Line data Source code
1 : // Copyright 2011 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 : "fmt"
10 : "io"
11 : "runtime"
12 : "strconv"
13 : "strings"
14 : "time"
15 : "unicode"
16 :
17 : "github.com/cockroachdb/errors"
18 : "github.com/cockroachdb/fifo"
19 : "github.com/cockroachdb/pebble/internal/base"
20 : "github.com/cockroachdb/pebble/internal/cache"
21 : "github.com/cockroachdb/pebble/internal/humanize"
22 : "github.com/cockroachdb/pebble/internal/keyspan"
23 : "github.com/cockroachdb/pebble/internal/manifest"
24 : "github.com/cockroachdb/pebble/objstorage/objstorageprovider"
25 : "github.com/cockroachdb/pebble/objstorage/remote"
26 : "github.com/cockroachdb/pebble/rangekey"
27 : "github.com/cockroachdb/pebble/sstable"
28 : "github.com/cockroachdb/pebble/sstable/block"
29 : "github.com/cockroachdb/pebble/vfs"
30 : "github.com/cockroachdb/pebble/wal"
31 : )
32 :
33 : const (
34 : cacheDefaultSize = 8 << 20 // 8 MB
35 : defaultLevelMultiplier = 10
36 : )
37 :
38 : // Compression exports the base.Compression type.
39 : type Compression = block.Compression
40 :
41 : // Exported Compression constants.
42 : const (
43 : DefaultCompression = block.DefaultCompression
44 : NoCompression = block.NoCompression
45 : SnappyCompression = block.SnappyCompression
46 : ZstdCompression = block.ZstdCompression
47 : )
48 :
49 : // FilterType exports the base.FilterType type.
50 : type FilterType = base.FilterType
51 :
52 : // Exported TableFilter constants.
53 : const (
54 : TableFilter = base.TableFilter
55 : )
56 :
57 : // FilterWriter exports the base.FilterWriter type.
58 : type FilterWriter = base.FilterWriter
59 :
60 : // FilterPolicy exports the base.FilterPolicy type.
61 : type FilterPolicy = base.FilterPolicy
62 :
63 : // BlockPropertyCollector exports the sstable.BlockPropertyCollector type.
64 : type BlockPropertyCollector = sstable.BlockPropertyCollector
65 :
66 : // BlockPropertyFilter exports the sstable.BlockPropertyFilter type.
67 : type BlockPropertyFilter = base.BlockPropertyFilter
68 :
69 : // ShortAttributeExtractor exports the base.ShortAttributeExtractor type.
70 : type ShortAttributeExtractor = base.ShortAttributeExtractor
71 :
72 : // UserKeyPrefixBound exports the sstable.UserKeyPrefixBound type.
73 : type UserKeyPrefixBound = sstable.UserKeyPrefixBound
74 :
75 : // IterKeyType configures which types of keys an iterator should surface.
76 : type IterKeyType int8
77 :
78 : const (
79 : // IterKeyTypePointsOnly configures an iterator to iterate over point keys
80 : // only.
81 : IterKeyTypePointsOnly IterKeyType = iota
82 : // IterKeyTypeRangesOnly configures an iterator to iterate over range keys
83 : // only.
84 : IterKeyTypeRangesOnly
85 : // IterKeyTypePointsAndRanges configures an iterator iterate over both point
86 : // keys and range keys simultaneously.
87 : IterKeyTypePointsAndRanges
88 : )
89 :
90 : // String implements fmt.Stringer.
91 1 : func (t IterKeyType) String() string {
92 1 : switch t {
93 1 : case IterKeyTypePointsOnly:
94 1 : return "points-only"
95 1 : case IterKeyTypeRangesOnly:
96 1 : return "ranges-only"
97 1 : case IterKeyTypePointsAndRanges:
98 1 : return "points-and-ranges"
99 0 : default:
100 0 : panic(fmt.Sprintf("unknown key type %d", t))
101 : }
102 : }
103 :
104 : // IterOptions hold the optional per-query parameters for NewIter.
105 : //
106 : // Like Options, a nil *IterOptions is valid and means to use the default
107 : // values.
108 : type IterOptions struct {
109 : // LowerBound specifies the smallest key (inclusive) that the iterator will
110 : // return during iteration. If the iterator is seeked or iterated past this
111 : // boundary the iterator will return Valid()==false. Setting LowerBound
112 : // effectively truncates the key space visible to the iterator.
113 : LowerBound []byte
114 : // UpperBound specifies the largest key (exclusive) that the iterator will
115 : // return during iteration. If the iterator is seeked or iterated past this
116 : // boundary the iterator will return Valid()==false. Setting UpperBound
117 : // effectively truncates the key space visible to the iterator.
118 : UpperBound []byte
119 : // SkipPoint may be used to skip over point keys that don't match an
120 : // arbitrary predicate during iteration. If set, the Iterator invokes
121 : // SkipPoint for keys encountered. If SkipPoint returns true, the iterator
122 : // will skip the key without yielding it to the iterator operation in
123 : // progress.
124 : //
125 : // SkipPoint must be a pure function and always return the same result when
126 : // provided the same arguments. The iterator may call SkipPoint multiple
127 : // times for the same user key.
128 : SkipPoint func(userKey []byte) bool
129 : // PointKeyFilters can be used to avoid scanning tables and blocks in tables
130 : // when iterating over point keys. This slice represents an intersection
131 : // across all filters, i.e., all filters must indicate that the block is
132 : // relevant.
133 : //
134 : // Performance note: When len(PointKeyFilters) > 0, the caller should ensure
135 : // that cap(PointKeyFilters) is at least len(PointKeyFilters)+1. This helps
136 : // avoid allocations in Pebble internal code that mutates the slice.
137 : PointKeyFilters []BlockPropertyFilter
138 : // RangeKeyFilters can be usefd to avoid scanning tables and blocks in tables
139 : // when iterating over range keys. The same requirements that apply to
140 : // PointKeyFilters apply here too.
141 : RangeKeyFilters []BlockPropertyFilter
142 : // KeyTypes configures which types of keys to iterate over: point keys,
143 : // range keys, or both.
144 : KeyTypes IterKeyType
145 : // RangeKeyMasking can be used to enable automatic masking of point keys by
146 : // range keys. Range key masking is only supported during combined range key
147 : // and point key iteration mode (IterKeyTypePointsAndRanges).
148 : RangeKeyMasking RangeKeyMasking
149 :
150 : // OnlyReadGuaranteedDurable is an advanced option that is only supported by
151 : // the Reader implemented by DB. When set to true, only the guaranteed to be
152 : // durable state is visible in the iterator.
153 : // - This definition is made under the assumption that the FS implementation
154 : // is providing a durability guarantee when data is synced.
155 : // - The visible state represents a consistent point in the history of the
156 : // DB.
157 : // - The implementation is free to choose a conservative definition of what
158 : // is guaranteed durable. For simplicity, the current implementation
159 : // ignores memtables. A more sophisticated implementation could track the
160 : // highest seqnum that is synced to the WAL and published and use that as
161 : // the visible seqnum for an iterator. Note that the latter approach is
162 : // not strictly better than the former since we can have DBs that are (a)
163 : // synced more rarely than memtable flushes, (b) have no WAL. (a) is
164 : // likely to be true in a future CockroachDB context where the DB
165 : // containing the state machine may be rarely synced.
166 : // NB: this current implementation relies on the fact that memtables are
167 : // flushed in seqnum order, and any ingested sstables that happen to have a
168 : // lower seqnum than a non-flushed memtable don't have any overlapping keys.
169 : // This is the fundamental level invariant used in other code too, like when
170 : // merging iterators.
171 : //
172 : // Semantically, using this option provides the caller a "snapshot" as of
173 : // the time the most recent memtable was flushed. An alternate interface
174 : // would be to add a NewSnapshot variant. Creating a snapshot is heavier
175 : // weight than creating an iterator, so we have opted to support this
176 : // iterator option.
177 : OnlyReadGuaranteedDurable bool
178 : // UseL6Filters allows the caller to opt into reading filter blocks for L6
179 : // sstables. Helpful if a lot of SeekPrefixGEs are expected in quick
180 : // succession, that are also likely to not yield a single key. Filter blocks in
181 : // L6 can be relatively large, often larger than data blocks, so the benefit of
182 : // loading them in the cache is minimized if the probability of the key
183 : // existing is not low or if we just expect a one-time Seek (where loading the
184 : // data block directly is better).
185 : UseL6Filters bool
186 : // CategoryAndQoS is used for categorized iterator stats. This should not be
187 : // changed by calling SetOptions.
188 : sstable.CategoryAndQoS
189 :
190 : DebugRangeKeyStack bool
191 :
192 : // Internal options.
193 :
194 : logger Logger
195 : // Layer corresponding to this file. Only passed in if constructed by a
196 : // levelIter.
197 : layer manifest.Layer
198 : // disableLazyCombinedIteration is an internal testing option.
199 : disableLazyCombinedIteration bool
200 : // snapshotForHideObsoletePoints is specified for/by levelIter when opening
201 : // files and is used to decide whether to hide obsolete points. A value of 0
202 : // implies obsolete points should not be hidden.
203 : snapshotForHideObsoletePoints base.SeqNum
204 :
205 : // NB: If adding new Options, you must account for them in iterator
206 : // construction and Iterator.SetOptions.
207 : }
208 :
209 : // GetLowerBound returns the LowerBound or nil if the receiver is nil.
210 1 : func (o *IterOptions) GetLowerBound() []byte {
211 1 : if o == nil {
212 1 : return nil
213 1 : }
214 1 : return o.LowerBound
215 : }
216 :
217 : // GetUpperBound returns the UpperBound or nil if the receiver is nil.
218 1 : func (o *IterOptions) GetUpperBound() []byte {
219 1 : if o == nil {
220 1 : return nil
221 1 : }
222 1 : return o.UpperBound
223 : }
224 :
225 1 : func (o *IterOptions) pointKeys() bool {
226 1 : if o == nil {
227 0 : return true
228 0 : }
229 1 : return o.KeyTypes == IterKeyTypePointsOnly || o.KeyTypes == IterKeyTypePointsAndRanges
230 : }
231 :
232 1 : func (o *IterOptions) rangeKeys() bool {
233 1 : if o == nil {
234 0 : return false
235 0 : }
236 1 : return o.KeyTypes == IterKeyTypeRangesOnly || o.KeyTypes == IterKeyTypePointsAndRanges
237 : }
238 :
239 1 : func (o *IterOptions) getLogger() Logger {
240 1 : if o == nil || o.logger == nil {
241 1 : return DefaultLogger
242 1 : }
243 1 : return o.logger
244 : }
245 :
246 : // SpanIterOptions creates a SpanIterOptions from this IterOptions.
247 1 : func (o *IterOptions) SpanIterOptions() keyspan.SpanIterOptions {
248 1 : if o == nil {
249 1 : return keyspan.SpanIterOptions{}
250 1 : }
251 1 : return keyspan.SpanIterOptions{
252 1 : RangeKeyFilters: o.RangeKeyFilters,
253 1 : }
254 : }
255 :
256 : // scanInternalOptions is similar to IterOptions, meant for use with
257 : // scanInternalIterator.
258 : type scanInternalOptions struct {
259 : sstable.CategoryAndQoS
260 : IterOptions
261 :
262 : visitPointKey func(key *InternalKey, value LazyValue, iterInfo IteratorLevel) error
263 : visitRangeDel func(start, end []byte, seqNum SeqNum) error
264 : visitRangeKey func(start, end []byte, keys []rangekey.Key) error
265 : visitSharedFile func(sst *SharedSSTMeta) error
266 : visitExternalFile func(sst *ExternalFile) error
267 :
268 : // includeObsoleteKeys specifies whether keys shadowed by newer internal keys
269 : // are exposed. If false, only one internal key per user key is exposed.
270 : includeObsoleteKeys bool
271 :
272 : // rateLimitFunc is used to limit the amount of bytes read per second.
273 : rateLimitFunc func(key *InternalKey, value LazyValue) error
274 : }
275 :
276 : // RangeKeyMasking configures automatic hiding of point keys by range keys. A
277 : // non-nil Suffix enables range-key masking. When enabled, range keys with
278 : // suffixes ≥ Suffix behave as masks. All point keys that are contained within a
279 : // masking range key's bounds and have suffixes greater than the range key's
280 : // suffix are automatically skipped.
281 : //
282 : // Specifically, when configured with a RangeKeyMasking.Suffix _s_, and there
283 : // exists a range key with suffix _r_ covering a point key with suffix _p_, and
284 : //
285 : // _s_ ≤ _r_ < _p_
286 : //
287 : // then the point key is elided.
288 : //
289 : // Range-key masking may only be used when iterating over both point keys and
290 : // range keys with IterKeyTypePointsAndRanges.
291 : type RangeKeyMasking struct {
292 : // Suffix configures which range keys may mask point keys. Only range keys
293 : // that are defined at suffixes greater than or equal to Suffix will mask
294 : // point keys.
295 : Suffix []byte
296 : // Filter is an optional field that may be used to improve performance of
297 : // range-key masking through a block-property filter defined over key
298 : // suffixes. If non-nil, Filter is called by Pebble to construct a
299 : // block-property filter mask at iterator creation. The filter is used to
300 : // skip whole point-key blocks containing point keys with suffixes greater
301 : // than a covering range-key's suffix.
302 : //
303 : // To use this functionality, the caller must create and configure (through
304 : // Options.BlockPropertyCollectors) a block-property collector that records
305 : // the maxmimum suffix contained within a block. The caller then must write
306 : // and provide a BlockPropertyFilterMask implementation on that same
307 : // property. See the BlockPropertyFilterMask type for more information.
308 : Filter func() BlockPropertyFilterMask
309 : }
310 :
311 : // BlockPropertyFilterMask extends the BlockPropertyFilter interface for use
312 : // with range-key masking. Unlike an ordinary block property filter, a
313 : // BlockPropertyFilterMask's filtering criteria is allowed to change when Pebble
314 : // invokes its SetSuffix method.
315 : //
316 : // When a Pebble iterator steps into a range key's bounds and the range key has
317 : // a suffix greater than or equal to RangeKeyMasking.Suffix, the range key acts
318 : // as a mask. The masking range key hides all point keys that fall within the
319 : // range key's bounds and have suffixes > the range key's suffix. Without a
320 : // filter mask configured, Pebble performs this hiding by stepping through point
321 : // keys and comparing suffixes. If large numbers of point keys are masked, this
322 : // requires Pebble to load, iterate through and discard a large number of
323 : // sstable blocks containing masked point keys.
324 : //
325 : // If a block-property collector and a filter mask are configured, Pebble may
326 : // skip loading some point-key blocks altogether. If a block's keys are known to
327 : // all fall within the bounds of the masking range key and the block was
328 : // annotated by a block-property collector with the maximal suffix, Pebble can
329 : // ask the filter mask to compare the property to the current masking range
330 : // key's suffix. If the mask reports no intersection, the block may be skipped.
331 : //
332 : // If unsuffixed and suffixed keys are written to the database, care must be
333 : // taken to avoid unintentionally masking un-suffixed keys located in the same
334 : // block as suffixed keys. One solution is to interpret unsuffixed keys as
335 : // containing the maximal suffix value, ensuring that blocks containing
336 : // unsuffixed keys are always loaded.
337 : type BlockPropertyFilterMask interface {
338 : BlockPropertyFilter
339 :
340 : // SetSuffix configures the mask with the suffix of a range key. The filter
341 : // should return false from Intersects whenever it's provided with a
342 : // property encoding a block's minimum suffix that's greater (according to
343 : // Compare) than the provided suffix.
344 : SetSuffix(suffix []byte) error
345 : }
346 :
347 : // WriteOptions hold the optional per-query parameters for Set and Delete
348 : // operations.
349 : //
350 : // Like Options, a nil *WriteOptions is valid and means to use the default
351 : // values.
352 : type WriteOptions struct {
353 : // Sync is whether to sync writes through the OS buffer cache and down onto
354 : // the actual disk, if applicable. Setting Sync is required for durability of
355 : // individual write operations but can result in slower writes.
356 : //
357 : // If false, and the process or machine crashes, then a recent write may be
358 : // lost. This is due to the recently written data being buffered inside the
359 : // process running Pebble. This differs from the semantics of a write system
360 : // call in which the data is buffered in the OS buffer cache and would thus
361 : // survive a process crash.
362 : //
363 : // The default value is true.
364 : Sync bool
365 : }
366 :
367 : // Sync specifies the default write options for writes which synchronize to
368 : // disk.
369 : var Sync = &WriteOptions{Sync: true}
370 :
371 : // NoSync specifies the default write options for writes which do not
372 : // synchronize to disk.
373 : var NoSync = &WriteOptions{Sync: false}
374 :
375 : // GetSync returns the Sync value or true if the receiver is nil.
376 1 : func (o *WriteOptions) GetSync() bool {
377 1 : return o == nil || o.Sync
378 1 : }
379 :
380 : // LevelOptions holds the optional per-level parameters.
381 : type LevelOptions struct {
382 : // BlockRestartInterval is the number of keys between restart points
383 : // for delta encoding of keys.
384 : //
385 : // The default value is 16.
386 : BlockRestartInterval int
387 :
388 : // BlockSize is the target uncompressed size in bytes of each table block.
389 : //
390 : // The default value is 4096.
391 : BlockSize int
392 :
393 : // BlockSizeThreshold finishes a block if the block size is larger than the
394 : // specified percentage of the target block size and adding the next entry
395 : // would cause the block to be larger than the target block size.
396 : //
397 : // The default value is 90
398 : BlockSizeThreshold int
399 :
400 : // Compression defines the per-block compression to use.
401 : //
402 : // The default value (DefaultCompression) uses snappy compression.
403 : Compression func() Compression
404 :
405 : // FilterPolicy defines a filter algorithm (such as a Bloom filter) that can
406 : // reduce disk reads for Get calls.
407 : //
408 : // One such implementation is bloom.FilterPolicy(10) from the pebble/bloom
409 : // package.
410 : //
411 : // The default value means to use no filter.
412 : FilterPolicy FilterPolicy
413 :
414 : // FilterType defines whether an existing filter policy is applied at a
415 : // block-level or table-level. Block-level filters use less memory to create,
416 : // but are slower to access as a check for the key in the index must first be
417 : // performed to locate the filter block. A table-level filter will require
418 : // memory proportional to the number of keys in an sstable to create, but
419 : // avoids the index lookup when determining if a key is present. Table-level
420 : // filters should be preferred except under constrained memory situations.
421 : FilterType FilterType
422 :
423 : // IndexBlockSize is the target uncompressed size in bytes of each index
424 : // block. When the index block size is larger than this target, two-level
425 : // indexes are automatically enabled. Setting this option to a large value
426 : // (such as math.MaxInt32) disables the automatic creation of two-level
427 : // indexes.
428 : //
429 : // The default value is the value of BlockSize.
430 : IndexBlockSize int
431 :
432 : // The target file size for the level.
433 : TargetFileSize int64
434 : }
435 :
436 : // EnsureDefaults ensures that the default values for all of the options have
437 : // been initialized. It is valid to call EnsureDefaults on a nil receiver. A
438 : // non-nil result will always be returned.
439 1 : func (o *LevelOptions) EnsureDefaults() *LevelOptions {
440 1 : if o == nil {
441 0 : o = &LevelOptions{}
442 0 : }
443 1 : if o.BlockRestartInterval <= 0 {
444 1 : o.BlockRestartInterval = base.DefaultBlockRestartInterval
445 1 : }
446 1 : if o.BlockSize <= 0 {
447 1 : o.BlockSize = base.DefaultBlockSize
448 1 : } else if o.BlockSize > sstable.MaximumBlockSize {
449 0 : panic(errors.Errorf("BlockSize %d exceeds MaximumBlockSize", o.BlockSize))
450 : }
451 1 : if o.BlockSizeThreshold <= 0 {
452 1 : o.BlockSizeThreshold = base.DefaultBlockSizeThreshold
453 1 : }
454 1 : if o.Compression == nil {
455 1 : o.Compression = func() Compression { return DefaultCompression }
456 : }
457 1 : if o.IndexBlockSize <= 0 {
458 1 : o.IndexBlockSize = o.BlockSize
459 1 : }
460 1 : if o.TargetFileSize <= 0 {
461 1 : o.TargetFileSize = 2 << 20 // 2 MB
462 1 : }
463 1 : return o
464 : }
465 :
466 : // Options holds the optional parameters for configuring pebble. These options
467 : // apply to the DB at large; per-query options are defined by the IterOptions
468 : // and WriteOptions types.
469 : type Options struct {
470 : // Sync sstables periodically in order to smooth out writes to disk. This
471 : // option does not provide any persistency guarantee, but is used to avoid
472 : // latency spikes if the OS automatically decides to write out a large chunk
473 : // of dirty filesystem buffers. This option only controls SSTable syncs; WAL
474 : // syncs are controlled by WALBytesPerSync.
475 : //
476 : // The default value is 512KB.
477 : BytesPerSync int
478 :
479 : // Cache is used to cache uncompressed blocks from sstables.
480 : //
481 : // The default cache size is 8 MB.
482 : Cache *cache.Cache
483 :
484 : // LoadBlockSema, if set, is used to limit the number of blocks that can be
485 : // loaded (i.e. read from the filesystem) in parallel. Each load acquires one
486 : // unit from the semaphore for the duration of the read.
487 : LoadBlockSema *fifo.Semaphore
488 :
489 : // Cleaner cleans obsolete files.
490 : //
491 : // The default cleaner uses the DeleteCleaner.
492 : Cleaner Cleaner
493 :
494 : // Local contains option that pertain to files stored on the local filesystem.
495 : Local struct {
496 : // ReadaheadConfig is used to retrieve the current readahead mode; it is
497 : // consulted whenever a read handle is initialized.
498 : ReadaheadConfig *ReadaheadConfig
499 :
500 : // TODO(radu): move BytesPerSync, LoadBlockSema, Cleaner here.
501 : }
502 :
503 : // Comparer defines a total ordering over the space of []byte keys: a 'less
504 : // than' relationship. The same comparison algorithm must be used for reads
505 : // and writes over the lifetime of the DB.
506 : //
507 : // The default value uses the same ordering as bytes.Compare.
508 : Comparer *Comparer
509 :
510 : // DebugCheck is invoked, if non-nil, whenever a new version is being
511 : // installed. Typically, this is set to pebble.DebugCheckLevels in tests
512 : // or tools only, to check invariants over all the data in the database.
513 : DebugCheck func(*DB) error
514 :
515 : // Disable the write-ahead log (WAL). Disabling the write-ahead log prohibits
516 : // crash recovery, but can improve performance if crash recovery is not
517 : // needed (e.g. when only temporary state is being stored in the database).
518 : //
519 : // TODO(peter): untested
520 : DisableWAL bool
521 :
522 : // ErrorIfExists causes an error on Open if the database already exists.
523 : // The error can be checked with errors.Is(err, ErrDBAlreadyExists).
524 : //
525 : // The default value is false.
526 : ErrorIfExists bool
527 :
528 : // ErrorIfNotExists causes an error on Open if the database does not already
529 : // exist. The error can be checked with errors.Is(err, ErrDBDoesNotExist).
530 : //
531 : // The default value is false which will cause a database to be created if it
532 : // does not already exist.
533 : ErrorIfNotExists bool
534 :
535 : // ErrorIfNotPristine causes an error on Open if the database already exists
536 : // and any operations have been performed on the database. The error can be
537 : // checked with errors.Is(err, ErrDBNotPristine).
538 : //
539 : // Note that a database that contained keys that were all subsequently deleted
540 : // may or may not trigger the error. Currently, we check if there are any live
541 : // SSTs or log records to replay.
542 : ErrorIfNotPristine bool
543 :
544 : // EventListener provides hooks to listening to significant DB events such as
545 : // flushes, compactions, and table deletion.
546 : EventListener *EventListener
547 :
548 : // Experimental contains experimental options which are off by default.
549 : // These options are temporary and will eventually either be deleted, moved
550 : // out of the experimental group, or made the non-adjustable default. These
551 : // options may change at any time, so do not rely on them.
552 : Experimental struct {
553 : // The threshold of L0 read-amplification at which compaction concurrency
554 : // is enabled (if CompactionDebtConcurrency was not already exceeded).
555 : // Every multiple of this value enables another concurrent
556 : // compaction up to MaxConcurrentCompactions.
557 : L0CompactionConcurrency int
558 :
559 : // CompactionDebtConcurrency controls the threshold of compaction debt
560 : // at which additional compaction concurrency slots are added. For every
561 : // multiple of this value in compaction debt bytes, an additional
562 : // concurrent compaction is added. This works "on top" of
563 : // L0CompactionConcurrency, so the higher of the count of compaction
564 : // concurrency slots as determined by the two options is chosen.
565 : CompactionDebtConcurrency uint64
566 :
567 : // IngestSplit, if it returns true, allows for ingest-time splitting of
568 : // existing sstables into two virtual sstables to allow ingestion sstables to
569 : // slot into a lower level than they otherwise would have.
570 : IngestSplit func() bool
571 :
572 : // ReadCompactionRate controls the frequency of read triggered
573 : // compactions by adjusting `AllowedSeeks` in manifest.FileMetadata:
574 : //
575 : // AllowedSeeks = FileSize / ReadCompactionRate
576 : //
577 : // From LevelDB:
578 : // ```
579 : // We arrange to automatically compact this file after
580 : // a certain number of seeks. Let's assume:
581 : // (1) One seek costs 10ms
582 : // (2) Writing or reading 1MB costs 10ms (100MB/s)
583 : // (3) A compaction of 1MB does 25MB of IO:
584 : // 1MB read from this level
585 : // 10-12MB read from next level (boundaries may be misaligned)
586 : // 10-12MB written to next level
587 : // This implies that 25 seeks cost the same as the compaction
588 : // of 1MB of data. I.e., one seek costs approximately the
589 : // same as the compaction of 40KB of data. We are a little
590 : // conservative and allow approximately one seek for every 16KB
591 : // of data before triggering a compaction.
592 : // ```
593 : ReadCompactionRate int64
594 :
595 : // ReadSamplingMultiplier is a multiplier for the readSamplingPeriod in
596 : // iterator.maybeSampleRead() to control the frequency of read sampling
597 : // to trigger a read triggered compaction. A value of -1 prevents sampling
598 : // and disables read triggered compactions. The default is 1 << 4. which
599 : // gets multiplied with a constant of 1 << 16 to yield 1 << 20 (1MB).
600 : ReadSamplingMultiplier int64
601 :
602 : // NumDeletionsThreshold defines the minimum number of point tombstones
603 : // that must be present in a single data block for that block to be
604 : // considered tombstone-dense for the purposes of triggering a
605 : // tombstone density compaction. Data blocks may also be considered
606 : // tombstone-dense if they meet the criteria defined by
607 : // DeletionSizeRatioThreshold below. Tombstone-dense blocks are identified
608 : // when sstables are written, and so this is effectively an option for
609 : // sstable writers. The default value is 100.
610 : NumDeletionsThreshold int
611 :
612 : // DeletionSizeRatioThreshold defines the minimum ratio of the size of
613 : // point tombstones to the size of a data block that must be reached
614 : // for that block to be considered tombstone-dense for the purposes of
615 : // triggering a tombstone density compaction. Data blocks may also be
616 : // considered tombstone-dense if they meet the criteria defined by
617 : // NumDeletionsThreshold above. Tombstone-dense blocks are identified
618 : // when sstables are written, and so this is effectively an option for
619 : // sstable writers. The default value is 0.5.
620 : DeletionSizeRatioThreshold float32
621 :
622 : // TombstoneDenseCompactionThreshold is the minimum percent of data
623 : // blocks in a table that must be tombstone-dense for that table to be
624 : // eligible for a tombstone density compaction. It should be defined as a
625 : // ratio out of 1. The default value is 0.05.
626 : //
627 : // If multiple tables are eligible for a tombstone density compaction, then
628 : // tables with a higher percent of tombstone-dense blocks are still
629 : // prioritized for compaction.
630 : //
631 : // A zero or negative value disables tombstone density compactions.
632 : TombstoneDenseCompactionThreshold float64
633 :
634 : // TableCacheShards is the number of shards per table cache.
635 : // Reducing the value can reduce the number of idle goroutines per DB
636 : // instance which can be useful in scenarios with a lot of DB instances
637 : // and a large number of CPUs, but doing so can lead to higher contention
638 : // in the table cache and reduced performance.
639 : //
640 : // The default value is the number of logical CPUs, which can be
641 : // limited by runtime.GOMAXPROCS.
642 : TableCacheShards int
643 :
644 : // KeyValidationFunc is a function to validate a user key in an SSTable.
645 : //
646 : // Currently, this function is used to validate the smallest and largest
647 : // keys in an SSTable undergoing compaction. In this case, returning an
648 : // error from the validation function will result in a panic at runtime,
649 : // given that there is rarely any way of recovering from malformed keys
650 : // present in compacted files. By default, validation is not performed.
651 : //
652 : // Additional use-cases may be added in the future.
653 : //
654 : // NOTE: callers should take care to not mutate the key being validated.
655 : KeyValidationFunc func(userKey []byte) error
656 :
657 : // ValidateOnIngest schedules validation of sstables after they have
658 : // been ingested.
659 : //
660 : // By default, this value is false.
661 : ValidateOnIngest bool
662 :
663 : // LevelMultiplier configures the size multiplier used to determine the
664 : // desired size of each level of the LSM. Defaults to 10.
665 : LevelMultiplier int
666 :
667 : // MultiLevelCompactionHeuristic determines whether to add an additional
668 : // level to a conventional two level compaction. If nil, a multilevel
669 : // compaction will never get triggered.
670 : MultiLevelCompactionHeuristic MultiLevelHeuristic
671 :
672 : // MaxWriterConcurrency is used to indicate the maximum number of
673 : // compression workers the compression queue is allowed to use. If
674 : // MaxWriterConcurrency > 0, then the Writer will use parallelism, to
675 : // compress and write blocks to disk. Otherwise, the writer will
676 : // compress and write blocks to disk synchronously.
677 : MaxWriterConcurrency int
678 :
679 : // ForceWriterParallelism is used to force parallelism in the sstable
680 : // Writer for the metamorphic tests. Even with the MaxWriterConcurrency
681 : // option set, we only enable parallelism in the sstable Writer if there
682 : // is enough CPU available, and this option bypasses that.
683 : ForceWriterParallelism bool
684 :
685 : // CPUWorkPermissionGranter should be set if Pebble should be given the
686 : // ability to optionally schedule additional CPU. See the documentation
687 : // for CPUWorkPermissionGranter for more details.
688 : CPUWorkPermissionGranter CPUWorkPermissionGranter
689 :
690 : // EnableValueBlocks is used to decide whether to enable writing
691 : // TableFormatPebblev3 sstables. This setting is only respected by a
692 : // specific subset of format major versions: FormatSSTableValueBlocks,
693 : // FormatFlushableIngest and FormatPrePebblev1MarkedCompacted. In lower
694 : // format major versions, value blocks are never enabled. In higher
695 : // format major versions, value blocks are always enabled.
696 : EnableValueBlocks func() bool
697 :
698 : // ShortAttributeExtractor is used iff EnableValueBlocks() returns true
699 : // (else ignored). If non-nil, a ShortAttribute can be extracted from the
700 : // value and stored with the key, when the value is stored elsewhere.
701 : ShortAttributeExtractor ShortAttributeExtractor
702 :
703 : // RequiredInPlaceValueBound specifies an optional span of user key
704 : // prefixes that are not-MVCC, but have a suffix. For these the values
705 : // must be stored with the key, since the concept of "older versions" is
706 : // not defined. It is also useful for statically known exclusions to value
707 : // separation. In CockroachDB, this will be used for the lock table key
708 : // space that has non-empty suffixes, but those locks don't represent
709 : // actual MVCC versions (the suffix ordering is arbitrary). We will also
710 : // need to add support for dynamically configured exclusions (we want the
711 : // default to be to allow Pebble to decide whether to separate the value
712 : // or not, hence this is structured as exclusions), for example, for users
713 : // of CockroachDB to dynamically exclude certain tables.
714 : //
715 : // Any change in exclusion behavior takes effect only on future written
716 : // sstables, and does not start rewriting existing sstables.
717 : //
718 : // Even ignoring changes in this setting, exclusions are interpreted as a
719 : // guidance by Pebble, and not necessarily honored. Specifically, user
720 : // keys with multiple Pebble-versions *may* have the older versions stored
721 : // in value blocks.
722 : RequiredInPlaceValueBound UserKeyPrefixBound
723 :
724 : // DisableIngestAsFlushable disables lazy ingestion of sstables through
725 : // a WAL write and memtable rotation. Only effectual if the format
726 : // major version is at least `FormatFlushableIngest`.
727 : DisableIngestAsFlushable func() bool
728 :
729 : // RemoteStorage enables use of remote storage (e.g. S3) for storing
730 : // sstables. Setting this option enables use of CreateOnShared option and
731 : // allows ingestion of external files.
732 : RemoteStorage remote.StorageFactory
733 :
734 : // If CreateOnShared is non-zero, new sstables are created on remote storage
735 : // (using CreateOnSharedLocator and with the appropriate
736 : // CreateOnSharedStrategy). These sstables can be shared between different
737 : // Pebble instances; the lifecycle of such objects is managed by the
738 : // remote.Storage constructed by options.RemoteStorage.
739 : //
740 : // Can only be used when RemoteStorage is set (and recognizes
741 : // CreateOnSharedLocator).
742 : CreateOnShared remote.CreateOnSharedStrategy
743 : CreateOnSharedLocator remote.Locator
744 :
745 : // CacheSizeBytesBytes is the size of the on-disk block cache for objects
746 : // on shared storage in bytes. If it is 0, no cache is used.
747 : SecondaryCacheSizeBytes int64
748 :
749 : // NB: DO NOT crash on SingleDeleteInvariantViolationCallback or
750 : // IneffectualSingleDeleteCallback, since these can be false positives
751 : // even if SingleDel has been used correctly.
752 : //
753 : // Pebble's delete-only compactions can cause a recent RANGEDEL to peek
754 : // below an older SINGLEDEL and delete an arbitrary subset of data below
755 : // that SINGLEDEL. When that SINGLEDEL gets compacted (without the
756 : // RANGEDEL), any of these callbacks can happen, without it being a real
757 : // correctness problem.
758 : //
759 : // Example 1:
760 : // RANGEDEL [a, c)#10 in L0
761 : // SINGLEDEL b#5 in L1
762 : // SET b#3 in L6
763 : //
764 : // If the L6 file containing the SET is narrow and the L1 file containing
765 : // the SINGLEDEL is wide, a delete-only compaction can remove the file in
766 : // L2 before the SINGLEDEL is compacted down. Then when the SINGLEDEL is
767 : // compacted down, it will not find any SET to delete, resulting in the
768 : // ineffectual callback.
769 : //
770 : // Example 2:
771 : // RANGEDEL [a, z)#60 in L0
772 : // SINGLEDEL g#50 in L1
773 : // SET g#40 in L2
774 : // RANGEDEL [g,h)#30 in L3
775 : // SET g#20 in L6
776 : //
777 : // In this example, the two SETs represent the same user write, and the
778 : // RANGEDELs are caused by the CockroachDB range being dropped. That is,
779 : // the user wrote to g once, range was dropped, then added back, which
780 : // caused the SET again, then at some point g was validly deleted using a
781 : // SINGLEDEL, and then the range was dropped again. The older RANGEDEL can
782 : // get fragmented due to compactions it has been part of. Say this L3 file
783 : // containing the RANGEDEL is very narrow, while the L1, L2, L6 files are
784 : // wider than the RANGEDEL in L0. Then the RANGEDEL in L3 can be dropped
785 : // using a delete-only compaction, resulting in an LSM with state:
786 : //
787 : // RANGEDEL [a, z)#60 in L0
788 : // SINGLEDEL g#50 in L1
789 : // SET g#40 in L2
790 : // SET g#20 in L6
791 : //
792 : // A multi-level compaction involving L1, L2, L6 will cause the invariant
793 : // violation callback. This example doesn't need multi-level compactions:
794 : // say there was a Pebble snapshot at g#21 preventing g#20 from being
795 : // dropped when it meets g#40 in a compaction. That snapshot will not save
796 : // RANGEDEL [g,h)#30, so we can have:
797 : //
798 : // SINGLEDEL g#50 in L1
799 : // SET g#40, SET g#20 in L6
800 : //
801 : // And say the snapshot is removed and then the L1 and L6 compaction
802 : // happens, resulting in the invariant violation callback.
803 : //
804 : // TODO(sumeer): rename SingleDeleteInvariantViolationCallback to remove
805 : // the word "invariant".
806 :
807 : // IneffectualPointDeleteCallback is called in compactions/flushes if any
808 : // single delete is being elided without deleting a point set/merge.
809 : IneffectualSingleDeleteCallback func(userKey []byte)
810 :
811 : // SingleDeleteInvariantViolationCallback is called in compactions/flushes if any
812 : // single delete has consumed a Set/Merge, and there is another immediately older
813 : // Set/SetWithDelete/Merge. The user of Pebble has violated the invariant under
814 : // which SingleDelete can be used correctly.
815 : //
816 : // Consider the sequence SingleDelete#3, Set#2, Set#1. There are three
817 : // ways some of these keys can first meet in a compaction.
818 : //
819 : // - All 3 keys in the same compaction: this callback will detect the
820 : // violation.
821 : //
822 : // - SingleDelete#3, Set#2 meet in a compaction first: Both keys will
823 : // disappear. The violation will not be detected, and the DB will have
824 : // Set#1 which is likely incorrect (from the user's perspective).
825 : //
826 : // - Set#2, Set#1 meet in a compaction first: The output will be Set#2,
827 : // which will later be consumed by SingleDelete#3. The violation will
828 : // not be detected and the DB will be correct.
829 : SingleDeleteInvariantViolationCallback func(userKey []byte)
830 : }
831 :
832 : // Filters is a map from filter policy name to filter policy. It is used for
833 : // debugging tools which may be used on multiple databases configured with
834 : // different filter policies. It is not necessary to populate this filters
835 : // map during normal usage of a DB (it will be done automatically by
836 : // EnsureDefaults).
837 : Filters map[string]FilterPolicy
838 :
839 : // FlushDelayDeleteRange configures how long the database should wait before
840 : // forcing a flush of a memtable that contains a range deletion. Disk space
841 : // cannot be reclaimed until the range deletion is flushed. No automatic
842 : // flush occurs if zero.
843 : FlushDelayDeleteRange time.Duration
844 :
845 : // FlushDelayRangeKey configures how long the database should wait before
846 : // forcing a flush of a memtable that contains a range key. Range keys in
847 : // the memtable prevent lazy combined iteration, so it's desirable to flush
848 : // range keys promptly. No automatic flush occurs if zero.
849 : FlushDelayRangeKey time.Duration
850 :
851 : // FlushSplitBytes denotes the target number of bytes per sublevel in
852 : // each flush split interval (i.e. range between two flush split keys)
853 : // in L0 sstables. When set to zero, only a single sstable is generated
854 : // by each flush. When set to a non-zero value, flushes are split at
855 : // points to meet L0's TargetFileSize, any grandparent-related overlap
856 : // options, and at boundary keys of L0 flush split intervals (which are
857 : // targeted to contain around FlushSplitBytes bytes in each sublevel
858 : // between pairs of boundary keys). Splitting sstables during flush
859 : // allows increased compaction flexibility and concurrency when those
860 : // tables are compacted to lower levels.
861 : FlushSplitBytes int64
862 :
863 : // FormatMajorVersion sets the format of on-disk files. It is
864 : // recommended to set the format major version to an explicit
865 : // version, as the default may change over time.
866 : //
867 : // At Open if the existing database is formatted using a later
868 : // format major version that is known to this version of Pebble,
869 : // Pebble will continue to use the later format major version. If
870 : // the existing database's version is unknown, the caller may use
871 : // FormatMostCompatible and will be able to open the database
872 : // regardless of its actual version.
873 : //
874 : // If the existing database is formatted using a format major
875 : // version earlier than the one specified, Open will automatically
876 : // ratchet the database to the specified format major version.
877 : FormatMajorVersion FormatMajorVersion
878 :
879 : // FS provides the interface for persistent file storage.
880 : //
881 : // The default value uses the underlying operating system's file system.
882 : FS vfs.FS
883 :
884 : // Lock, if set, must be a database lock acquired through LockDirectory for
885 : // the same directory passed to Open. If provided, Open will skip locking
886 : // the directory. Closing the database will not release the lock, and it's
887 : // the responsibility of the caller to release the lock after closing the
888 : // database.
889 : //
890 : // Open will enforce that the Lock passed locks the same directory passed to
891 : // Open. Concurrent calls to Open using the same Lock are detected and
892 : // prohibited.
893 : Lock *Lock
894 :
895 : // The count of L0 files necessary to trigger an L0 compaction.
896 : L0CompactionFileThreshold int
897 :
898 : // The amount of L0 read-amplification necessary to trigger an L0 compaction.
899 : L0CompactionThreshold int
900 :
901 : // Hard limit on L0 read-amplification, computed as the number of L0
902 : // sublevels. Writes are stopped when this threshold is reached.
903 : L0StopWritesThreshold int
904 :
905 : // The maximum number of bytes for LBase. The base level is the level which
906 : // L0 is compacted into. The base level is determined dynamically based on
907 : // the existing data in the LSM. The maximum number of bytes for other levels
908 : // is computed dynamically based on the base level's maximum size. When the
909 : // maximum number of bytes for a level is exceeded, compaction is requested.
910 : LBaseMaxBytes int64
911 :
912 : // Per-level options. Options for at least one level must be specified. The
913 : // options for the last level are used for all subsequent levels.
914 : Levels []LevelOptions
915 :
916 : // LoggerAndTracer will be used, if non-nil, else Logger will be used and
917 : // tracing will be a noop.
918 :
919 : // Logger used to write log messages.
920 : //
921 : // The default logger uses the Go standard library log package.
922 : Logger Logger
923 : // LoggerAndTracer is used for writing log messages and traces.
924 : LoggerAndTracer LoggerAndTracer
925 :
926 : // MaxManifestFileSize is the maximum size the MANIFEST file is allowed to
927 : // become. When the MANIFEST exceeds this size it is rolled over and a new
928 : // MANIFEST is created.
929 : MaxManifestFileSize int64
930 :
931 : // MaxOpenFiles is a soft limit on the number of open files that can be
932 : // used by the DB.
933 : //
934 : // The default value is 1000.
935 : MaxOpenFiles int
936 :
937 : // The size of a MemTable in steady state. The actual MemTable size starts at
938 : // min(256KB, MemTableSize) and doubles for each subsequent MemTable up to
939 : // MemTableSize. This reduces the memory pressure caused by MemTables for
940 : // short lived (test) DB instances. Note that more than one MemTable can be
941 : // in existence since flushing a MemTable involves creating a new one and
942 : // writing the contents of the old one in the
943 : // background. MemTableStopWritesThreshold places a hard limit on the size of
944 : // the queued MemTables.
945 : //
946 : // The default value is 4MB.
947 : MemTableSize uint64
948 :
949 : // Hard limit on the number of queued of MemTables. Writes are stopped when
950 : // the sum of the queued memtable sizes exceeds:
951 : // MemTableStopWritesThreshold * MemTableSize.
952 : //
953 : // This value should be at least 2 or writes will stop whenever a MemTable is
954 : // being flushed.
955 : //
956 : // The default value is 2.
957 : MemTableStopWritesThreshold int
958 :
959 : // Merger defines the associative merge operation to use for merging values
960 : // written with {Batch,DB}.Merge.
961 : //
962 : // The default merger concatenates values.
963 : Merger *Merger
964 :
965 : // MaxConcurrentCompactions specifies the maximum number of concurrent
966 : // compactions (not including download compactions).
967 : //
968 : // Concurrent compactions are performed:
969 : // - when L0 read-amplification passes the L0CompactionConcurrency threshold;
970 : // - for automatic background compactions;
971 : // - when a manual compaction for a level is split and parallelized.
972 : //
973 : // MaxConcurrentCompactions() must be greater than 0.
974 : //
975 : // The default value is 1.
976 : MaxConcurrentCompactions func() int
977 :
978 : // MaxConcurrentDownloads specifies the maximum number of download
979 : // compactions. These are compactions that copy an external file to the local
980 : // store.
981 : //
982 : // This limit is independent of MaxConcurrentCompactions; at any point in
983 : // time, we may be running MaxConcurrentCompactions non-download compactions
984 : // and MaxConcurrentDownloads download compactions.
985 : //
986 : // MaxConcurrentDownloads() must be greater than 0.
987 : //
988 : // The default value is 1.
989 : MaxConcurrentDownloads func() int
990 :
991 : // DisableAutomaticCompactions dictates whether automatic compactions are
992 : // scheduled or not. The default is false (enabled). This option is only used
993 : // externally when running a manual compaction, and internally for tests.
994 : DisableAutomaticCompactions bool
995 :
996 : // DisableConsistencyCheck disables the consistency check that is performed on
997 : // open. Should only be used when a database cannot be opened normally (e.g.
998 : // some of the tables don't exist / aren't accessible).
999 : DisableConsistencyCheck bool
1000 :
1001 : // DisableTableStats dictates whether tables should be loaded asynchronously
1002 : // to compute statistics that inform compaction heuristics. The collection
1003 : // of table stats improves compaction of tombstones, reclaiming disk space
1004 : // more quickly and in some cases reducing write amplification in the
1005 : // presence of tombstones. Disabling table stats may be useful in tests
1006 : // that require determinism as the asynchronicity of table stats collection
1007 : // introduces significant nondeterminism.
1008 : DisableTableStats bool
1009 :
1010 : // NoSyncOnClose decides whether the Pebble instance will enforce a
1011 : // close-time synchronization (e.g., fdatasync() or sync_file_range())
1012 : // on files it writes to. Setting this to true removes the guarantee for a
1013 : // sync on close. Some implementations can still issue a non-blocking sync.
1014 : NoSyncOnClose bool
1015 :
1016 : // NumPrevManifest is the number of non-current or older manifests which
1017 : // we want to keep around for debugging purposes. By default, we're going
1018 : // to keep one older manifest.
1019 : NumPrevManifest int
1020 :
1021 : // ReadOnly indicates that the DB should be opened in read-only mode. Writes
1022 : // to the DB will return an error, background compactions are disabled, and
1023 : // the flush that normally occurs after replaying the WAL at startup is
1024 : // disabled.
1025 : ReadOnly bool
1026 :
1027 : // TableCache is an initialized TableCache which should be set as an
1028 : // option if the DB needs to be initialized with a pre-existing table cache.
1029 : // If TableCache is nil, then a table cache which is unique to the DB instance
1030 : // is created. TableCache can be shared between db instances by setting it here.
1031 : // The TableCache set here must use the same underlying cache as Options.Cache
1032 : // and pebble will panic otherwise.
1033 : TableCache *TableCache
1034 :
1035 : // BlockPropertyCollectors is a list of BlockPropertyCollector creation
1036 : // functions. A new BlockPropertyCollector is created for each sstable
1037 : // built and lives for the lifetime of writing that table.
1038 : BlockPropertyCollectors []func() BlockPropertyCollector
1039 :
1040 : // WALBytesPerSync sets the number of bytes to write to a WAL before calling
1041 : // Sync on it in the background. Just like with BytesPerSync above, this
1042 : // helps smooth out disk write latencies, and avoids cases where the OS
1043 : // writes a lot of buffered data to disk at once. However, this is less
1044 : // necessary with WALs, as many write operations already pass in
1045 : // Sync = true.
1046 : //
1047 : // The default value is 0, i.e. no background syncing. This matches the
1048 : // default behaviour in RocksDB.
1049 : WALBytesPerSync int
1050 :
1051 : // WALDir specifies the directory to store write-ahead logs (WALs) in. If
1052 : // empty (the default), WALs will be stored in the same directory as sstables
1053 : // (i.e. the directory passed to pebble.Open).
1054 : WALDir string
1055 :
1056 : // WALFailover may be set to configure Pebble to monitor writes to its
1057 : // write-ahead log and failover to writing write-ahead log entries to a
1058 : // secondary location (eg, a separate physical disk). WALFailover may be
1059 : // used to improve write availability in the presence of transient disk
1060 : // unavailability.
1061 : WALFailover *WALFailoverOptions
1062 :
1063 : // WALRecoveryDirs is a list of additional directories that should be
1064 : // scanned for the existence of additional write-ahead logs. WALRecoveryDirs
1065 : // is expected to be used when starting Pebble with a new WALDir or a new
1066 : // WALFailover configuration. The directories associated with the previous
1067 : // configuration may still contain WALs that are required for recovery of
1068 : // the current database state.
1069 : //
1070 : // If a previous WAL configuration may have stored WALs elsewhere but there
1071 : // is not a corresponding entry in WALRecoveryDirs, Open will error.
1072 : WALRecoveryDirs []wal.Dir
1073 :
1074 : // WALMinSyncInterval is the minimum duration between syncs of the WAL. If
1075 : // WAL syncs are requested faster than this interval, they will be
1076 : // artificially delayed. Introducing a small artificial delay (500us) between
1077 : // WAL syncs can allow more operations to arrive and reduce IO operations
1078 : // while having a minimal impact on throughput. This option is supplied as a
1079 : // closure in order to allow the value to be changed dynamically. The default
1080 : // value is 0.
1081 : //
1082 : // TODO(peter): rather than a closure, should there be another mechanism for
1083 : // changing options dynamically?
1084 : WALMinSyncInterval func() time.Duration
1085 :
1086 : // TargetByteDeletionRate is the rate (in bytes per second) at which sstable file
1087 : // deletions are limited to (under normal circumstances).
1088 : //
1089 : // Deletion pacing is used to slow down deletions when compactions finish up
1090 : // or readers close and newly-obsolete files need cleaning up. Deleting lots
1091 : // of files at once can cause disk latency to go up on some SSDs, which this
1092 : // functionality guards against.
1093 : //
1094 : // This value is only a best-effort target; the effective rate can be
1095 : // higher if deletions are falling behind or disk space is running low.
1096 : //
1097 : // Setting this to 0 disables deletion pacing, which is also the default.
1098 : TargetByteDeletionRate int
1099 :
1100 : // EnableSQLRowSpillMetrics specifies whether the Pebble instance will only be used
1101 : // to temporarily persist data spilled to disk for row-oriented SQL query execution.
1102 : EnableSQLRowSpillMetrics bool
1103 :
1104 : // AllocatorSizeClasses provides a sorted list containing the supported size
1105 : // classes of the underlying memory allocator. This provides hints to the
1106 : // sstable block writer's flushing policy to select block sizes that
1107 : // preemptively reduce internal fragmentation when loaded into the block cache.
1108 : AllocatorSizeClasses []int
1109 :
1110 : // private options are only used by internal tests or are used internally
1111 : // for facilitating upgrade paths of unconfigurable functionality.
1112 : private struct {
1113 : // disableDeleteOnlyCompactions prevents the scheduling of delete-only
1114 : // compactions that drop sstables wholy covered by range tombstones or
1115 : // range key tombstones.
1116 : disableDeleteOnlyCompactions bool
1117 :
1118 : // disableElisionOnlyCompactions prevents the scheduling of elision-only
1119 : // compactions that rewrite sstables in place in order to elide obsolete
1120 : // keys.
1121 : disableElisionOnlyCompactions bool
1122 :
1123 : // disableLazyCombinedIteration is a private option used by the
1124 : // metamorphic tests to test equivalence between lazy-combined iteration
1125 : // and constructing the range-key iterator upfront. It's a private
1126 : // option to avoid littering the public interface with options that we
1127 : // do not want to allow users to actually configure.
1128 : disableLazyCombinedIteration bool
1129 :
1130 : // testingAlwaysWaitForCleanup is set by some tests to force waiting for
1131 : // obsolete file deletion (to make events deterministic).
1132 : testingAlwaysWaitForCleanup bool
1133 :
1134 : // fsCloser holds a closer that should be invoked after a DB using these
1135 : // Options is closed. This is used to automatically stop the
1136 : // long-running goroutine associated with the disk-health-checking FS.
1137 : // See the initialization of FS in EnsureDefaults. Note that care has
1138 : // been taken to ensure that it is still safe to continue using the FS
1139 : // after this closer has been invoked. However, if write operations
1140 : // against the FS are made after the DB is closed, the FS may leak a
1141 : // goroutine indefinitely.
1142 : fsCloser io.Closer
1143 : }
1144 : }
1145 :
1146 : // WALFailoverOptions configures the WAL failover mechanics to use during
1147 : // transient write unavailability on the primary WAL volume.
1148 : type WALFailoverOptions struct {
1149 : // Secondary indicates the secondary directory and VFS to use in the event a
1150 : // write to the primary WAL stalls.
1151 : Secondary wal.Dir
1152 : // FailoverOptions provides configuration of the thresholds and intervals
1153 : // involved in WAL failover. If any of its fields are left unspecified,
1154 : // reasonable defaults will be used.
1155 : wal.FailoverOptions
1156 : }
1157 :
1158 : // ReadaheadConfig controls the use of read-ahead.
1159 : type ReadaheadConfig = objstorageprovider.ReadaheadConfig
1160 :
1161 : // DebugCheckLevels calls CheckLevels on the provided database.
1162 : // It may be set in the DebugCheck field of Options to check
1163 : // level invariants whenever a new version is installed.
1164 1 : func DebugCheckLevels(db *DB) error {
1165 1 : return db.CheckLevels(nil)
1166 1 : }
1167 :
1168 : // EnsureDefaults ensures that the default values for all options are set if a
1169 : // valid value was not already specified. Returns the new options.
1170 1 : func (o *Options) EnsureDefaults() *Options {
1171 1 : if o == nil {
1172 1 : o = &Options{}
1173 1 : }
1174 1 : o.Comparer = o.Comparer.EnsureDefaults()
1175 1 :
1176 1 : if o.BytesPerSync <= 0 {
1177 1 : o.BytesPerSync = 512 << 10 // 512 KB
1178 1 : }
1179 1 : if o.Cleaner == nil {
1180 1 : o.Cleaner = DeleteCleaner{}
1181 1 : }
1182 :
1183 1 : if o.Experimental.DisableIngestAsFlushable == nil {
1184 1 : o.Experimental.DisableIngestAsFlushable = func() bool { return false }
1185 : }
1186 1 : if o.Experimental.L0CompactionConcurrency <= 0 {
1187 1 : o.Experimental.L0CompactionConcurrency = 10
1188 1 : }
1189 1 : if o.Experimental.CompactionDebtConcurrency <= 0 {
1190 1 : o.Experimental.CompactionDebtConcurrency = 1 << 30 // 1 GB
1191 1 : }
1192 1 : if o.Experimental.KeyValidationFunc == nil {
1193 1 : o.Experimental.KeyValidationFunc = func([]byte) error { return nil }
1194 : }
1195 1 : if o.L0CompactionThreshold <= 0 {
1196 1 : o.L0CompactionThreshold = 4
1197 1 : }
1198 1 : if o.L0CompactionFileThreshold <= 0 {
1199 1 : // Some justification for the default of 500:
1200 1 : // Why not smaller?:
1201 1 : // - The default target file size for L0 is 2MB, so 500 files is <= 1GB
1202 1 : // of data. At observed compaction speeds of > 20MB/s, L0 can be
1203 1 : // cleared of all files in < 1min, so this backlog is not huge.
1204 1 : // - 500 files is low overhead for instantiating L0 sublevels from
1205 1 : // scratch.
1206 1 : // - Lower values were observed to cause excessive and inefficient
1207 1 : // compactions out of L0 in a TPCC import benchmark.
1208 1 : // Why not larger?:
1209 1 : // - More than 1min to compact everything out of L0.
1210 1 : // - CockroachDB's admission control system uses a threshold of 1000
1211 1 : // files to start throttling writes to Pebble. Using 500 here gives
1212 1 : // us headroom between when Pebble should start compacting L0 and
1213 1 : // when the admission control threshold is reached.
1214 1 : //
1215 1 : // We can revisit this default in the future based on better
1216 1 : // experimental understanding.
1217 1 : //
1218 1 : // TODO(jackson): Experiment with slightly lower thresholds [or higher
1219 1 : // admission control thresholds] to see whether a higher L0 score at the
1220 1 : // threshold (currently 2.0) is necessary for some workloads to avoid
1221 1 : // starving L0 in favor of lower-level compactions.
1222 1 : o.L0CompactionFileThreshold = 500
1223 1 : }
1224 1 : if o.L0StopWritesThreshold <= 0 {
1225 1 : o.L0StopWritesThreshold = 12
1226 1 : }
1227 1 : if o.LBaseMaxBytes <= 0 {
1228 1 : o.LBaseMaxBytes = 64 << 20 // 64 MB
1229 1 : }
1230 1 : if o.Levels == nil {
1231 1 : o.Levels = make([]LevelOptions, 1)
1232 1 : for i := range o.Levels {
1233 1 : if i > 0 {
1234 0 : l := &o.Levels[i]
1235 0 : if l.TargetFileSize <= 0 {
1236 0 : l.TargetFileSize = o.Levels[i-1].TargetFileSize * 2
1237 0 : }
1238 : }
1239 1 : o.Levels[i].EnsureDefaults()
1240 : }
1241 1 : } else {
1242 1 : for i := range o.Levels {
1243 1 : o.Levels[i].EnsureDefaults()
1244 1 : }
1245 : }
1246 1 : if o.Logger == nil {
1247 1 : o.Logger = DefaultLogger
1248 1 : }
1249 1 : if o.EventListener == nil {
1250 1 : o.EventListener = &EventListener{}
1251 1 : }
1252 1 : o.EventListener.EnsureDefaults(o.Logger)
1253 1 : if o.MaxManifestFileSize == 0 {
1254 1 : o.MaxManifestFileSize = 128 << 20 // 128 MB
1255 1 : }
1256 1 : if o.MaxOpenFiles == 0 {
1257 1 : o.MaxOpenFiles = 1000
1258 1 : }
1259 1 : if o.MemTableSize <= 0 {
1260 1 : o.MemTableSize = 4 << 20 // 4 MB
1261 1 : }
1262 1 : if o.MemTableStopWritesThreshold <= 0 {
1263 1 : o.MemTableStopWritesThreshold = 2
1264 1 : }
1265 1 : if o.Merger == nil {
1266 1 : o.Merger = DefaultMerger
1267 1 : }
1268 1 : if o.MaxConcurrentCompactions == nil {
1269 1 : o.MaxConcurrentCompactions = func() int { return 1 }
1270 : }
1271 1 : if o.MaxConcurrentDownloads == nil {
1272 1 : o.MaxConcurrentDownloads = func() int { return 1 }
1273 : }
1274 1 : if o.NumPrevManifest <= 0 {
1275 1 : o.NumPrevManifest = 1
1276 1 : }
1277 :
1278 1 : if o.FormatMajorVersion == FormatDefault {
1279 1 : o.FormatMajorVersion = FormatMinSupported
1280 1 : if o.Experimental.CreateOnShared != remote.CreateOnSharedNone {
1281 1 : o.FormatMajorVersion = FormatMinForSharedObjects
1282 1 : }
1283 : }
1284 :
1285 1 : if o.FS == nil {
1286 1 : o.WithFSDefaults()
1287 1 : }
1288 1 : if o.FlushSplitBytes <= 0 {
1289 1 : o.FlushSplitBytes = 2 * o.Levels[0].TargetFileSize
1290 1 : }
1291 1 : if o.WALFailover != nil {
1292 1 : o.WALFailover.FailoverOptions.EnsureDefaults()
1293 1 : }
1294 1 : if o.Experimental.LevelMultiplier <= 0 {
1295 1 : o.Experimental.LevelMultiplier = defaultLevelMultiplier
1296 1 : }
1297 1 : if o.Experimental.ReadCompactionRate == 0 {
1298 1 : o.Experimental.ReadCompactionRate = 16000
1299 1 : }
1300 1 : if o.Experimental.ReadSamplingMultiplier == 0 {
1301 1 : o.Experimental.ReadSamplingMultiplier = 1 << 4
1302 1 : }
1303 1 : if o.Experimental.NumDeletionsThreshold == 0 {
1304 1 : o.Experimental.NumDeletionsThreshold = sstable.DefaultNumDeletionsThreshold
1305 1 : }
1306 1 : if o.Experimental.DeletionSizeRatioThreshold == 0 {
1307 1 : o.Experimental.DeletionSizeRatioThreshold = sstable.DefaultDeletionSizeRatioThreshold
1308 1 : }
1309 1 : if o.Experimental.TombstoneDenseCompactionThreshold == 0 {
1310 1 : o.Experimental.TombstoneDenseCompactionThreshold = 0.05
1311 1 : }
1312 1 : if o.Experimental.TableCacheShards <= 0 {
1313 1 : o.Experimental.TableCacheShards = runtime.GOMAXPROCS(0)
1314 1 : }
1315 1 : if o.Experimental.CPUWorkPermissionGranter == nil {
1316 1 : o.Experimental.CPUWorkPermissionGranter = defaultCPUWorkGranter{}
1317 1 : }
1318 1 : if o.Experimental.MultiLevelCompactionHeuristic == nil {
1319 1 : o.Experimental.MultiLevelCompactionHeuristic = WriteAmpHeuristic{}
1320 1 : }
1321 :
1322 1 : o.initMaps()
1323 1 : return o
1324 : }
1325 :
1326 : // WithFSDefaults configures the Options to wrap the configured filesystem with
1327 : // the default virtual file system middleware, like disk-health checking.
1328 1 : func (o *Options) WithFSDefaults() *Options {
1329 1 : if o.FS == nil {
1330 1 : o.FS = vfs.Default
1331 1 : }
1332 1 : o.FS, o.private.fsCloser = vfs.WithDiskHealthChecks(o.FS, 5*time.Second, nil,
1333 1 : func(info vfs.DiskSlowInfo) {
1334 0 : o.EventListener.DiskSlow(info)
1335 0 : })
1336 1 : return o
1337 : }
1338 :
1339 : // AddEventListener adds the provided event listener to the Options, in addition
1340 : // to any existing event listener.
1341 1 : func (o *Options) AddEventListener(l EventListener) {
1342 1 : if o.EventListener != nil {
1343 1 : l = TeeEventListener(l, *o.EventListener)
1344 1 : }
1345 1 : o.EventListener = &l
1346 : }
1347 :
1348 : // initMaps initializes the Comparers, Filters, and Mergers maps.
1349 1 : func (o *Options) initMaps() {
1350 1 : for i := range o.Levels {
1351 1 : l := &o.Levels[i]
1352 1 : if l.FilterPolicy != nil {
1353 1 : if o.Filters == nil {
1354 1 : o.Filters = make(map[string]FilterPolicy)
1355 1 : }
1356 1 : name := l.FilterPolicy.Name()
1357 1 : if _, ok := o.Filters[name]; !ok {
1358 1 : o.Filters[name] = l.FilterPolicy
1359 1 : }
1360 : }
1361 : }
1362 : }
1363 :
1364 : // Level returns the LevelOptions for the specified level.
1365 1 : func (o *Options) Level(level int) LevelOptions {
1366 1 : if level < len(o.Levels) {
1367 1 : return o.Levels[level]
1368 1 : }
1369 1 : n := len(o.Levels) - 1
1370 1 : l := o.Levels[n]
1371 1 : for i := n; i < level; i++ {
1372 1 : l.TargetFileSize *= 2
1373 1 : }
1374 1 : return l
1375 : }
1376 :
1377 : // Clone creates a shallow-copy of the supplied options.
1378 1 : func (o *Options) Clone() *Options {
1379 1 : n := &Options{}
1380 1 : if o != nil {
1381 1 : *n = *o
1382 1 : }
1383 1 : return n
1384 : }
1385 :
1386 1 : func filterPolicyName(p FilterPolicy) string {
1387 1 : if p == nil {
1388 1 : return "none"
1389 1 : }
1390 1 : return p.Name()
1391 : }
1392 :
1393 1 : func (o *Options) String() string {
1394 1 : var buf bytes.Buffer
1395 1 :
1396 1 : cacheSize := int64(cacheDefaultSize)
1397 1 : if o.Cache != nil {
1398 1 : cacheSize = o.Cache.MaxSize()
1399 1 : }
1400 :
1401 1 : fmt.Fprintf(&buf, "[Version]\n")
1402 1 : fmt.Fprintf(&buf, " pebble_version=0.1\n")
1403 1 : fmt.Fprintf(&buf, "\n")
1404 1 : fmt.Fprintf(&buf, "[Options]\n")
1405 1 : fmt.Fprintf(&buf, " bytes_per_sync=%d\n", o.BytesPerSync)
1406 1 : fmt.Fprintf(&buf, " cache_size=%d\n", cacheSize)
1407 1 : fmt.Fprintf(&buf, " cleaner=%s\n", o.Cleaner)
1408 1 : fmt.Fprintf(&buf, " compaction_debt_concurrency=%d\n", o.Experimental.CompactionDebtConcurrency)
1409 1 : fmt.Fprintf(&buf, " comparer=%s\n", o.Comparer.Name)
1410 1 : fmt.Fprintf(&buf, " disable_wal=%t\n", o.DisableWAL)
1411 1 : if o.Experimental.DisableIngestAsFlushable != nil && o.Experimental.DisableIngestAsFlushable() {
1412 1 : fmt.Fprintf(&buf, " disable_ingest_as_flushable=%t\n", true)
1413 1 : }
1414 1 : fmt.Fprintf(&buf, " flush_delay_delete_range=%s\n", o.FlushDelayDeleteRange)
1415 1 : fmt.Fprintf(&buf, " flush_delay_range_key=%s\n", o.FlushDelayRangeKey)
1416 1 : fmt.Fprintf(&buf, " flush_split_bytes=%d\n", o.FlushSplitBytes)
1417 1 : fmt.Fprintf(&buf, " format_major_version=%d\n", o.FormatMajorVersion)
1418 1 : fmt.Fprintf(&buf, " l0_compaction_concurrency=%d\n", o.Experimental.L0CompactionConcurrency)
1419 1 : fmt.Fprintf(&buf, " l0_compaction_file_threshold=%d\n", o.L0CompactionFileThreshold)
1420 1 : fmt.Fprintf(&buf, " l0_compaction_threshold=%d\n", o.L0CompactionThreshold)
1421 1 : fmt.Fprintf(&buf, " l0_stop_writes_threshold=%d\n", o.L0StopWritesThreshold)
1422 1 : fmt.Fprintf(&buf, " lbase_max_bytes=%d\n", o.LBaseMaxBytes)
1423 1 : if o.Experimental.LevelMultiplier != defaultLevelMultiplier {
1424 1 : fmt.Fprintf(&buf, " level_multiplier=%d\n", o.Experimental.LevelMultiplier)
1425 1 : }
1426 1 : fmt.Fprintf(&buf, " max_concurrent_compactions=%d\n", o.MaxConcurrentCompactions())
1427 1 : fmt.Fprintf(&buf, " max_concurrent_downloads=%d\n", o.MaxConcurrentDownloads())
1428 1 : fmt.Fprintf(&buf, " max_manifest_file_size=%d\n", o.MaxManifestFileSize)
1429 1 : fmt.Fprintf(&buf, " max_open_files=%d\n", o.MaxOpenFiles)
1430 1 : fmt.Fprintf(&buf, " mem_table_size=%d\n", o.MemTableSize)
1431 1 : fmt.Fprintf(&buf, " mem_table_stop_writes_threshold=%d\n", o.MemTableStopWritesThreshold)
1432 1 : fmt.Fprintf(&buf, " min_deletion_rate=%d\n", o.TargetByteDeletionRate)
1433 1 : fmt.Fprintf(&buf, " merger=%s\n", o.Merger.Name)
1434 1 : if o.Experimental.MultiLevelCompactionHeuristic != nil {
1435 1 : fmt.Fprintf(&buf, " multilevel_compaction_heuristic=%s\n", o.Experimental.MultiLevelCompactionHeuristic.String())
1436 1 : }
1437 1 : fmt.Fprintf(&buf, " read_compaction_rate=%d\n", o.Experimental.ReadCompactionRate)
1438 1 : fmt.Fprintf(&buf, " read_sampling_multiplier=%d\n", o.Experimental.ReadSamplingMultiplier)
1439 1 : fmt.Fprintf(&buf, " num_deletions_threshold=%d\n", o.Experimental.NumDeletionsThreshold)
1440 1 : fmt.Fprintf(&buf, " deletion_size_ratio_threshold=%f\n", o.Experimental.DeletionSizeRatioThreshold)
1441 1 : fmt.Fprintf(&buf, " tombstone_dense_compaction_threshold=%f\n", o.Experimental.TombstoneDenseCompactionThreshold)
1442 1 : // We no longer care about strict_wal_tail, but set it to true in case an
1443 1 : // older version reads the options.
1444 1 : fmt.Fprintf(&buf, " strict_wal_tail=%t\n", true)
1445 1 : fmt.Fprintf(&buf, " table_cache_shards=%d\n", o.Experimental.TableCacheShards)
1446 1 : fmt.Fprintf(&buf, " validate_on_ingest=%t\n", o.Experimental.ValidateOnIngest)
1447 1 : fmt.Fprintf(&buf, " wal_dir=%s\n", o.WALDir)
1448 1 : fmt.Fprintf(&buf, " wal_bytes_per_sync=%d\n", o.WALBytesPerSync)
1449 1 : fmt.Fprintf(&buf, " max_writer_concurrency=%d\n", o.Experimental.MaxWriterConcurrency)
1450 1 : fmt.Fprintf(&buf, " force_writer_parallelism=%t\n", o.Experimental.ForceWriterParallelism)
1451 1 : fmt.Fprintf(&buf, " secondary_cache_size_bytes=%d\n", o.Experimental.SecondaryCacheSizeBytes)
1452 1 : fmt.Fprintf(&buf, " create_on_shared=%d\n", o.Experimental.CreateOnShared)
1453 1 :
1454 1 : // Private options.
1455 1 : //
1456 1 : // These options are only encoded if true, because we do not want them to
1457 1 : // appear in production serialized Options files, since they're testing-only
1458 1 : // options. They're only serialized when true, which still ensures that the
1459 1 : // metamorphic tests may propagate them to subprocesses.
1460 1 : if o.private.disableDeleteOnlyCompactions {
1461 1 : fmt.Fprintln(&buf, " disable_delete_only_compactions=true")
1462 1 : }
1463 1 : if o.private.disableElisionOnlyCompactions {
1464 1 : fmt.Fprintln(&buf, " disable_elision_only_compactions=true")
1465 1 : }
1466 1 : if o.private.disableLazyCombinedIteration {
1467 1 : fmt.Fprintln(&buf, " disable_lazy_combined_iteration=true")
1468 1 : }
1469 :
1470 1 : if o.WALFailover != nil {
1471 1 : unhealthyThreshold, _ := o.WALFailover.FailoverOptions.UnhealthyOperationLatencyThreshold()
1472 1 : fmt.Fprintf(&buf, "\n")
1473 1 : fmt.Fprintf(&buf, "[WAL Failover]\n")
1474 1 : fmt.Fprintf(&buf, " secondary_dir=%s\n", o.WALFailover.Secondary.Dirname)
1475 1 : fmt.Fprintf(&buf, " primary_dir_probe_interval=%s\n", o.WALFailover.FailoverOptions.PrimaryDirProbeInterval)
1476 1 : fmt.Fprintf(&buf, " healthy_probe_latency_threshold=%s\n", o.WALFailover.FailoverOptions.HealthyProbeLatencyThreshold)
1477 1 : fmt.Fprintf(&buf, " healthy_interval=%s\n", o.WALFailover.FailoverOptions.HealthyInterval)
1478 1 : fmt.Fprintf(&buf, " unhealthy_sampling_interval=%s\n", o.WALFailover.FailoverOptions.UnhealthySamplingInterval)
1479 1 : fmt.Fprintf(&buf, " unhealthy_operation_latency_threshold=%s\n", unhealthyThreshold)
1480 1 : fmt.Fprintf(&buf, " elevated_write_stall_threshold_lag=%s\n", o.WALFailover.FailoverOptions.ElevatedWriteStallThresholdLag)
1481 1 : }
1482 :
1483 1 : for i := range o.Levels {
1484 1 : l := &o.Levels[i]
1485 1 : fmt.Fprintf(&buf, "\n")
1486 1 : fmt.Fprintf(&buf, "[Level \"%d\"]\n", i)
1487 1 : fmt.Fprintf(&buf, " block_restart_interval=%d\n", l.BlockRestartInterval)
1488 1 : fmt.Fprintf(&buf, " block_size=%d\n", l.BlockSize)
1489 1 : fmt.Fprintf(&buf, " block_size_threshold=%d\n", l.BlockSizeThreshold)
1490 1 : fmt.Fprintf(&buf, " compression=%s\n", resolveDefaultCompression(l.Compression()))
1491 1 : fmt.Fprintf(&buf, " filter_policy=%s\n", filterPolicyName(l.FilterPolicy))
1492 1 : fmt.Fprintf(&buf, " filter_type=%s\n", l.FilterType)
1493 1 : fmt.Fprintf(&buf, " index_block_size=%d\n", l.IndexBlockSize)
1494 1 : fmt.Fprintf(&buf, " target_file_size=%d\n", l.TargetFileSize)
1495 1 : }
1496 :
1497 1 : return buf.String()
1498 : }
1499 :
1500 : // parseOptions takes options serialized by Options.String() and parses them into
1501 : // keys and values, calling fn for each one.
1502 1 : func parseOptions(s string, fn func(section, key, value string) error) error {
1503 1 : var section string
1504 1 : for _, line := range strings.Split(s, "\n") {
1505 1 : line = strings.TrimSpace(line)
1506 1 : if len(line) == 0 {
1507 1 : // Skip blank lines.
1508 1 : continue
1509 : }
1510 1 : if line[0] == ';' || line[0] == '#' {
1511 0 : // Skip comments.
1512 0 : continue
1513 : }
1514 1 : n := len(line)
1515 1 : if line[0] == '[' && line[n-1] == ']' {
1516 1 : // Parse section.
1517 1 : section = line[1 : n-1]
1518 1 : continue
1519 : }
1520 :
1521 1 : pos := strings.Index(line, "=")
1522 1 : if pos < 0 {
1523 1 : const maxLen = 50
1524 1 : if len(line) > maxLen {
1525 0 : line = line[:maxLen-3] + "..."
1526 0 : }
1527 1 : return base.CorruptionErrorf("invalid key=value syntax: %q", errors.Safe(line))
1528 : }
1529 :
1530 1 : key := strings.TrimSpace(line[:pos])
1531 1 : value := strings.TrimSpace(line[pos+1:])
1532 1 :
1533 1 : // RocksDB uses a similar (INI-style) syntax for the OPTIONS file, but
1534 1 : // different section names and keys. The "CFOptions ..." paths are the
1535 1 : // RocksDB versions which we map to the Pebble paths.
1536 1 : mappedSection := section
1537 1 : if section == `CFOptions "default"` {
1538 1 : mappedSection = "Options"
1539 1 : switch key {
1540 1 : case "comparator":
1541 1 : key = "comparer"
1542 1 : case "merge_operator":
1543 1 : key = "merger"
1544 : }
1545 : }
1546 :
1547 1 : if err := fn(mappedSection, key, value); err != nil {
1548 1 : return err
1549 1 : }
1550 : }
1551 1 : return nil
1552 : }
1553 :
1554 : // ParseHooks contains callbacks to create options fields which can have
1555 : // user-defined implementations.
1556 : type ParseHooks struct {
1557 : NewCache func(size int64) *Cache
1558 : NewCleaner func(name string) (Cleaner, error)
1559 : NewComparer func(name string) (*Comparer, error)
1560 : NewFilterPolicy func(name string) (FilterPolicy, error)
1561 : NewMerger func(name string) (*Merger, error)
1562 : SkipUnknown func(name, value string) bool
1563 : }
1564 :
1565 : // Parse parses the options from the specified string. Note that certain
1566 : // options cannot be parsed into populated fields. For example, comparer and
1567 : // merger.
1568 1 : func (o *Options) Parse(s string, hooks *ParseHooks) error {
1569 1 : return parseOptions(s, func(section, key, value string) error {
1570 1 : // WARNING: DO NOT remove entries from the switches below because doing so
1571 1 : // causes a key previously written to the OPTIONS file to be considered unknown,
1572 1 : // a backwards incompatible change. Instead, leave in support for parsing the
1573 1 : // key but simply don't parse the value.
1574 1 :
1575 1 : switch {
1576 1 : case section == "Version":
1577 1 : switch key {
1578 1 : case "pebble_version":
1579 0 : default:
1580 0 : if hooks != nil && hooks.SkipUnknown != nil && hooks.SkipUnknown(section+"."+key, value) {
1581 0 : return nil
1582 0 : }
1583 0 : return errors.Errorf("pebble: unknown option: %s.%s",
1584 0 : errors.Safe(section), errors.Safe(key))
1585 : }
1586 1 : return nil
1587 :
1588 1 : case section == "Options":
1589 1 : var err error
1590 1 : switch key {
1591 1 : case "bytes_per_sync":
1592 1 : o.BytesPerSync, err = strconv.Atoi(value)
1593 1 : case "cache_size":
1594 1 : var n int64
1595 1 : n, err = strconv.ParseInt(value, 10, 64)
1596 1 : if err == nil && hooks != nil && hooks.NewCache != nil {
1597 1 : if o.Cache != nil {
1598 0 : o.Cache.Unref()
1599 0 : }
1600 1 : o.Cache = hooks.NewCache(n)
1601 : }
1602 : // We avoid calling cache.New in parsing because it makes it
1603 : // too easy to leak a cache.
1604 1 : case "cleaner":
1605 1 : switch value {
1606 1 : case "archive":
1607 1 : o.Cleaner = ArchiveCleaner{}
1608 1 : case "delete":
1609 1 : o.Cleaner = DeleteCleaner{}
1610 0 : default:
1611 0 : if hooks != nil && hooks.NewCleaner != nil {
1612 0 : o.Cleaner, err = hooks.NewCleaner(value)
1613 0 : }
1614 : }
1615 1 : case "comparer":
1616 1 : switch value {
1617 1 : case "leveldb.BytewiseComparator":
1618 1 : o.Comparer = DefaultComparer
1619 1 : default:
1620 1 : if hooks != nil && hooks.NewComparer != nil {
1621 1 : o.Comparer, err = hooks.NewComparer(value)
1622 1 : }
1623 : }
1624 1 : case "compaction_debt_concurrency":
1625 1 : o.Experimental.CompactionDebtConcurrency, err = strconv.ParseUint(value, 10, 64)
1626 0 : case "delete_range_flush_delay":
1627 0 : // NB: This is a deprecated serialization of the
1628 0 : // `flush_delay_delete_range`.
1629 0 : o.FlushDelayDeleteRange, err = time.ParseDuration(value)
1630 1 : case "disable_delete_only_compactions":
1631 1 : o.private.disableDeleteOnlyCompactions, err = strconv.ParseBool(value)
1632 1 : case "disable_elision_only_compactions":
1633 1 : o.private.disableElisionOnlyCompactions, err = strconv.ParseBool(value)
1634 1 : case "disable_ingest_as_flushable":
1635 1 : var v bool
1636 1 : v, err = strconv.ParseBool(value)
1637 1 : if err == nil {
1638 1 : o.Experimental.DisableIngestAsFlushable = func() bool { return v }
1639 : }
1640 1 : case "disable_lazy_combined_iteration":
1641 1 : o.private.disableLazyCombinedIteration, err = strconv.ParseBool(value)
1642 1 : case "disable_wal":
1643 1 : o.DisableWAL, err = strconv.ParseBool(value)
1644 1 : case "flush_delay_delete_range":
1645 1 : o.FlushDelayDeleteRange, err = time.ParseDuration(value)
1646 1 : case "flush_delay_range_key":
1647 1 : o.FlushDelayRangeKey, err = time.ParseDuration(value)
1648 1 : case "flush_split_bytes":
1649 1 : o.FlushSplitBytes, err = strconv.ParseInt(value, 10, 64)
1650 1 : case "format_major_version":
1651 1 : // NB: The version written here may be stale. Open does
1652 1 : // not use the format major version encoded in the
1653 1 : // OPTIONS file other than to validate that the encoded
1654 1 : // version is valid right here.
1655 1 : var v uint64
1656 1 : v, err = strconv.ParseUint(value, 10, 64)
1657 1 : if vers := FormatMajorVersion(v); vers > internalFormatNewest || vers == FormatDefault {
1658 0 : err = errors.Newf("unsupported format major version %d", o.FormatMajorVersion)
1659 0 : }
1660 1 : if err == nil {
1661 1 : o.FormatMajorVersion = FormatMajorVersion(v)
1662 1 : }
1663 1 : case "l0_compaction_concurrency":
1664 1 : o.Experimental.L0CompactionConcurrency, err = strconv.Atoi(value)
1665 1 : case "l0_compaction_file_threshold":
1666 1 : o.L0CompactionFileThreshold, err = strconv.Atoi(value)
1667 1 : case "l0_compaction_threshold":
1668 1 : o.L0CompactionThreshold, err = strconv.Atoi(value)
1669 1 : case "l0_stop_writes_threshold":
1670 1 : o.L0StopWritesThreshold, err = strconv.Atoi(value)
1671 0 : case "l0_sublevel_compactions":
1672 : // Do nothing; option existed in older versions of pebble.
1673 1 : case "lbase_max_bytes":
1674 1 : o.LBaseMaxBytes, err = strconv.ParseInt(value, 10, 64)
1675 1 : case "level_multiplier":
1676 1 : o.Experimental.LevelMultiplier, err = strconv.Atoi(value)
1677 1 : case "max_concurrent_compactions":
1678 1 : var concurrentCompactions int
1679 1 : concurrentCompactions, err = strconv.Atoi(value)
1680 1 : if concurrentCompactions <= 0 {
1681 0 : err = errors.New("max_concurrent_compactions cannot be <= 0")
1682 1 : } else {
1683 1 : o.MaxConcurrentCompactions = func() int { return concurrentCompactions }
1684 : }
1685 1 : case "max_concurrent_downloads":
1686 1 : var concurrentDownloads int
1687 1 : concurrentDownloads, err = strconv.Atoi(value)
1688 1 : if concurrentDownloads <= 0 {
1689 0 : err = errors.New("max_concurrent_compactions cannot be <= 0")
1690 1 : } else {
1691 1 : o.MaxConcurrentDownloads = func() int { return concurrentDownloads }
1692 : }
1693 1 : case "max_manifest_file_size":
1694 1 : o.MaxManifestFileSize, err = strconv.ParseInt(value, 10, 64)
1695 1 : case "max_open_files":
1696 1 : o.MaxOpenFiles, err = strconv.Atoi(value)
1697 1 : case "mem_table_size":
1698 1 : o.MemTableSize, err = strconv.ParseUint(value, 10, 64)
1699 1 : case "mem_table_stop_writes_threshold":
1700 1 : o.MemTableStopWritesThreshold, err = strconv.Atoi(value)
1701 0 : case "min_compaction_rate":
1702 : // Do nothing; option existed in older versions of pebble, and
1703 : // may be meaningful again eventually.
1704 1 : case "min_deletion_rate":
1705 1 : o.TargetByteDeletionRate, err = strconv.Atoi(value)
1706 0 : case "min_flush_rate":
1707 : // Do nothing; option existed in older versions of pebble, and
1708 : // may be meaningful again eventually.
1709 1 : case "multilevel_compaction_heuristic":
1710 1 : switch {
1711 1 : case value == "none":
1712 1 : o.Experimental.MultiLevelCompactionHeuristic = NoMultiLevel{}
1713 1 : case strings.HasPrefix(value, "wamp"):
1714 1 : fields := strings.FieldsFunc(strings.TrimPrefix(value, "wamp"), func(r rune) bool {
1715 1 : return unicode.IsSpace(r) || r == ',' || r == '(' || r == ')'
1716 1 : })
1717 1 : if len(fields) != 2 {
1718 0 : err = errors.Newf("require 2 arguments")
1719 0 : }
1720 1 : var h WriteAmpHeuristic
1721 1 : if err == nil {
1722 1 : h.AddPropensity, err = strconv.ParseFloat(fields[0], 64)
1723 1 : }
1724 1 : if err == nil {
1725 1 : h.AllowL0, err = strconv.ParseBool(fields[1])
1726 1 : }
1727 1 : if err == nil {
1728 1 : o.Experimental.MultiLevelCompactionHeuristic = h
1729 1 : } else {
1730 0 : err = errors.Wrapf(err, "unexpected wamp heuristic arguments: %s", value)
1731 0 : }
1732 0 : default:
1733 0 : err = errors.Newf("unrecognized multilevel compaction heuristic: %s", value)
1734 : }
1735 0 : case "point_tombstone_weight":
1736 : // Do nothing; deprecated.
1737 1 : case "strict_wal_tail":
1738 1 : var strictWALTail bool
1739 1 : strictWALTail, err = strconv.ParseBool(value)
1740 1 : if err == nil && !strictWALTail {
1741 0 : err = errors.Newf("reading from versions with strict_wal_tail=false no longer supported")
1742 0 : }
1743 1 : case "merger":
1744 1 : switch value {
1745 0 : case "nullptr":
1746 0 : o.Merger = nil
1747 1 : case "pebble.concatenate":
1748 1 : o.Merger = DefaultMerger
1749 1 : default:
1750 1 : if hooks != nil && hooks.NewMerger != nil {
1751 1 : o.Merger, err = hooks.NewMerger(value)
1752 1 : }
1753 : }
1754 1 : case "read_compaction_rate":
1755 1 : o.Experimental.ReadCompactionRate, err = strconv.ParseInt(value, 10, 64)
1756 1 : case "read_sampling_multiplier":
1757 1 : o.Experimental.ReadSamplingMultiplier, err = strconv.ParseInt(value, 10, 64)
1758 1 : case "num_deletions_threshold":
1759 1 : o.Experimental.NumDeletionsThreshold, err = strconv.Atoi(value)
1760 1 : case "deletion_size_ratio_threshold":
1761 1 : val, parseErr := strconv.ParseFloat(value, 32)
1762 1 : o.Experimental.DeletionSizeRatioThreshold = float32(val)
1763 1 : err = parseErr
1764 1 : case "tombstone_dense_compaction_threshold":
1765 1 : o.Experimental.TombstoneDenseCompactionThreshold, err = strconv.ParseFloat(value, 64)
1766 1 : case "table_cache_shards":
1767 1 : o.Experimental.TableCacheShards, err = strconv.Atoi(value)
1768 0 : case "table_format":
1769 0 : switch value {
1770 0 : case "leveldb":
1771 0 : case "rocksdbv2":
1772 0 : default:
1773 0 : return errors.Errorf("pebble: unknown table format: %q", errors.Safe(value))
1774 : }
1775 1 : case "table_property_collectors":
1776 : // No longer implemented; ignore.
1777 1 : case "validate_on_ingest":
1778 1 : o.Experimental.ValidateOnIngest, err = strconv.ParseBool(value)
1779 1 : case "wal_dir":
1780 1 : o.WALDir = value
1781 1 : case "wal_bytes_per_sync":
1782 1 : o.WALBytesPerSync, err = strconv.Atoi(value)
1783 1 : case "max_writer_concurrency":
1784 1 : o.Experimental.MaxWriterConcurrency, err = strconv.Atoi(value)
1785 1 : case "force_writer_parallelism":
1786 1 : o.Experimental.ForceWriterParallelism, err = strconv.ParseBool(value)
1787 1 : case "secondary_cache_size_bytes":
1788 1 : o.Experimental.SecondaryCacheSizeBytes, err = strconv.ParseInt(value, 10, 64)
1789 1 : case "create_on_shared":
1790 1 : var createOnSharedInt int64
1791 1 : createOnSharedInt, err = strconv.ParseInt(value, 10, 64)
1792 1 : o.Experimental.CreateOnShared = remote.CreateOnSharedStrategy(createOnSharedInt)
1793 0 : default:
1794 0 : if hooks != nil && hooks.SkipUnknown != nil && hooks.SkipUnknown(section+"."+key, value) {
1795 0 : return nil
1796 0 : }
1797 0 : return errors.Errorf("pebble: unknown option: %s.%s",
1798 0 : errors.Safe(section), errors.Safe(key))
1799 : }
1800 1 : return err
1801 :
1802 1 : case section == "WAL Failover":
1803 1 : if o.WALFailover == nil {
1804 1 : o.WALFailover = new(WALFailoverOptions)
1805 1 : }
1806 1 : var err error
1807 1 : switch key {
1808 1 : case "secondary_dir":
1809 1 : o.WALFailover.Secondary = wal.Dir{Dirname: value, FS: vfs.Default}
1810 1 : case "primary_dir_probe_interval":
1811 1 : o.WALFailover.PrimaryDirProbeInterval, err = time.ParseDuration(value)
1812 1 : case "healthy_probe_latency_threshold":
1813 1 : o.WALFailover.HealthyProbeLatencyThreshold, err = time.ParseDuration(value)
1814 1 : case "healthy_interval":
1815 1 : o.WALFailover.HealthyInterval, err = time.ParseDuration(value)
1816 1 : case "unhealthy_sampling_interval":
1817 1 : o.WALFailover.UnhealthySamplingInterval, err = time.ParseDuration(value)
1818 1 : case "unhealthy_operation_latency_threshold":
1819 1 : var threshold time.Duration
1820 1 : threshold, err = time.ParseDuration(value)
1821 1 : o.WALFailover.UnhealthyOperationLatencyThreshold = func() (time.Duration, bool) {
1822 1 : return threshold, true
1823 1 : }
1824 1 : case "elevated_write_stall_threshold_lag":
1825 1 : o.WALFailover.ElevatedWriteStallThresholdLag, err = time.ParseDuration(value)
1826 0 : default:
1827 0 : if hooks != nil && hooks.SkipUnknown != nil && hooks.SkipUnknown(section+"."+key, value) {
1828 0 : return nil
1829 0 : }
1830 0 : return errors.Errorf("pebble: unknown option: %s.%s",
1831 0 : errors.Safe(section), errors.Safe(key))
1832 : }
1833 1 : return err
1834 :
1835 1 : case strings.HasPrefix(section, "Level "):
1836 1 : var index int
1837 1 : if n, err := fmt.Sscanf(section, `Level "%d"`, &index); err != nil {
1838 0 : return err
1839 1 : } else if n != 1 {
1840 0 : if hooks != nil && hooks.SkipUnknown != nil && hooks.SkipUnknown(section, value) {
1841 0 : return nil
1842 0 : }
1843 0 : return errors.Errorf("pebble: unknown section: %q", errors.Safe(section))
1844 : }
1845 :
1846 1 : if len(o.Levels) <= index {
1847 1 : newLevels := make([]LevelOptions, index+1)
1848 1 : copy(newLevels, o.Levels)
1849 1 : o.Levels = newLevels
1850 1 : }
1851 1 : l := &o.Levels[index]
1852 1 :
1853 1 : var err error
1854 1 : switch key {
1855 1 : case "block_restart_interval":
1856 1 : l.BlockRestartInterval, err = strconv.Atoi(value)
1857 1 : case "block_size":
1858 1 : l.BlockSize, err = strconv.Atoi(value)
1859 1 : case "block_size_threshold":
1860 1 : l.BlockSizeThreshold, err = strconv.Atoi(value)
1861 1 : case "compression":
1862 1 : switch value {
1863 0 : case "Default":
1864 0 : l.Compression = func() Compression { return DefaultCompression }
1865 1 : case "NoCompression":
1866 1 : l.Compression = func() Compression { return NoCompression }
1867 1 : case "Snappy":
1868 1 : l.Compression = func() Compression { return SnappyCompression }
1869 1 : case "ZSTD":
1870 1 : l.Compression = func() Compression { return ZstdCompression }
1871 0 : default:
1872 0 : return errors.Errorf("pebble: unknown compression: %q", errors.Safe(value))
1873 : }
1874 1 : case "filter_policy":
1875 1 : if hooks != nil && hooks.NewFilterPolicy != nil {
1876 1 : l.FilterPolicy, err = hooks.NewFilterPolicy(value)
1877 1 : }
1878 1 : case "filter_type":
1879 1 : switch value {
1880 1 : case "table":
1881 1 : l.FilterType = TableFilter
1882 0 : default:
1883 0 : return errors.Errorf("pebble: unknown filter type: %q", errors.Safe(value))
1884 : }
1885 1 : case "index_block_size":
1886 1 : l.IndexBlockSize, err = strconv.Atoi(value)
1887 1 : case "target_file_size":
1888 1 : l.TargetFileSize, err = strconv.ParseInt(value, 10, 64)
1889 0 : default:
1890 0 : if hooks != nil && hooks.SkipUnknown != nil && hooks.SkipUnknown(section+"."+key, value) {
1891 0 : return nil
1892 0 : }
1893 0 : return errors.Errorf("pebble: unknown option: %s.%s", errors.Safe(section), errors.Safe(key))
1894 : }
1895 1 : return err
1896 : }
1897 1 : if hooks != nil && hooks.SkipUnknown != nil && hooks.SkipUnknown(section+"."+key, value) {
1898 1 : return nil
1899 1 : }
1900 0 : return errors.Errorf("pebble: unknown section: %q", errors.Safe(section))
1901 : })
1902 : }
1903 :
1904 : // ErrMissingWALRecoveryDir is an error returned when a database is attempted to be
1905 : // opened without supplying a Options.WALRecoveryDir entry for a directory that
1906 : // may contain WALs required to recover a consistent database state.
1907 : type ErrMissingWALRecoveryDir struct {
1908 : Dir string
1909 : }
1910 :
1911 : // Error implements error.
1912 1 : func (e ErrMissingWALRecoveryDir) Error() string {
1913 1 : return fmt.Sprintf("directory %q may contain relevant WALs", e.Dir)
1914 1 : }
1915 :
1916 : // CheckCompatibility verifies the options are compatible with the previous options
1917 : // serialized by Options.String(). For example, the Comparer and Merger must be
1918 : // the same, or data will not be able to be properly read from the DB.
1919 : //
1920 : // This function only looks at specific keys and does not error out if the
1921 : // options are newer and contain unknown keys.
1922 1 : func (o *Options) CheckCompatibility(previousOptions string) error {
1923 1 : return parseOptions(previousOptions, func(section, key, value string) error {
1924 1 : switch section + "." + key {
1925 1 : case "Options.comparer":
1926 1 : if value != o.Comparer.Name {
1927 1 : return errors.Errorf("pebble: comparer name from file %q != comparer name from options %q",
1928 1 : errors.Safe(value), errors.Safe(o.Comparer.Name))
1929 1 : }
1930 1 : case "Options.merger":
1931 1 : // RocksDB allows the merge operator to be unspecified, in which case it
1932 1 : // shows up as "nullptr".
1933 1 : if value != "nullptr" && value != o.Merger.Name {
1934 1 : return errors.Errorf("pebble: merger name from file %q != merger name from options %q",
1935 1 : errors.Safe(value), errors.Safe(o.Merger.Name))
1936 1 : }
1937 1 : case "Options.wal_dir", "WAL Failover.secondary_dir":
1938 1 : switch {
1939 1 : case o.WALDir == value:
1940 1 : return nil
1941 1 : case o.WALFailover != nil && o.WALFailover.Secondary.Dirname == value:
1942 1 : return nil
1943 1 : default:
1944 1 : for _, d := range o.WALRecoveryDirs {
1945 1 : if d.Dirname == value {
1946 1 : return nil
1947 1 : }
1948 : }
1949 1 : return ErrMissingWALRecoveryDir{Dir: value}
1950 : }
1951 : }
1952 1 : return nil
1953 : })
1954 : }
1955 :
1956 : // Validate verifies that the options are mutually consistent. For example,
1957 : // L0StopWritesThreshold must be >= L0CompactionThreshold, otherwise a write
1958 : // stall would persist indefinitely.
1959 1 : func (o *Options) Validate() error {
1960 1 : // Note that we can presume Options.EnsureDefaults has been called, so there
1961 1 : // is no need to check for zero values.
1962 1 :
1963 1 : var buf strings.Builder
1964 1 : if o.Experimental.L0CompactionConcurrency < 1 {
1965 1 : fmt.Fprintf(&buf, "L0CompactionConcurrency (%d) must be >= 1\n",
1966 1 : o.Experimental.L0CompactionConcurrency)
1967 1 : }
1968 1 : if o.L0StopWritesThreshold < o.L0CompactionThreshold {
1969 1 : fmt.Fprintf(&buf, "L0StopWritesThreshold (%d) must be >= L0CompactionThreshold (%d)\n",
1970 1 : o.L0StopWritesThreshold, o.L0CompactionThreshold)
1971 1 : }
1972 1 : if uint64(o.MemTableSize) >= maxMemTableSize {
1973 1 : fmt.Fprintf(&buf, "MemTableSize (%s) must be < %s\n",
1974 1 : humanize.Bytes.Uint64(uint64(o.MemTableSize)), humanize.Bytes.Uint64(maxMemTableSize))
1975 1 : }
1976 1 : if o.MemTableStopWritesThreshold < 2 {
1977 1 : fmt.Fprintf(&buf, "MemTableStopWritesThreshold (%d) must be >= 2\n",
1978 1 : o.MemTableStopWritesThreshold)
1979 1 : }
1980 1 : if o.FormatMajorVersion < FormatMinSupported || o.FormatMajorVersion > internalFormatNewest {
1981 0 : fmt.Fprintf(&buf, "FormatMajorVersion (%d) must be between %d and %d\n",
1982 0 : o.FormatMajorVersion, FormatMinSupported, internalFormatNewest)
1983 0 : }
1984 1 : if o.Experimental.CreateOnShared != remote.CreateOnSharedNone && o.FormatMajorVersion < FormatMinForSharedObjects {
1985 0 : fmt.Fprintf(&buf, "FormatMajorVersion (%d) when CreateOnShared is set must be at least %d\n",
1986 0 : o.FormatMajorVersion, FormatMinForSharedObjects)
1987 0 :
1988 0 : }
1989 1 : if o.TableCache != nil && o.Cache != o.TableCache.cache {
1990 1 : fmt.Fprintf(&buf, "underlying cache in the TableCache and the Cache dont match\n")
1991 1 : }
1992 1 : if buf.Len() == 0 {
1993 1 : return nil
1994 1 : }
1995 1 : return errors.New(buf.String())
1996 : }
1997 :
1998 : // MakeReaderOptions constructs sstable.ReaderOptions from the corresponding
1999 : // options in the receiver.
2000 1 : func (o *Options) MakeReaderOptions() sstable.ReaderOptions {
2001 1 : var readerOpts sstable.ReaderOptions
2002 1 : if o != nil {
2003 1 : readerOpts.LoadBlockSema = o.LoadBlockSema
2004 1 : readerOpts.Comparer = o.Comparer
2005 1 : readerOpts.Merger = o.Merger
2006 1 : readerOpts.Filters = o.Filters
2007 1 : readerOpts.LoggerAndTracer = o.LoggerAndTracer
2008 1 : }
2009 1 : return readerOpts
2010 : }
2011 :
2012 : // MakeWriterOptions constructs sstable.WriterOptions for the specified level
2013 : // from the corresponding options in the receiver.
2014 1 : func (o *Options) MakeWriterOptions(level int, format sstable.TableFormat) sstable.WriterOptions {
2015 1 : var writerOpts sstable.WriterOptions
2016 1 : writerOpts.TableFormat = format
2017 1 : if o != nil {
2018 1 : writerOpts.Comparer = o.Comparer
2019 1 : if o.Merger != nil {
2020 1 : writerOpts.MergerName = o.Merger.Name
2021 1 : }
2022 1 : writerOpts.BlockPropertyCollectors = o.BlockPropertyCollectors
2023 : }
2024 1 : if format >= sstable.TableFormatPebblev3 {
2025 1 : writerOpts.ShortAttributeExtractor = o.Experimental.ShortAttributeExtractor
2026 1 : writerOpts.RequiredInPlaceValueBound = o.Experimental.RequiredInPlaceValueBound
2027 1 : if format >= sstable.TableFormatPebblev4 && level == numLevels-1 {
2028 1 : writerOpts.WritingToLowestLevel = true
2029 1 : }
2030 : }
2031 1 : levelOpts := o.Level(level)
2032 1 : writerOpts.BlockRestartInterval = levelOpts.BlockRestartInterval
2033 1 : writerOpts.BlockSize = levelOpts.BlockSize
2034 1 : writerOpts.BlockSizeThreshold = levelOpts.BlockSizeThreshold
2035 1 : writerOpts.Compression = resolveDefaultCompression(levelOpts.Compression())
2036 1 : writerOpts.FilterPolicy = levelOpts.FilterPolicy
2037 1 : writerOpts.FilterType = levelOpts.FilterType
2038 1 : writerOpts.IndexBlockSize = levelOpts.IndexBlockSize
2039 1 : writerOpts.AllocatorSizeClasses = o.AllocatorSizeClasses
2040 1 : writerOpts.NumDeletionsThreshold = o.Experimental.NumDeletionsThreshold
2041 1 : writerOpts.DeletionSizeRatioThreshold = o.Experimental.DeletionSizeRatioThreshold
2042 1 : return writerOpts
2043 : }
2044 :
2045 1 : func resolveDefaultCompression(c Compression) Compression {
2046 1 : if c <= DefaultCompression || c >= block.NCompression {
2047 1 : c = SnappyCompression
2048 1 : }
2049 1 : return c
2050 : }
|
