memTable) {
2718 1 : // Create a new memtable, scheduling the previous one for flushing. We do
2719 1 : // this even if the previous memtable was empty because the DB.Flush
2720 1 : // mechanism is dependent on being able to wait for the empty memtable to
2721 1 : // flush. We can't just mark the empty memtable as flushed here because we
2722 1 : // also have to wait for all previous immutable tables to
2723 1 : // flush. Additionally, the memtable is tied to particular WAL file and we
2724 1 : // want to go through the flush path in order to recycle that WAL file.
2725 1 : //
2726 1 : // NB: newLogNum corresponds to the WAL that contains mutations that are
2727 1 : // present in the new memtable. When immutable memtables are flushed to
2728 1 : // disk, a VersionEdit will be created telling the manifest the minimum
2729 1 : // unflushed log number (which will be the next one in d.mu.mem.mutable
2730 1 : // that was not flushed).
2731 1 : //
2732 1 : // NB: prev should be the current mutable memtable.
2733 1 : var entry *flushableEntry
2734 1 : d.mu.mem.mutable, entry = d.newMemTable(newLogNum, logSeqNum)
2735 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
2736 1 : d.updateReadStateLocked(nil)
2737 1 : if prev.writerUnref() {
2738 1 : d.maybeScheduleFlush()
2739 1 : }
2740 : }
2741 :
2742 : // Both DB.mu and commitPipeline.mu must be held by the caller. Note that DB.mu
2743 : // may be released and reacquired.
2744 1 : func (d *DB) recycleWAL() (newLogNum base.DiskFileNum, prevLogSize uint64) {
2745 1 : if d.opts.DisableWAL {
2746 0 : panic("pebble: invalid function call")
2747 : }
2748 1 : jobID := d.newJobIDLocked()
2749 1 : newLogNum = d.mu.versions.getNextDiskFileNum()
2750 1 :
2751 1 : d.mu.Unlock()
2752 1 : // Close the previous log first. This writes an EOF trailer
2753 1 : // signifying the end of the file and syncs it to disk. We must
2754 1 : // close the previous log before linking the new log file,
2755 1 : // otherwise a crash could leave both logs with unclean tails, and
2756 1 : // Open will treat the previous log as corrupt.
2757 1 : offset, err := d.mu.log.writer.Close()
2758 1 : if err != nil {
2759 0 : // What to do here? Stumbling on doesn't seem worthwhile. If we failed to
2760 0 : // close the previous log it is possible we lost a write.
2761 0 : panic(err)
2762 : }
2763 1 : prevLogSize = uint64(offset)
2764 1 : metrics := d.mu.log.writer.Metrics()
2765 1 :
2766 1 : d.mu.Lock()
2767 1 : if err := d.mu.log.metrics.LogWriterMetrics.Merge(&metrics); err != nil {
2768 0 : d.opts.Logger.Errorf("metrics error: %s", err)
2769 0 : }
2770 :
2771 1 : d.mu.Unlock()
2772 1 : writer, err := d.mu.log.manager.Create(wal.NumWAL(newLogNum), int(jobID))
2773 1 : if err != nil {
2774 0 : panic(err)
2775 : }
2776 :
2777 1 : d.mu.Lock()
2778 1 : d.mu.log.writer = writer
2779 1 : return newLogNum, prevLogSize
2780 : }
2781 :
2782 1 : func (d *DB) getEarliestUnflushedSeqNumLocked() uint64 {
2783 1 : seqNum := InternalKeySeqNumMax
2784 1 : for i := range d.mu.mem.queue {
2785 1 : logSeqNum := d.mu.mem.queue[i].logSeqNum
2786 1 : if seqNum > logSeqNum {
2787 1 : seqNum = logSeqNum
2788 1 : }
2789 : }
2790 1 : return seqNum
2791 : }
2792 :
2793 1 : func (d *DB) getInProgressCompactionInfoLocked(finishing *compaction) (rv []compactionInfo) {
2794 1 : for c := range d.mu.compact.inProgress {
2795 1 : if len(c.flushing) == 0 && (finishing == nil || c != finishing) {
2796 1 : info := compactionInfo{
2797 1 : versionEditApplied: c.versionEditApplied,
2798 1 : inputs: c.inputs,
2799 1 : smallest: c.smallest,
2800 1 : largest: c.largest,
2801 1 : outputLevel: -1,
2802 1 : }
2803 1 : if c.outputLevel != nil {
2804 1 : info.outputLevel = c.outputLevel.level
2805 1 : }
2806 1 : rv = append(rv, info)
2807 : }
2808 : }
2809 1 : return
2810 : }
2811 :
2812 1 : func inProgressL0Compactions(inProgress []compactionInfo) []manifest.L0Compaction {
2813 1 : var compactions []manifest.L0Compaction
2814 1 : for _, info := range inProgress {
2815 1 : // Skip in-progress compactions that have already committed; the L0
2816 1 : // sublevels initialization code requires the set of in-progress
2817 1 : // compactions to be consistent with the current version. Compactions
2818 1 : // with versionEditApplied=true are already applied to the current
2819 1 : // version and but are performing cleanup without the database mutex.
2820 1 : if info.versionEditApplied {
2821 1 : continue
2822 : }
2823 1 : l0 := false
2824 1 : for _, cl := range info.inputs {
2825 1 : l0 = l0 || cl.level == 0
2826 1 : }
2827 1 : if !l0 {
2828 1 : continue
2829 : }
2830 1 : compactions = append(compactions, manifest.L0Compaction{
2831 1 : Smallest: info.smallest,
2832 1 : Largest: info.largest,
2833 1 : IsIntraL0: info.outputLevel == 0,
2834 1 : })
2835 : }
2836 1 : return compactions
2837 : }
2838 :
2839 : // firstError returns the first non-nil error of err0 and err1, or nil if both
2840 : // are nil.
2841 1 : func firstError(err0, err1 error) error {
2842 1 : if err0 != nil {
2843 1 : return err0
2844 1 : }
2845 1 : return err1
2846 : }
2847 :
2848 : // SetCreatorID sets the CreatorID which is needed in order to use shared objects.
2849 : // Remote object usage is disabled until this method is called the first time.
2850 : // Once set, the Creator ID is persisted and cannot change.
2851 : //
2852 : // Does nothing if SharedStorage was not set in the options when the DB was
2853 : // opened or if the DB is in read-only mode.
2854 1 : func (d *DB) SetCreatorID(creatorID uint64) error {
2855 1 : if d.opts.Experimental.RemoteStorage == nil || d.opts.ReadOnly {
2856 0 : return nil
2857 0 : }
2858 1 : return d.objProvider.SetCreatorID(objstorage.CreatorID(creatorID))
2859 : }
2860 :
2861 : // KeyStatistics keeps track of the number of keys that have been pinned by a
2862 : // snapshot as well as counts of the different key kinds in the lsm.
2863 : //
2864 : // One way of using the accumulated stats, when we only have sets and dels,
2865 : // and say the counts are represented as del_count, set_count,
2866 : // del_latest_count, set_latest_count, snapshot_pinned_count.
2867 : //
2868 : // - del_latest_count + set_latest_count is the set of unique user keys
2869 : // (unique).
2870 : //
2871 : // - set_latest_count is the set of live unique user keys (live_unique).
2872 : //
2873 : // - Garbage is del_count + set_count - live_unique.
2874 : //
2875 : // - If everything were in the LSM, del_count+set_count-snapshot_pinned_count
2876 : // would also be the set of unique user keys (note that
2877 : // snapshot_pinned_count is counting something different -- see comment below).
2878 : // But snapshot_pinned_count only counts keys in the LSM so the excess here
2879 : // must be keys in memtables.
2880 : type KeyStatistics struct {
2881 : // TODO(sumeer): the SnapshotPinned* are incorrect in that these older
2882 : // versions can be in a different level. Either fix the accounting or
2883 : // rename these fields.
2884 :
2885 : // SnapshotPinnedKeys represents obsolete keys that cannot be elided during
2886 : // a compaction, because they are required by an open snapshot.
2887 : SnapshotPinnedKeys int
2888 : // SnapshotPinnedKeysBytes is the total number of bytes of all snapshot
2889 : // pinned keys.
2890 : SnapshotPinnedKeysBytes uint64
2891 : // KindsCount is the count for each kind of key. It includes point keys,
2892 : // range deletes and range keys.
2893 : KindsCount [InternalKeyKindMax + 1]int
2894 : // LatestKindsCount is the count for each kind of key when it is the latest
2895 : // kind for a user key. It is only populated for point keys.
2896 : LatestKindsCount [InternalKeyKindMax + 1]int
2897 : }
2898 :
2899 : // LSMKeyStatistics is used by DB.ScanStatistics.
2900 : type LSMKeyStatistics struct {
2901 : Accumulated KeyStatistics
2902 : // Levels contains statistics only for point keys. Range deletions and range keys will
2903 : // appear in Accumulated but not Levels.
2904 : Levels [numLevels]KeyStatistics
2905 : // BytesRead represents the logical, pre-compression size of keys and values read
2906 : BytesRead uint64
2907 : }
2908 :
2909 : // ScanStatisticsOptions is used by DB.ScanStatistics.
2910 : type ScanStatisticsOptions struct {
2911 : // LimitBytesPerSecond indicates the number of bytes that are able to be read
2912 : // per second using ScanInternal.
2913 : // A value of 0 indicates that there is no limit set.
2914 : LimitBytesPerSecond int64
2915 : }
2916 :
2917 : // ScanStatistics returns the count of different key kinds within the lsm for a
2918 : // key span [lower, upper) as well as the number of snapshot keys.
2919 : func (d *DB) ScanStatistics(
2920 : ctx context.Context, lower, upper []byte, opts ScanStatisticsOptions,
2921 1 : ) (LSMKeyStatistics, error) {
2922 1 : stats := LSMKeyStatistics{}
2923 1 : var prevKey InternalKey
2924 1 : var rateLimitFunc func(key *InternalKey, val LazyValue) error
2925 1 : tb := tokenbucket.TokenBucket{}
2926 1 :
2927 1 : if opts.LimitBytesPerSecond != 0 {
2928 0 : // Each "token" roughly corresponds to a byte that was read.
2929 0 : tb.Init(tokenbucket.TokensPerSecond(opts.LimitBytesPerSecond), tokenbucket.Tokens(1024))
2930 0 : rateLimitFunc = func(key *InternalKey, val LazyValue) error {
2931 0 : return tb.WaitCtx(ctx, tokenbucket.Tokens(key.Size()+val.Len()))
2932 0 : }
2933 : }
2934 :
2935 1 : scanInternalOpts := &scanInternalOptions{
2936 1 : visitPointKey: func(key *InternalKey, value LazyValue, iterInfo IteratorLevel) error {
2937 1 : // If the previous key is equal to the current point key, the current key was
2938 1 : // pinned by a snapshot.
2939 1 : size := uint64(key.Size())
2940 1 : kind := key.Kind()
2941 1 : sameKey := d.equal(prevKey.UserKey, key.UserKey)
2942 1 : if iterInfo.Kind == IteratorLevelLSM && sameKey {
2943 1 : stats.Levels[iterInfo.Level].SnapshotPinnedKeys++
2944 1 : stats.Levels[iterInfo.Level].SnapshotPinnedKeysBytes += size
2945 1 : stats.Accumulated.SnapshotPinnedKeys++
2946 1 : stats.Accumulated.SnapshotPinnedKeysBytes += size
2947 1 : }
2948 1 : if iterInfo.Kind == IteratorLevelLSM {
2949 1 : stats.Levels[iterInfo.Level].KindsCount[kind]++
2950 1 : }
2951 1 : if !sameKey {
2952 1 : if iterInfo.Kind == IteratorLevelLSM {
2953 1 : stats.Levels[iterInfo.Level].LatestKindsCount[kind]++
2954 1 : }
2955 1 : stats.Accumulated.LatestKindsCount[kind]++
2956 : }
2957 :
2958 1 : stats.Accumulated.KindsCount[kind]++
2959 1 : prevKey.CopyFrom(*key)
2960 1 : stats.BytesRead += uint64(key.Size() + value.Len())
2961 1 : return nil
2962 : },
2963 0 : visitRangeDel: func(start, end []byte, seqNum uint64) error {
2964 0 : stats.Accumulated.KindsCount[InternalKeyKindRangeDelete]++
2965 0 : stats.BytesRead += uint64(len(start) + len(end))
2966 0 : return nil
2967 0 : },
2968 0 : visitRangeKey: func(start, end []byte, keys []rangekey.Key) error {
2969 0 : stats.BytesRead += uint64(len(start) + len(end))
2970 0 : for _, key := range keys {
2971 0 : stats.Accumulated.KindsCount[key.Kind()]++
2972 0 : stats.BytesRead += uint64(len(key.Value) + len(key.Suffix))
2973 0 : }
2974 0 : return nil
2975 : },
2976 : includeObsoleteKeys: true,
2977 : IterOptions: IterOptions{
2978 : KeyTypes: IterKeyTypePointsAndRanges,
2979 : LowerBound: lower,
2980 : UpperBound: upper,
2981 : },
2982 : rateLimitFunc: rateLimitFunc,
2983 : }
2984 1 : iter, err := d.newInternalIter(ctx, snapshotIterOpts{}, scanInternalOpts)
2985 1 : if err != nil {
2986 0 : return LSMKeyStatistics{}, err
2987 0 : }
2988 1 : defer iter.close()
2989 1 :
2990 1 : err = scanInternalImpl(ctx, lower, upper, iter, scanInternalOpts)
2991 1 :
2992 1 : if err != nil {
2993 0 : return LSMKeyStatistics{}, err
2994 0 : }
2995 :
2996 1 : return stats, nil
2997 : }
2998 :
2999 : // ObjProvider returns the objstorage.Provider for this database. Meant to be
3000 : // used for internal purposes only.
3001 1 : func (d *DB) ObjProvider() objstorage.Provider {
3002 1 : return d.objProvider
3003 1 : }
3004 :
3005 1 : func (d *DB) checkVirtualBounds(m *fileMetadata) {
3006 1 : if !invariants.Enabled {
3007 0 : return
3008 0 : }
3009 :
3010 1 : objMeta, err := d.objProvider.Lookup(fileTypeTable, m.FileBacking.DiskFileNum)
3011 1 : if err != nil {
3012 0 : panic(err)
3013 : }
3014 1 : if objMeta.IsExternal() {
3015 0 : // Nothing to do; bounds are expected to be loose.
3016 0 : return
3017 0 : }
3018 :
3019 1 : iters, err := d.newIters(context.TODO(), m, nil, internalIterOpts{}, iterPointKeys|iterRangeDeletions|iterRangeKeys)
3020 1 : if err != nil {
3021 0 : panic(errors.Wrap(err, "pebble: error creating iterators"))
3022 : }
3023 1 : defer iters.CloseAll()
3024 1 :
3025 1 : if m.HasPointKeys {
3026 1 : pointIter := iters.Point()
3027 1 : rangeDelIter := iters.RangeDeletion()
3028 1 :
3029 1 : // Check that the lower bound is tight.
3030 1 : pointKey, _ := pointIter.First()
3031 1 : rangeDel, err := rangeDelIter.First()
3032 1 : if err != nil {
3033 0 : panic(err)
3034 : }
3035 1 : if (rangeDel == nil || d.cmp(rangeDel.SmallestKey().UserKey, m.SmallestPointKey.UserKey) != 0) &&
3036 1 : (pointKey == nil || d.cmp(pointKey.UserKey, m.SmallestPointKey.UserKey) != 0) {
3037 0 : panic(errors.Newf("pebble: virtual sstable %s lower point key bound is not tight", m.FileNum))
3038 : }
3039 :
3040 : // Check that the upper bound is tight.
3041 1 : pointKey, _ = pointIter.Last()
3042 1 : rangeDel, err = rangeDelIter.Last()
3043 1 : if err != nil {
3044 0 : panic(err)
3045 : }
3046 1 : if (rangeDel == nil || d.cmp(rangeDel.LargestKey().UserKey, m.LargestPointKey.UserKey) != 0) &&
3047 1 : (pointKey == nil || d.cmp(pointKey.UserKey, m.LargestPointKey.UserKey) != 0) {
3048 0 : panic(errors.Newf("pebble: virtual sstable %s upper point key bound is not tight", m.FileNum))
3049 : }
3050 :
3051 : // Check that iterator keys are within bounds.
3052 1 : for key, _ := pointIter.First(); key != nil; key, _ = pointIter.Next() {
3053 1 : if d.cmp(key.UserKey, m.SmallestPointKey.UserKey) < 0 || d.cmp(key.UserKey, m.LargestPointKey.UserKey) > 0 {
3054 0 : panic(errors.Newf("pebble: virtual sstable %s point key %s is not within bounds", m.FileNum, key.UserKey))
3055 : }
3056 : }
3057 1 : s, err := rangeDelIter.First()
3058 1 : for ; s != nil; s, err = rangeDelIter.Next() {
3059 1 : if d.cmp(s.SmallestKey().UserKey, m.SmallestPointKey.UserKey) < 0 {
3060 0 : panic(errors.Newf("pebble: virtual sstable %s point key %s is not within bounds", m.FileNum, s.SmallestKey().UserKey))
3061 : }
3062 1 : if d.cmp(s.LargestKey().UserKey, m.LargestPointKey.UserKey) > 0 {
3063 0 : panic(errors.Newf("pebble: virtual sstable %s point key %s is not within bounds", m.FileNum, s.LargestKey().UserKey))
3064 : }
3065 : }
3066 1 : if err != nil {
3067 0 : panic(err)
3068 : }
3069 : }
3070 :
3071 1 : if !m.HasRangeKeys {
3072 1 : return
3073 1 : }
3074 0 : rangeKeyIter := iters.RangeKey()
3075 0 :
3076 0 : // Check that the lower bound is tight.
3077 0 : if s, err := rangeKeyIter.First(); err != nil {
3078 0 : panic(err)
3079 0 : } else if d.cmp(s.SmallestKey().UserKey, m.SmallestRangeKey.UserKey) != 0 {
3080 0 : panic(errors.Newf("pebble: virtual sstable %s lower range key bound is not tight", m.FileNum))
3081 : }
3082 :
3083 : // Check that upper bound is tight.
3084 0 : if s, err := rangeKeyIter.Last(); err != nil {
3085 0 : panic(err)
3086 0 : } else if d.cmp(s.LargestKey().UserKey, m.LargestRangeKey.UserKey) != 0 {
3087 0 : panic(errors.Newf("pebble: virtual sstable %s upper range key bound is not tight", m.FileNum))
3088 : }
3089 :
3090 0 : s, err := rangeKeyIter.First()
3091 0 : for ; s != nil; s, err = rangeKeyIter.Next() {
3092 0 : if d.cmp(s.SmallestKey().UserKey, m.SmallestRangeKey.UserKey) < 0 {
3093 0 : panic(errors.Newf("pebble: virtual sstable %s point key %s is not within bounds", m.FileNum, s.SmallestKey().UserKey))
3094 : }
3095 0 : if d.cmp(s.LargestKey().UserKey, m.LargestRangeKey.UserKey) > 0 {
3096 0 : panic(errors.Newf("pebble: virtual sstable %s point key %s is not within bounds", m.FileNum, s.LargestKey().UserKey))
3097 : }
3098 : }
3099 0 : if err != nil {
3100 0 : panic(err)
3101 : }
3102 : }
3103 :
3104 : // DebugString returns a debugging string describing the LSM.
3105 0 : func (d *DB) DebugString() string {
3106 0 : d.mu.Lock()
3107 0 : defer d.mu.Unlock()
3108 0 : return d.mu.versions.currentVersion().DebugString()
3109 0 : }
|
