Line data Source code
1 : // Copyright 2012 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "bytes"
9 : "context"
10 : "encoding/binary"
11 : "fmt"
12 : "io"
13 : "math"
14 : "os"
15 : "slices"
16 : "sync/atomic"
17 : "time"
18 :
19 : "github.com/cockroachdb/crlib/crtime"
20 : "github.com/cockroachdb/errors"
21 : "github.com/cockroachdb/errors/oserror"
22 : "github.com/cockroachdb/pebble/batchrepr"
23 : "github.com/cockroachdb/pebble/internal/arenaskl"
24 : "github.com/cockroachdb/pebble/internal/base"
25 : "github.com/cockroachdb/pebble/internal/cache"
26 : "github.com/cockroachdb/pebble/internal/invariants"
27 : "github.com/cockroachdb/pebble/internal/keyspan"
28 : "github.com/cockroachdb/pebble/internal/manifest"
29 : "github.com/cockroachdb/pebble/internal/manual"
30 : "github.com/cockroachdb/pebble/objstorage"
31 : "github.com/cockroachdb/pebble/objstorage/objstorageprovider"
32 : "github.com/cockroachdb/pebble/objstorage/remote"
33 : "github.com/cockroachdb/pebble/record"
34 : "github.com/cockroachdb/pebble/vfs"
35 : "github.com/cockroachdb/pebble/wal"
36 : "github.com/cockroachdb/redact"
37 : "github.com/prometheus/client_golang/prometheus"
38 : )
39 :
40 : const (
41 : initialMemTableSize = 256 << 10 // 256 KB
42 :
43 : // The max batch size is limited by the uint32 offsets stored in
44 : // internal/batchskl.node, DeferredBatchOp, and flushableBatchEntry.
45 : //
46 : // We limit the size to MaxUint32 (just short of 4GB) so that the exclusive
47 : // end of an allocation fits in uint32.
48 : //
49 : // On 32-bit systems, slices are naturally limited to MaxInt (just short of
50 : // 2GB).
51 : maxBatchSize = min(math.MaxUint32, math.MaxInt)
52 :
53 : // The max memtable size is limited by the uint32 offsets stored in
54 : // internal/arenaskl.node, DeferredBatchOp, and flushableBatchEntry.
55 : //
56 : // We limit the size to MaxUint32 (just short of 4GB) so that the exclusive
57 : // end of an allocation fits in uint32.
58 : //
59 : // On 32-bit systems, slices are naturally limited to MaxInt (just short of
60 : // 2GB).
61 : maxMemTableSize = min(math.MaxUint32, math.MaxInt)
62 : )
63 :
64 : // FileCacheSize can be used to determine the file
65 : // cache size for a single db, given the maximum open
66 : // files which can be used by a file cache which is
67 : // only used by a single db.
68 1 : func FileCacheSize(maxOpenFiles int) int {
69 1 : fileCacheSize := maxOpenFiles - numNonFileCacheFiles
70 1 : if fileCacheSize < minFileCacheSize {
71 1 : fileCacheSize = minFileCacheSize
72 1 : }
73 1 : return fileCacheSize
74 : }
75 :
76 : // Open opens a DB whose files live in the given directory.
77 : //
78 : // IsCorruptionError() can be use to determine if the error is caused by on-disk
79 : // corruption.
80 1 : func Open(dirname string, opts *Options) (db *DB, err error) {
81 1 : // Make a copy of the options so that we don't mutate the passed in options.
82 1 : opts = opts.Clone()
83 1 : opts.EnsureDefaults()
84 1 : if opts.Experimental.CompactionScheduler == nil {
85 0 : opts.Experimental.CompactionScheduler = newConcurrencyLimitScheduler(defaultTimeSource{})
86 0 : }
87 1 : if err := opts.Validate(); err != nil {
88 0 : return nil, err
89 0 : }
90 1 : if opts.LoggerAndTracer == nil {
91 1 : opts.LoggerAndTracer = &base.LoggerWithNoopTracer{Logger: opts.Logger}
92 1 : } else {
93 0 : opts.Logger = opts.LoggerAndTracer
94 0 : }
95 :
96 1 : if invariants.Sometimes(5) {
97 1 : assertComparer := base.MakeAssertComparer(*opts.Comparer)
98 1 : opts.Comparer = &assertComparer
99 1 : }
100 :
101 : // In all error cases, we return db = nil; this is used by various
102 : // deferred cleanups.
103 1 : maybeCleanUp := func(fn func() error) {
104 1 : if db == nil {
105 0 : err = errors.CombineErrors(err, fn())
106 0 : }
107 : }
108 :
109 : // Open the database and WAL directories first.
110 1 : walDirname, secondaryWalDirName, dataDir, err := prepareAndOpenDirs(dirname, opts)
111 1 : if err != nil {
112 0 : return nil, errors.Wrapf(err, "error opening database at %q", dirname)
113 0 : }
114 1 : defer maybeCleanUp(dataDir.Close)
115 1 :
116 1 : // Lock the database directory.
117 1 : fileLock, err := base.AcquireOrValidateDirectoryLock(opts.Lock, dirname, opts.FS)
118 1 : if err != nil {
119 0 : return nil, err
120 0 : }
121 1 : defer maybeCleanUp(fileLock.Close)
122 1 :
123 1 : // List the directory contents. This also happens to include WAL log files, if
124 1 : // they are in the same dir, but we will ignore those below. The provider is
125 1 : // also given this list, but it ignores non sstable files.
126 1 : ls, err := opts.FS.List(dirname)
127 1 : if err != nil {
128 0 : return nil, err
129 0 : }
130 :
131 : // Establish the format major version.
132 1 : formatVersion, formatVersionMarker, err := lookupFormatMajorVersion(opts.FS, dirname, ls)
133 1 : if err != nil {
134 0 : return nil, err
135 0 : }
136 1 : defer maybeCleanUp(formatVersionMarker.Close)
137 1 :
138 1 : noFormatVersionMarker := formatVersion == FormatDefault
139 1 : if noFormatVersionMarker {
140 1 : // We will initialize the store at the minimum possible format, then upgrade
141 1 : // the format to the desired one. This helps test the format upgrade code.
142 1 : formatVersion = FormatMinSupported
143 1 : if opts.Experimental.CreateOnShared != remote.CreateOnSharedNone {
144 1 : formatVersion = FormatMinForSharedObjects
145 1 : }
146 : // There is no format version marker file. There are three cases:
147 : // - we are trying to open an existing store that was created at
148 : // FormatMostCompatible (the only one without a version marker file)
149 : // - we are creating a new store;
150 : // - we are retrying a failed creation.
151 : //
152 : // To error in the first case, we set ErrorIfNotPristine.
153 1 : opts.ErrorIfNotPristine = true
154 1 : defer func() {
155 1 : if err != nil && errors.Is(err, ErrDBNotPristine) {
156 0 : // We must be trying to open an existing store at FormatMostCompatible.
157 0 : // Correct the error in this case -we
158 0 : err = errors.Newf(
159 0 : "pebble: database %q written in format major version 1 which is no longer supported",
160 0 : dirname)
161 0 : }
162 : }()
163 : }
164 :
165 : // Find the currently active manifest, if there is one.
166 1 : manifestMarker, manifestFileNum, manifestExists, err := findCurrentManifest(opts.FS, dirname, ls)
167 1 : if err != nil {
168 0 : return nil, errors.Wrapf(err, "pebble: database %q", dirname)
169 0 : }
170 1 : defer maybeCleanUp(manifestMarker.Close)
171 1 :
172 1 : // Atomic markers may leave behind obsolete files if there's a crash
173 1 : // mid-update. Clean these up if we're not in read-only mode.
174 1 : if !opts.ReadOnly {
175 1 : if err := formatVersionMarker.RemoveObsolete(); err != nil {
176 0 : return nil, err
177 0 : }
178 1 : if err := manifestMarker.RemoveObsolete(); err != nil {
179 0 : return nil, err
180 0 : }
181 : }
182 :
183 1 : if opts.Cache == nil {
184 1 : opts.Cache = cache.New(opts.CacheSize)
185 1 : defer opts.Cache.Unref()
186 1 : }
187 :
188 1 : d := &DB{
189 1 : cacheHandle: opts.Cache.NewHandle(),
190 1 : dirname: dirname,
191 1 : opts: opts,
192 1 : cmp: opts.Comparer.Compare,
193 1 : equal: opts.Comparer.Equal,
194 1 : merge: opts.Merger.Merge,
195 1 : split: opts.Comparer.Split,
196 1 : abbreviatedKey: opts.Comparer.AbbreviatedKey,
197 1 : largeBatchThreshold: (opts.MemTableSize - uint64(memTableEmptySize)) / 2,
198 1 : dataDirLock: fileLock,
199 1 : dataDir: dataDir,
200 1 : closed: new(atomic.Value),
201 1 : closedCh: make(chan struct{}),
202 1 : }
203 1 : d.mu.versions = &versionSet{}
204 1 : d.diskAvailBytes.Store(math.MaxUint64)
205 1 : d.problemSpans.Init(manifest.NumLevels, opts.Comparer.Compare)
206 1 :
207 1 : defer func() {
208 1 : // If an error or panic occurs during open, attempt to release the manually
209 1 : // allocated memory resources. Note that rather than look for an error, we
210 1 : // look for the return of a nil DB pointer.
211 1 : if r := recover(); db == nil {
212 0 : // If there's an unused, recycled memtable, we need to release its memory.
213 0 : if obsoleteMemTable := d.memTableRecycle.Swap(nil); obsoleteMemTable != nil {
214 0 : d.freeMemTable(obsoleteMemTable)
215 0 : }
216 :
217 0 : if d.fileCache != nil {
218 0 : _ = d.fileCache.Close()
219 0 : }
220 0 : d.cacheHandle.Close()
221 0 :
222 0 : for _, mem := range d.mu.mem.queue {
223 0 : switch t := mem.flushable.(type) {
224 0 : case *memTable:
225 0 : manual.Free(manual.MemTable, t.arenaBuf)
226 0 : t.arenaBuf = manual.Buf{}
227 : }
228 : }
229 0 : if d.cleanupManager != nil {
230 0 : d.cleanupManager.Close()
231 0 : }
232 0 : if d.objProvider != nil {
233 0 : _ = d.objProvider.Close()
234 0 : }
235 0 : if d.mu.versions.manifestFile != nil {
236 0 : _ = d.mu.versions.manifestFile.Close()
237 0 : }
238 0 : if r != nil {
239 0 : panic(r)
240 : }
241 : }
242 : }()
243 :
244 1 : d.commit = newCommitPipeline(commitEnv{
245 1 : logSeqNum: &d.mu.versions.logSeqNum,
246 1 : visibleSeqNum: &d.mu.versions.visibleSeqNum,
247 1 : apply: d.commitApply,
248 1 : write: d.commitWrite,
249 1 : })
250 1 : d.mu.nextJobID = 1
251 1 : d.mu.mem.nextSize = opts.MemTableSize
252 1 : if d.mu.mem.nextSize > initialMemTableSize {
253 1 : d.mu.mem.nextSize = initialMemTableSize
254 1 : }
255 1 : d.mu.compact.cond.L = &d.mu.Mutex
256 1 : d.mu.compact.inProgress = make(map[compaction]struct{})
257 1 : d.mu.compact.noOngoingFlushStartTime = crtime.NowMono()
258 1 : d.mu.snapshots.init()
259 1 : // logSeqNum is the next sequence number that will be assigned.
260 1 : // Start assigning sequence numbers from base.SeqNumStart to leave
261 1 : // room for reserved sequence numbers (see comments around
262 1 : // SeqNumStart).
263 1 : d.mu.versions.logSeqNum.Store(base.SeqNumStart)
264 1 : d.mu.formatVers.vers.Store(uint64(formatVersion))
265 1 : d.mu.formatVers.marker = formatVersionMarker
266 1 :
267 1 : d.timeNow = time.Now
268 1 : d.openedAt = d.timeNow()
269 1 :
270 1 : d.mu.Lock()
271 1 : defer d.mu.Unlock()
272 1 :
273 1 : jobID := d.newJobIDLocked()
274 1 :
275 1 : providerSettings := opts.MakeObjStorageProviderSettings(dirname)
276 1 : providerSettings.FSDirInitialListing = ls
277 1 : d.objProvider, err = objstorageprovider.Open(providerSettings)
278 1 : if err != nil {
279 0 : return nil, err
280 0 : }
281 :
282 1 : blobRewriteHeuristic := manifest.BlobRewriteHeuristic{
283 1 : CurrentTime: d.timeNow,
284 1 : MinimumAge: opts.Experimental.ValueSeparationPolicy().RewriteMinimumAge,
285 1 : }
286 1 :
287 1 : if !manifestExists {
288 1 : // DB does not exist.
289 1 : if d.opts.ErrorIfNotExists || d.opts.ReadOnly {
290 0 : return nil, errors.Wrapf(ErrDBDoesNotExist, "dirname=%q", dirname)
291 0 : }
292 :
293 : // Create the DB.
294 1 : if err := d.mu.versions.create(
295 1 : jobID, dirname, d.objProvider, opts, manifestMarker, d.FormatMajorVersion, blobRewriteHeuristic, &d.mu.Mutex); err != nil {
296 0 : return nil, err
297 0 : }
298 1 : } else {
299 1 : if opts.ErrorIfExists {
300 0 : return nil, errors.Wrapf(ErrDBAlreadyExists, "dirname=%q", dirname)
301 0 : }
302 : // Load the version set.
303 1 : if err := d.mu.versions.load(
304 1 : dirname, d.objProvider, opts, manifestFileNum, manifestMarker, d.FormatMajorVersion, blobRewriteHeuristic, &d.mu.Mutex); err != nil {
305 0 : return nil, err
306 0 : }
307 1 : if opts.ErrorIfNotPristine {
308 0 : liveFileNums := make(map[base.DiskFileNum]struct{})
309 0 : d.mu.versions.addLiveFileNums(liveFileNums)
310 0 : if len(liveFileNums) != 0 {
311 0 : return nil, errors.Wrapf(ErrDBNotPristine, "dirname=%q", dirname)
312 0 : }
313 : }
314 : }
315 :
316 : // In read-only mode, we replay directly into the mutable memtable but never
317 : // flush it. We need to delay creation of the memtable until we know the
318 : // sequence number of the first batch that will be inserted.
319 1 : if !d.opts.ReadOnly {
320 1 : var entry *flushableEntry
321 1 : d.mu.mem.mutable, entry = d.newMemTable(0 /* logNum */, d.mu.versions.logSeqNum.Load(), 0 /* minSize */)
322 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
323 1 : }
324 :
325 1 : d.mu.log.metrics.fsyncLatency = prometheus.NewHistogram(prometheus.HistogramOpts{
326 1 : Buckets: FsyncLatencyBuckets,
327 1 : })
328 1 :
329 1 : walOpts := wal.Options{
330 1 : Primary: wal.Dir{FS: opts.FS, Dirname: walDirname},
331 1 : Secondary: wal.Dir{},
332 1 : MinUnflushedWALNum: wal.NumWAL(d.mu.versions.minUnflushedLogNum),
333 1 : MaxNumRecyclableLogs: opts.MemTableStopWritesThreshold + 1,
334 1 : NoSyncOnClose: opts.NoSyncOnClose,
335 1 : BytesPerSync: opts.WALBytesPerSync,
336 1 : PreallocateSize: d.walPreallocateSize,
337 1 : MinSyncInterval: opts.WALMinSyncInterval,
338 1 : FsyncLatency: d.mu.log.metrics.fsyncLatency,
339 1 : QueueSemChan: d.commit.logSyncQSem,
340 1 : Logger: opts.Logger,
341 1 : EventListener: walEventListenerAdaptor{l: opts.EventListener},
342 1 : WriteWALSyncOffsets: func() bool { return d.FormatMajorVersion() >= FormatWALSyncChunks },
343 : }
344 : // Ensure we release the WAL directory locks if we fail to open the
345 : // database. If we fail before initializing the WAL manager, this defer is
346 : // responsible for releasing the locks. If we fail after initializing the
347 : // WAL manager, closing the WAL manager will release the locks.
348 : //
349 : // TODO(jackson): Open's cleanup error handling logic is convoluted; can we
350 : // simplify it?
351 1 : defer maybeCleanUp(func() (err error) {
352 0 : if d.mu.log.manager == nil {
353 0 : if walOpts.Primary.Lock != nil {
354 0 : err = errors.CombineErrors(err, walOpts.Primary.Lock.Close())
355 0 : }
356 0 : if walOpts.Secondary.Lock != nil {
357 0 : err = errors.CombineErrors(err, walOpts.Secondary.Lock.Close())
358 0 : }
359 0 : return err
360 : }
361 0 : return nil
362 : })
363 :
364 : // Lock the dedicated WAL directory, if configured.
365 1 : if walDirname != dirname {
366 1 : walOpts.Primary.Lock, err = base.AcquireOrValidateDirectoryLock(opts.WALDirLock, walDirname, opts.FS)
367 1 : if err != nil {
368 0 : return nil, err
369 0 : }
370 : }
371 1 : if opts.WALFailover != nil {
372 1 : walOpts.Secondary = opts.WALFailover.Secondary
373 1 : // Lock the secondary WAL directory, if distinct from the data directory
374 1 : // and primary WAL directory.
375 1 : if secondaryWalDirName != dirname && secondaryWalDirName != walDirname {
376 1 : walOpts.Secondary.Lock, err = base.AcquireOrValidateDirectoryLock(
377 1 : opts.WALFailover.Secondary.Lock, secondaryWalDirName, opts.WALFailover.Secondary.FS)
378 1 : if err != nil {
379 0 : return nil, err
380 0 : }
381 : }
382 1 : walOpts.Secondary.Dirname = secondaryWalDirName
383 1 : walOpts.FailoverOptions = opts.WALFailover.FailoverOptions
384 1 : walOpts.FailoverWriteAndSyncLatency = prometheus.NewHistogram(prometheus.HistogramOpts{
385 1 : Buckets: FsyncLatencyBuckets,
386 1 : })
387 : }
388 1 : walDirs := walOpts.Dirs()
389 1 : walRecoveryLocks := make([]*base.DirLock, len(opts.WALRecoveryDirs))
390 1 : defer func() {
391 1 : // We only need the recovery WALs during Open, so we can release
392 1 : // the locks after the WALs have been scanned.
393 1 : for _, l := range walRecoveryLocks {
394 0 : if l != nil {
395 0 : _ = l.Close()
396 0 : }
397 : }
398 : }()
399 1 : for i, dir := range opts.WALRecoveryDirs {
400 0 : dir.Dirname = resolveStorePath(dirname, dir.Dirname)
401 0 : if dir.Dirname != dirname {
402 0 : // Acquire a lock on the WAL recovery directory.
403 0 : walRecoveryLocks[i], err = base.AcquireOrValidateDirectoryLock(dir.Lock, dir.Dirname, dir.FS)
404 0 : if err != nil {
405 0 : return nil, errors.Wrapf(err, "error acquiring lock on WAL recovery directory %q", dir.Dirname)
406 0 : }
407 : }
408 0 : walDirs = append(walDirs, dir)
409 : }
410 1 : wals, err := wal.Scan(walDirs...)
411 1 : if err != nil {
412 0 : return nil, err
413 0 : }
414 1 : d.opts.Logger.Infof("Found %d WALs", redact.Safe(len(wals)))
415 1 : for i := range wals {
416 1 : d.opts.Logger.Infof(" - %s", wals[i])
417 1 : }
418 1 : walManager, err := wal.Init(walOpts, wals)
419 1 : if err != nil {
420 0 : return nil, err
421 0 : }
422 1 : defer maybeCleanUp(walManager.Close)
423 1 : d.mu.log.manager = walManager
424 1 :
425 1 : d.cleanupManager = openCleanupManager(opts, d.objProvider, d.getDeletionPacerInfo)
426 1 :
427 1 : if manifestExists && !opts.DisableConsistencyCheck {
428 1 : curVersion := d.mu.versions.currentVersion()
429 1 : if err := checkConsistency(curVersion, d.objProvider); err != nil {
430 0 : return nil, err
431 0 : }
432 : }
433 :
434 1 : fileCacheSize := FileCacheSize(opts.MaxOpenFiles)
435 1 : if opts.FileCache == nil {
436 1 : opts.FileCache = NewFileCache(opts.Experimental.FileCacheShards, fileCacheSize)
437 1 : defer opts.FileCache.Unref()
438 1 : }
439 1 : d.fileCache = opts.FileCache.newHandle(d.cacheHandle, d.objProvider, d.opts.LoggerAndTracer, d.opts.MakeReaderOptions(), d.reportCorruption)
440 1 : d.newIters = d.fileCache.newIters
441 1 : d.tableNewRangeKeyIter = tableNewRangeKeyIter(d.newIters)
442 1 :
443 1 : d.mu.annotators.totalFileSize = d.makeFileSizeAnnotator(func(f *manifest.TableMetadata) bool {
444 0 : return true
445 0 : })
446 1 : d.mu.annotators.remoteSize = d.makeFileSizeAnnotator(func(f *manifest.TableMetadata) bool {
447 0 : meta, err := d.objProvider.Lookup(base.FileTypeTable, f.TableBacking.DiskFileNum)
448 0 : if err != nil {
449 0 : return false
450 0 : }
451 0 : return meta.IsRemote()
452 : })
453 1 : d.mu.annotators.externalSize = d.makeFileSizeAnnotator(func(f *manifest.TableMetadata) bool {
454 0 : meta, err := d.objProvider.Lookup(base.FileTypeTable, f.TableBacking.DiskFileNum)
455 0 : if err != nil {
456 0 : return false
457 0 : }
458 0 : return meta.IsRemote() && meta.Remote.CleanupMethod == objstorage.SharedNoCleanup
459 : })
460 :
461 1 : var previousOptionsFileNum base.DiskFileNum
462 1 : var previousOptionsFilename string
463 1 : for _, filename := range ls {
464 1 : ft, fn, ok := base.ParseFilename(opts.FS, filename)
465 1 : if !ok {
466 1 : continue
467 : }
468 :
469 : // Don't reuse any obsolete file numbers to avoid modifying an
470 : // ingested sstable's original external file.
471 1 : d.mu.versions.markFileNumUsed(fn)
472 1 :
473 1 : switch ft {
474 0 : case base.FileTypeLog:
475 : // Ignore.
476 1 : case base.FileTypeOptions:
477 1 : if previousOptionsFileNum < fn {
478 1 : previousOptionsFileNum = fn
479 1 : previousOptionsFilename = filename
480 1 : }
481 0 : case base.FileTypeTemp, base.FileTypeOldTemp:
482 0 : if !d.opts.ReadOnly {
483 0 : // Some codepaths write to a temporary file and then
484 0 : // rename it to its final location when complete. A
485 0 : // temp file is leftover if a process exits before the
486 0 : // rename. Remove it.
487 0 : err := opts.FS.Remove(opts.FS.PathJoin(dirname, filename))
488 0 : if err != nil {
489 0 : return nil, err
490 0 : }
491 : }
492 : }
493 : }
494 1 : if n := len(wals); n > 0 {
495 1 : // Don't reuse any obsolete file numbers to avoid modifying an
496 1 : // ingested sstable's original external file.
497 1 : d.mu.versions.markFileNumUsed(base.DiskFileNum(wals[n-1].Num))
498 1 : }
499 :
500 : // Ratchet d.mu.versions.nextFileNum ahead of all known objects in the
501 : // objProvider. This avoids FileNum collisions with obsolete sstables.
502 1 : objects := d.objProvider.List()
503 1 : for _, obj := range objects {
504 1 : d.mu.versions.markFileNumUsed(obj.DiskFileNum)
505 1 : }
506 :
507 : // Validate the most-recent OPTIONS file, if there is one.
508 1 : if previousOptionsFilename != "" {
509 1 : path := opts.FS.PathJoin(dirname, previousOptionsFilename)
510 1 : previousOptions, err := readOptionsFile(opts, path)
511 1 : if err != nil {
512 0 : return nil, err
513 0 : }
514 1 : if err := opts.CheckCompatibility(dirname, previousOptions); err != nil {
515 0 : return nil, err
516 0 : }
517 : }
518 :
519 : // Replay any newer log files than the ones named in the manifest.
520 1 : var replayWALs wal.Logs
521 1 : for i, w := range wals {
522 1 : if base.DiskFileNum(w.Num) >= d.mu.versions.minUnflushedLogNum {
523 1 : replayWALs = wals[i:]
524 1 : break
525 : }
526 : }
527 1 : var flushableIngests []*ingestedFlushable
528 1 : for i, lf := range replayWALs {
529 1 : // WALs other than the last one would have been closed cleanly.
530 1 : //
531 1 : // Note: we used to never require strict WAL tails when reading from older
532 1 : // versions: RocksDB 6.2.1 and the version of Pebble included in CockroachDB
533 1 : // 20.1 do not guarantee that closed WALs end cleanly. But the earliest
534 1 : // compatible Pebble format is newer and guarantees a clean EOF.
535 1 : strictWALTail := i < len(replayWALs)-1
536 1 : fi, maxSeqNum, err := d.replayWAL(jobID, lf, strictWALTail)
537 1 : if err != nil {
538 0 : return nil, err
539 0 : }
540 1 : if len(fi) > 0 {
541 1 : flushableIngests = append(flushableIngests, fi...)
542 1 : }
543 1 : if d.mu.versions.logSeqNum.Load() < maxSeqNum {
544 1 : d.mu.versions.logSeqNum.Store(maxSeqNum)
545 1 : }
546 : }
547 1 : if d.mu.mem.mutable == nil {
548 1 : // Recreate the mutable memtable if replayWAL got rid of it.
549 1 : var entry *flushableEntry
550 1 : d.mu.mem.mutable, entry = d.newMemTable(d.mu.versions.getNextDiskFileNum(), d.mu.versions.logSeqNum.Load(), 0 /* minSize */)
551 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
552 1 : }
553 1 : d.mu.versions.visibleSeqNum.Store(d.mu.versions.logSeqNum.Load())
554 1 :
555 1 : // Register with the CompactionScheduler before calling
556 1 : // d.maybeScheduleFlush, since completion of the flush can trigger
557 1 : // compactions.
558 1 : d.opts.Experimental.CompactionScheduler.Register(2, d)
559 1 : if !d.opts.ReadOnly {
560 1 : d.maybeScheduleFlush()
561 1 : for d.mu.compact.flushing {
562 1 : d.mu.compact.cond.Wait()
563 1 : }
564 :
565 : // Create an empty .log file for the mutable memtable.
566 1 : newLogNum := d.mu.versions.getNextDiskFileNum()
567 1 : d.mu.log.writer, err = d.mu.log.manager.Create(wal.NumWAL(newLogNum), int(jobID))
568 1 : if err != nil {
569 0 : return nil, err
570 0 : }
571 :
572 : // This isn't strictly necessary as we don't use the log number for
573 : // memtables being flushed, only for the next unflushed memtable.
574 1 : d.mu.mem.queue[len(d.mu.mem.queue)-1].logNum = newLogNum
575 : }
576 1 : d.updateReadStateLocked(d.opts.DebugCheck)
577 1 :
578 1 : if !d.opts.ReadOnly {
579 1 : // If the Options specify a format major version higher than the
580 1 : // loaded database's, upgrade it. If this is a new database, this
581 1 : // code path also performs an initial upgrade from the starting
582 1 : // implicit MinSupported version.
583 1 : //
584 1 : // We ratchet the version this far into Open so that migrations have a read
585 1 : // state available. Note that this also results in creating/updating the
586 1 : // format version marker file.
587 1 : if opts.FormatMajorVersion > d.FormatMajorVersion() {
588 1 : if err := d.ratchetFormatMajorVersionLocked(opts.FormatMajorVersion); err != nil {
589 0 : return nil, err
590 0 : }
591 1 : } else if noFormatVersionMarker {
592 1 : // We are creating a new store. Create the format version marker file.
593 1 : if err := d.writeFormatVersionMarker(d.FormatMajorVersion()); err != nil {
594 0 : return nil, err
595 0 : }
596 : }
597 :
598 : // Write the current options to disk.
599 1 : d.optionsFileNum = d.mu.versions.getNextDiskFileNum()
600 1 : tmpPath := base.MakeFilepath(opts.FS, dirname, base.FileTypeTemp, d.optionsFileNum)
601 1 : optionsPath := base.MakeFilepath(opts.FS, dirname, base.FileTypeOptions, d.optionsFileNum)
602 1 :
603 1 : // Write them to a temporary file first, in case we crash before
604 1 : // we're done. A corrupt options file prevents opening the
605 1 : // database.
606 1 : optionsFile, err := opts.FS.Create(tmpPath, vfs.WriteCategoryUnspecified)
607 1 : if err != nil {
608 0 : return nil, err
609 0 : }
610 1 : serializedOpts := []byte(opts.String())
611 1 : if _, err := optionsFile.Write(serializedOpts); err != nil {
612 0 : return nil, errors.CombineErrors(err, optionsFile.Close())
613 0 : }
614 1 : d.optionsFileSize = uint64(len(serializedOpts))
615 1 : if err := optionsFile.Sync(); err != nil {
616 0 : return nil, errors.CombineErrors(err, optionsFile.Close())
617 0 : }
618 1 : if err := optionsFile.Close(); err != nil {
619 0 : return nil, err
620 0 : }
621 : // Atomically rename to the OPTIONS-XXXXXX path. This rename is
622 : // guaranteed to be atomic because the destination path does not
623 : // exist.
624 1 : if err := opts.FS.Rename(tmpPath, optionsPath); err != nil {
625 0 : return nil, err
626 0 : }
627 1 : if err := d.dataDir.Sync(); err != nil {
628 0 : return nil, err
629 0 : }
630 : }
631 :
632 1 : if !d.opts.ReadOnly {
633 1 : // Get a fresh list of files, in case some of the earlier flushes/compactions
634 1 : // have deleted some files.
635 1 : ls, err := opts.FS.List(dirname)
636 1 : if err != nil {
637 0 : return nil, err
638 0 : }
639 1 : d.scanObsoleteFiles(ls, flushableIngests)
640 1 : d.deleteObsoleteFiles(jobID)
641 : }
642 : // Else, nothing is obsolete.
643 :
644 1 : d.mu.tableStats.cond.L = &d.mu.Mutex
645 1 : d.mu.tableValidation.cond.L = &d.mu.Mutex
646 1 : if !d.opts.ReadOnly {
647 1 : d.maybeCollectTableStatsLocked()
648 1 : }
649 1 : d.calculateDiskAvailableBytes()
650 1 :
651 1 : d.maybeScheduleFlush()
652 1 : d.maybeScheduleCompaction()
653 1 :
654 1 : // Note: this is a no-op if invariants are disabled or race is enabled.
655 1 : //
656 1 : // Setting a finalizer on *DB causes *DB to never be reclaimed and the
657 1 : // finalizer to never be run. The problem is due to this limitation of
658 1 : // finalizers mention in the SetFinalizer docs:
659 1 : //
660 1 : // If a cyclic structure includes a block with a finalizer, that cycle is
661 1 : // not guaranteed to be garbage collected and the finalizer is not
662 1 : // guaranteed to run, because there is no ordering that respects the
663 1 : // dependencies.
664 1 : //
665 1 : // DB has cycles with several of its internal structures: readState,
666 1 : // newIters, fileCache, versions, etc. Each of this individually cause a
667 1 : // cycle and prevent the finalizer from being run. But we can workaround this
668 1 : // finializer limitation by setting a finalizer on another object that is
669 1 : // tied to the lifetime of DB: the DB.closed atomic.Value.
670 1 : dPtr := fmt.Sprintf("%p", d)
671 1 : invariants.SetFinalizer(d.closed, func(obj interface{}) {
672 0 : v := obj.(*atomic.Value)
673 0 : if err := v.Load(); err == nil {
674 0 : fmt.Fprintf(os.Stderr, "%s: unreferenced DB not closed\n", dPtr)
675 0 : os.Exit(1)
676 0 : }
677 : })
678 :
679 1 : return d, nil
680 : }
681 :
682 : // prepareAndOpenDirs opens the directories for the store (and creates them if
683 : // necessary).
684 : //
685 : // Returns an error if ReadOnly is set and the directories don't exist.
686 : func prepareAndOpenDirs(
687 : dirname string, opts *Options,
688 1 : ) (walDirname string, secondaryWalDirName string, dataDir vfs.File, err error) {
689 1 : walDirname = dirname
690 1 : if opts.WALDir != "" {
691 1 : walDirname = resolveStorePath(dirname, opts.WALDir)
692 1 : }
693 1 : if opts.WALFailover != nil {
694 1 : secondaryWalDirName = resolveStorePath(dirname, opts.WALFailover.Secondary.Dirname)
695 1 : }
696 :
697 : // Create directories if needed.
698 1 : if !opts.ReadOnly {
699 1 : f, err := mkdirAllAndSyncParents(opts.FS, dirname)
700 1 : if err != nil {
701 0 : return "", "", nil, err
702 0 : }
703 1 : f.Close()
704 1 : if walDirname != dirname {
705 1 : f, err := mkdirAllAndSyncParents(opts.FS, walDirname)
706 1 : if err != nil {
707 0 : return "", "", nil, err
708 0 : }
709 1 : f.Close()
710 : }
711 1 : if opts.WALFailover != nil {
712 1 : f, err := mkdirAllAndSyncParents(opts.WALFailover.Secondary.FS, secondaryWalDirName)
713 1 : if err != nil {
714 0 : return "", "", nil, err
715 0 : }
716 1 : f.Close()
717 : }
718 : }
719 :
720 1 : dataDir, err = opts.FS.OpenDir(dirname)
721 1 : if err != nil {
722 0 : if opts.ReadOnly && oserror.IsNotExist(err) {
723 0 : return "", "", nil, errors.Errorf("pebble: database %q does not exist", dirname)
724 0 : }
725 0 : return "", "", nil, err
726 : }
727 1 : if opts.ReadOnly && walDirname != dirname {
728 0 : // Check that the wal dir exists.
729 0 : walDir, err := opts.FS.OpenDir(walDirname)
730 0 : if err != nil {
731 0 : dataDir.Close()
732 0 : return "", "", nil, err
733 0 : }
734 0 : walDir.Close()
735 : }
736 :
737 1 : return walDirname, secondaryWalDirName, dataDir, nil
738 : }
739 :
740 : // GetVersion returns the engine version string from the latest options
741 : // file present in dir. Used to check what Pebble or RocksDB version was last
742 : // used to write to the database stored in this directory. An empty string is
743 : // returned if no valid OPTIONS file with a version key was found.
744 0 : func GetVersion(dir string, fs vfs.FS) (string, error) {
745 0 : ls, err := fs.List(dir)
746 0 : if err != nil {
747 0 : return "", err
748 0 : }
749 0 : var version string
750 0 : lastOptionsSeen := base.DiskFileNum(0)
751 0 : for _, filename := range ls {
752 0 : ft, fn, ok := base.ParseFilename(fs, filename)
753 0 : if !ok {
754 0 : continue
755 : }
756 0 : switch ft {
757 0 : case base.FileTypeOptions:
758 0 : // If this file has a higher number than the last options file
759 0 : // processed, reset version. This is because rocksdb often
760 0 : // writes multiple options files without deleting previous ones.
761 0 : // Otherwise, skip parsing this options file.
762 0 : if fn > lastOptionsSeen {
763 0 : version = ""
764 0 : lastOptionsSeen = fn
765 0 : } else {
766 0 : continue
767 : }
768 0 : f, err := fs.Open(fs.PathJoin(dir, filename))
769 0 : if err != nil {
770 0 : return "", err
771 0 : }
772 0 : data, err := io.ReadAll(f)
773 0 : f.Close()
774 0 :
775 0 : if err != nil {
776 0 : return "", err
777 0 : }
778 0 : err = parseOptions(string(data), parseOptionsFuncs{
779 0 : visitKeyValue: func(i, j int, section, key, value string) error {
780 0 : switch {
781 0 : case section == "Version":
782 0 : switch key {
783 0 : case "pebble_version":
784 0 : version = value
785 0 : case "rocksdb_version":
786 0 : version = fmt.Sprintf("rocksdb v%s", value)
787 : }
788 : }
789 0 : return nil
790 : },
791 : })
792 0 : if err != nil {
793 0 : return "", err
794 0 : }
795 : }
796 : }
797 0 : return version, nil
798 : }
799 :
800 : func (d *DB) replayIngestedFlushable(
801 : b *Batch, logNum base.DiskFileNum,
802 1 : ) (entry *flushableEntry, err error) {
803 1 : br := b.Reader()
804 1 : seqNum := b.SeqNum()
805 1 :
806 1 : fileNums := make([]base.DiskFileNum, 0, b.Count())
807 1 : var exciseSpan KeyRange
808 1 : addFileNum := func(encodedFileNum []byte) {
809 1 : fileNum, n := binary.Uvarint(encodedFileNum)
810 1 : if n <= 0 {
811 0 : panic("pebble: ingest sstable file num is invalid")
812 : }
813 1 : fileNums = append(fileNums, base.DiskFileNum(fileNum))
814 : }
815 :
816 1 : for i := 0; i < int(b.Count()); i++ {
817 1 : kind, key, val, ok, err := br.Next()
818 1 : if err != nil {
819 0 : return nil, err
820 0 : }
821 1 : if kind != InternalKeyKindIngestSST && kind != InternalKeyKindExcise {
822 0 : panic("pebble: invalid batch key kind")
823 : }
824 1 : if !ok {
825 0 : panic("pebble: invalid batch count")
826 : }
827 1 : if kind == base.InternalKeyKindExcise {
828 1 : if exciseSpan.Valid() {
829 0 : panic("pebble: multiple excise spans in a single batch")
830 : }
831 1 : exciseSpan.Start = slices.Clone(key)
832 1 : exciseSpan.End = slices.Clone(val)
833 1 : continue
834 : }
835 1 : addFileNum(key)
836 : }
837 :
838 1 : if _, _, _, ok, err := br.Next(); err != nil {
839 0 : return nil, err
840 1 : } else if ok {
841 0 : panic("pebble: invalid number of entries in batch")
842 : }
843 :
844 1 : meta := make([]*manifest.TableMetadata, len(fileNums))
845 1 : var lastRangeKey keyspan.Span
846 1 : for i, n := range fileNums {
847 1 : readable, err := d.objProvider.OpenForReading(context.TODO(), base.FileTypeTable, n,
848 1 : objstorage.OpenOptions{MustExist: true})
849 1 : if err != nil {
850 0 : return nil, errors.Wrap(err, "pebble: error when opening flushable ingest files")
851 0 : }
852 : // NB: ingestLoad1 will close readable.
853 1 : meta[i], lastRangeKey, err = ingestLoad1(context.TODO(), d.opts, d.FormatMajorVersion(),
854 1 : readable, d.cacheHandle, base.PhysicalTableFileNum(n), disableRangeKeyChecks())
855 1 : if err != nil {
856 0 : return nil, errors.Wrap(err, "pebble: error when loading flushable ingest files")
857 0 : }
858 : }
859 1 : if lastRangeKey.Valid() && d.opts.Comparer.Split.HasSuffix(lastRangeKey.End) {
860 0 : return nil, errors.AssertionFailedf("pebble: last ingest sstable has suffixed range key end %s",
861 0 : d.opts.Comparer.FormatKey(lastRangeKey.End))
862 0 : }
863 :
864 1 : numFiles := len(meta)
865 1 : if exciseSpan.Valid() {
866 1 : numFiles++
867 1 : }
868 1 : if uint32(numFiles) != b.Count() {
869 0 : panic("pebble: couldn't load all files in WAL entry")
870 : }
871 :
872 1 : return d.newIngestedFlushableEntry(meta, seqNum, logNum, exciseSpan)
873 : }
874 :
875 : // replayWAL replays the edits in the specified WAL. If the DB is in read
876 : // only mode, then the WALs are replayed into memtables and not flushed. If
877 : // the DB is not in read only mode, then the contents of the WAL are
878 : // guaranteed to be flushed when a flush is scheduled after this method is run.
879 : // Note that this flushing is very important for guaranteeing durability:
880 : // the application may have had a number of pending
881 : // fsyncs to the WAL before the process crashed, and those fsyncs may not have
882 : // happened but the corresponding data may now be readable from the WAL (while
883 : // sitting in write-back caches in the kernel or the storage device). By
884 : // reading the WAL (including the non-fsynced data) and then flushing all
885 : // these changes (flush does fsyncs), we are able to guarantee that the
886 : // initial state of the DB is durable.
887 : //
888 : // This method mutates d.mu.mem.queue and possibly d.mu.mem.mutable and replays
889 : // WALs into the flushable queue. Flushing of the queue is expected to be handled
890 : // by callers. A list of flushable ingests (but not memtables) replayed is returned.
891 : //
892 : // d.mu must be held when calling this, but the mutex may be dropped and
893 : // re-acquired during the course of this method.
894 : func (d *DB) replayWAL(
895 : jobID JobID, ll wal.LogicalLog, strictWALTail bool,
896 1 : ) (flushableIngests []*ingestedFlushable, maxSeqNum base.SeqNum, err error) {
897 1 : rr := ll.OpenForRead()
898 1 : defer func() { _ = rr.Close() }()
899 1 : var (
900 1 : b Batch
901 1 : buf bytes.Buffer
902 1 : mem *memTable
903 1 : entry *flushableEntry
904 1 : offset wal.Offset
905 1 : lastFlushOffset int64
906 1 : keysReplayed int64 // number of keys replayed
907 1 : batchesReplayed int64 // number of batches replayed
908 1 : )
909 1 :
910 1 : // TODO(jackson): This function is interspersed with panics, in addition to
911 1 : // corruption error propagation. Audit them to ensure we're truly only
912 1 : // panicking where the error points to Pebble bug and not user or
913 1 : // hardware-induced corruption.
914 1 :
915 1 : // "Flushes" (ie. closes off) the current memtable, if not nil.
916 1 : flushMem := func() {
917 1 : if mem == nil {
918 1 : return
919 1 : }
920 1 : mem.writerUnref()
921 1 : if d.mu.mem.mutable == mem {
922 1 : d.mu.mem.mutable = nil
923 1 : }
924 1 : entry.flushForced = !d.opts.ReadOnly
925 1 : var logSize uint64
926 1 : mergedOffset := offset.Physical + offset.PreviousFilesBytes
927 1 : if mergedOffset >= lastFlushOffset {
928 1 : logSize = uint64(mergedOffset - lastFlushOffset)
929 1 : }
930 : // Else, this was the initial memtable in the read-only case which must have
931 : // been empty, but we need to flush it since we don't want to add to it later.
932 1 : lastFlushOffset = mergedOffset
933 1 : entry.logSize = logSize
934 1 : mem, entry = nil, nil
935 : }
936 :
937 1 : mem = d.mu.mem.mutable
938 1 : if mem != nil {
939 1 : entry = d.mu.mem.queue[len(d.mu.mem.queue)-1]
940 1 : if !d.opts.ReadOnly {
941 1 : flushMem()
942 1 : }
943 : }
944 :
945 : // Creates a new memtable if there is no current memtable.
946 1 : ensureMem := func(seqNum base.SeqNum) {
947 1 : if mem != nil {
948 1 : return
949 1 : }
950 1 : mem, entry = d.newMemTable(base.DiskFileNum(ll.Num), seqNum, 0 /* minSize */)
951 1 : d.mu.mem.mutable = mem
952 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
953 : }
954 :
955 1 : defer func() {
956 1 : if err != nil {
957 0 : err = errors.WithDetailf(err, "replaying wal %d, offset %s", ll.Num, offset)
958 0 : }
959 : }()
960 :
961 1 : for {
962 1 : var r io.Reader
963 1 : var err error
964 1 : r, offset, err = rr.NextRecord()
965 1 : if err == nil {
966 1 : _, err = io.Copy(&buf, r)
967 1 : }
968 1 : if err != nil {
969 1 : // It is common to encounter a zeroed or invalid chunk due to WAL
970 1 : // preallocation and WAL recycling. However zeroed or invalid chunks
971 1 : // can also be a consequence of corruption / disk rot. When the log
972 1 : // reader encounters one of these cases, it attempts to disambiguate
973 1 : // by reading ahead looking for a future record. If a future chunk
974 1 : // indicates the chunk at the original offset should've been valid, it
975 1 : // surfaces record.ErrInvalidChunk or record.ErrZeroedChunk. These
976 1 : // errors are always indicative of corruption and data loss.
977 1 : //
978 1 : // Otherwise, the reader surfaces record.ErrUnexpectedEOF indicating
979 1 : // that the WAL terminated uncleanly and ambiguously. If the WAL is
980 1 : // the most recent logical WAL, the caller passes in
981 1 : // (strictWALTail=false), indicating we should tolerate the unclean
982 1 : // ending. If the WAL is an older WAL, the caller passes in
983 1 : // (strictWALTail=true), indicating that the WAL should have been
984 1 : // closed cleanly, and we should interpret the
985 1 : // `record.ErrUnexpectedEOF` as corruption and stop recovery.
986 1 : if errors.Is(err, io.EOF) {
987 1 : break
988 0 : } else if errors.Is(err, record.ErrUnexpectedEOF) && !strictWALTail {
989 0 : break
990 0 : } else if (errors.Is(err, record.ErrUnexpectedEOF) && strictWALTail) ||
991 0 : errors.Is(err, record.ErrInvalidChunk) || errors.Is(err, record.ErrZeroedChunk) {
992 0 : // If a read-ahead returns record.ErrInvalidChunk or
993 0 : // record.ErrZeroedChunk, then there's definitively corruption.
994 0 : //
995 0 : // If strictWALTail=true, then record.ErrUnexpectedEOF should
996 0 : // also be considered corruption because the strictWALTail
997 0 : // indicates we expect a clean end to the WAL.
998 0 : //
999 0 : // Other I/O related errors should not be marked with corruption
1000 0 : // and simply returned.
1001 0 : err = errors.Mark(err, ErrCorruption)
1002 0 : }
1003 :
1004 0 : return nil, 0, errors.Wrap(err, "pebble: error when replaying WAL")
1005 : }
1006 :
1007 1 : if buf.Len() < batchrepr.HeaderLen {
1008 0 : return nil, 0, base.CorruptionErrorf("pebble: corrupt wal %s (offset %s)",
1009 0 : errors.Safe(base.DiskFileNum(ll.Num)), offset)
1010 0 : }
1011 :
1012 1 : if d.opts.ErrorIfNotPristine {
1013 0 : return nil, 0, errors.WithDetailf(ErrDBNotPristine, "location: %q", d.dirname)
1014 0 : }
1015 :
1016 : // Specify Batch.db so that Batch.SetRepr will compute Batch.memTableSize
1017 : // which is used below.
1018 1 : b = Batch{}
1019 1 : b.db = d
1020 1 : if err := b.SetRepr(buf.Bytes()); err != nil {
1021 0 : return nil, 0, err
1022 0 : }
1023 1 : seqNum := b.SeqNum()
1024 1 : maxSeqNum = seqNum + base.SeqNum(b.Count())
1025 1 : keysReplayed += int64(b.Count())
1026 1 : batchesReplayed++
1027 1 : {
1028 1 : br := b.Reader()
1029 1 : if kind, _, _, ok, err := br.Next(); err != nil {
1030 0 : return nil, 0, err
1031 1 : } else if ok && (kind == InternalKeyKindIngestSST || kind == InternalKeyKindExcise) {
1032 1 : // We're in the flushable ingests (+ possibly excises) case.
1033 1 : //
1034 1 : // Ingests require an up-to-date view of the LSM to determine the target
1035 1 : // level of ingested sstables, and to accurately compute excises. Instead of
1036 1 : // doing an ingest in this function, we just enqueue a flushable ingest
1037 1 : // in the flushables queue and run a regular flush.
1038 1 : flushMem()
1039 1 : // mem is nil here.
1040 1 : entry, err = d.replayIngestedFlushable(&b, base.DiskFileNum(ll.Num))
1041 1 : if err != nil {
1042 0 : return nil, 0, err
1043 0 : }
1044 1 : fi := entry.flushable.(*ingestedFlushable)
1045 1 : flushableIngests = append(flushableIngests, fi)
1046 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
1047 1 : // A flushable ingest is always followed by a WAL rotation.
1048 1 : break
1049 : }
1050 : }
1051 :
1052 1 : if b.memTableSize >= uint64(d.largeBatchThreshold) {
1053 1 : flushMem()
1054 1 : // Make a copy of the data slice since it is currently owned by buf and will
1055 1 : // be reused in the next iteration.
1056 1 : b.data = slices.Clone(b.data)
1057 1 : b.flushable, err = newFlushableBatch(&b, d.opts.Comparer)
1058 1 : if err != nil {
1059 0 : return nil, 0, err
1060 0 : }
1061 1 : entry := d.newFlushableEntry(b.flushable, base.DiskFileNum(ll.Num), b.SeqNum())
1062 1 : // Disable memory accounting by adding a reader ref that will never be
1063 1 : // removed.
1064 1 : entry.readerRefs.Add(1)
1065 1 : d.mu.mem.queue = append(d.mu.mem.queue, entry)
1066 1 : } else {
1067 1 : ensureMem(seqNum)
1068 1 : if err = mem.prepare(&b); err != nil && err != arenaskl.ErrArenaFull {
1069 0 : return nil, 0, err
1070 0 : }
1071 : // We loop since DB.newMemTable() slowly grows the size of allocated memtables, so the
1072 : // batch may not initially fit, but will eventually fit (since it is smaller than
1073 : // largeBatchThreshold).
1074 1 : for err == arenaskl.ErrArenaFull {
1075 1 : flushMem()
1076 1 : ensureMem(seqNum)
1077 1 : err = mem.prepare(&b)
1078 1 : if err != nil && err != arenaskl.ErrArenaFull {
1079 0 : return nil, 0, err
1080 0 : }
1081 : }
1082 1 : if err = mem.apply(&b, seqNum); err != nil {
1083 0 : return nil, 0, err
1084 0 : }
1085 1 : mem.writerUnref()
1086 : }
1087 1 : buf.Reset()
1088 : }
1089 :
1090 1 : d.opts.Logger.Infof("[JOB %d] WAL %s stopped reading at offset: %s; replayed %d keys in %d batches",
1091 1 : jobID, ll.String(), offset, keysReplayed, batchesReplayed)
1092 1 : if !d.opts.ReadOnly {
1093 1 : flushMem()
1094 1 : }
1095 :
1096 : // mem is nil here, if !ReadOnly.
1097 1 : return flushableIngests, maxSeqNum, err
1098 : }
1099 :
1100 1 : func readOptionsFile(opts *Options, path string) (string, error) {
1101 1 : f, err := opts.FS.Open(path)
1102 1 : if err != nil {
1103 0 : return "", err
1104 0 : }
1105 1 : defer f.Close()
1106 1 :
1107 1 : data, err := io.ReadAll(f)
1108 1 : if err != nil {
1109 0 : return "", err
1110 0 : }
1111 1 : return string(data), nil
1112 : }
1113 :
1114 : // DBDesc briefly describes high-level state about a database.
1115 : type DBDesc struct {
1116 : // Exists is true if an existing database was found.
1117 : Exists bool
1118 : // FormatMajorVersion indicates the database's current format
1119 : // version.
1120 : FormatMajorVersion FormatMajorVersion
1121 : // ManifestFilename is the filename of the current active manifest,
1122 : // if the database exists.
1123 : ManifestFilename string
1124 : // OptionsFilename is the filename of the most recent OPTIONS file, if it
1125 : // exists.
1126 : OptionsFilename string
1127 : }
1128 :
1129 : // String implements fmt.Stringer.
1130 0 : func (d *DBDesc) String() string {
1131 0 : if !d.Exists {
1132 0 : return "uninitialized"
1133 0 : }
1134 0 : var buf bytes.Buffer
1135 0 : fmt.Fprintf(&buf, "initialized at format major version %s\n", d.FormatMajorVersion)
1136 0 : fmt.Fprintf(&buf, "manifest: %s\n", d.ManifestFilename)
1137 0 : fmt.Fprintf(&buf, "options: %s", d.OptionsFilename)
1138 0 : return buf.String()
1139 : }
1140 :
1141 : // Peek looks for an existing database in dirname on the provided FS. It
1142 : // returns a brief description of the database. Peek is read-only and
1143 : // does not open the database
1144 0 : func Peek(dirname string, fs vfs.FS) (*DBDesc, error) {
1145 0 : ls, err := fs.List(dirname)
1146 0 : if err != nil {
1147 0 : return nil, err
1148 0 : }
1149 :
1150 0 : vers, versMarker, err := lookupFormatMajorVersion(fs, dirname, ls)
1151 0 : if err != nil {
1152 0 : return nil, err
1153 0 : }
1154 : // TODO(jackson): Immediately closing the marker is clunky. Add a
1155 : // PeekMarker variant that avoids opening the directory.
1156 0 : if err := versMarker.Close(); err != nil {
1157 0 : return nil, err
1158 0 : }
1159 :
1160 : // Find the currently active manifest, if there is one.
1161 0 : manifestMarker, manifestFileNum, exists, err := findCurrentManifest(fs, dirname, ls)
1162 0 : if err != nil {
1163 0 : return nil, err
1164 0 : }
1165 : // TODO(jackson): Immediately closing the marker is clunky. Add a
1166 : // PeekMarker variant that avoids opening the directory.
1167 0 : if err := manifestMarker.Close(); err != nil {
1168 0 : return nil, err
1169 0 : }
1170 :
1171 0 : desc := &DBDesc{
1172 0 : Exists: exists,
1173 0 : FormatMajorVersion: vers,
1174 0 : }
1175 0 :
1176 0 : // Find the OPTIONS file with the highest file number within the list of
1177 0 : // directory entries.
1178 0 : var previousOptionsFileNum base.DiskFileNum
1179 0 : for _, filename := range ls {
1180 0 : ft, fn, ok := base.ParseFilename(fs, filename)
1181 0 : if !ok || ft != base.FileTypeOptions || fn < previousOptionsFileNum {
1182 0 : continue
1183 : }
1184 0 : previousOptionsFileNum = fn
1185 0 : desc.OptionsFilename = fs.PathJoin(dirname, filename)
1186 : }
1187 :
1188 0 : if exists {
1189 0 : desc.ManifestFilename = base.MakeFilepath(fs, dirname, base.FileTypeManifest, manifestFileNum)
1190 0 : }
1191 0 : return desc, nil
1192 : }
1193 :
1194 : // ErrDBDoesNotExist is generated when ErrorIfNotExists is set and the database
1195 : // does not exist.
1196 : //
1197 : // Note that errors can be wrapped with more details; use errors.Is().
1198 : var ErrDBDoesNotExist = errors.New("pebble: database does not exist")
1199 :
1200 : // ErrDBAlreadyExists is generated when ErrorIfExists is set and the database
1201 : // already exists.
1202 : //
1203 : // Note that errors can be wrapped with more details; use errors.Is().
1204 : var ErrDBAlreadyExists = errors.New("pebble: database already exists")
1205 :
1206 : // ErrDBNotPristine is generated when ErrorIfNotPristine is set and the database
1207 : // already exists and is not pristine.
1208 : //
1209 : // Note that errors can be wrapped with more details; use errors.Is().
1210 : var ErrDBNotPristine = errors.New("pebble: database already exists and is not pristine")
1211 :
1212 1 : func checkConsistency(v *manifest.Version, objProvider objstorage.Provider) error {
1213 1 : var errs []error
1214 1 : dedup := make(map[base.DiskFileNum]struct{})
1215 1 : for level, files := range v.Levels {
1216 1 : for f := range files.All() {
1217 1 : backingState := f.TableBacking
1218 1 : if _, ok := dedup[backingState.DiskFileNum]; ok {
1219 1 : continue
1220 : }
1221 1 : dedup[backingState.DiskFileNum] = struct{}{}
1222 1 : fileNum := backingState.DiskFileNum
1223 1 : fileSize := backingState.Size
1224 1 : // We skip over remote objects; those are instead checked asynchronously
1225 1 : // by the table stats loading job.
1226 1 : meta, err := objProvider.Lookup(base.FileTypeTable, fileNum)
1227 1 : var size int64
1228 1 : if err == nil {
1229 1 : if meta.IsRemote() {
1230 1 : continue
1231 : }
1232 1 : size, err = objProvider.Size(meta)
1233 : }
1234 1 : if err != nil {
1235 0 : errs = append(errs, errors.Wrapf(err, "L%d: %s", errors.Safe(level), fileNum))
1236 0 : continue
1237 : }
1238 :
1239 1 : if size != int64(fileSize) {
1240 0 : errs = append(errs, errors.Errorf(
1241 0 : "L%d: %s: object size mismatch (%s): %d (disk) != %d (MANIFEST)",
1242 0 : errors.Safe(level), fileNum, objProvider.Path(meta),
1243 0 : errors.Safe(size), errors.Safe(fileSize)))
1244 0 : continue
1245 : }
1246 : }
1247 : }
1248 1 : return errors.Join(errs...)
1249 : }
1250 :
1251 : type walEventListenerAdaptor struct {
1252 : l *EventListener
1253 : }
1254 :
1255 1 : func (l walEventListenerAdaptor) LogCreated(ci wal.CreateInfo) {
1256 1 : // TODO(sumeer): extend WALCreateInfo for the failover case in case the path
1257 1 : // is insufficient to infer whether primary or secondary.
1258 1 : wci := WALCreateInfo{
1259 1 : JobID: ci.JobID,
1260 1 : Path: ci.Path,
1261 1 : FileNum: base.DiskFileNum(ci.Num),
1262 1 : RecycledFileNum: ci.RecycledFileNum,
1263 1 : Err: ci.Err,
1264 1 : }
1265 1 : l.l.WALCreated(wci)
1266 1 : }
|
