Line data Source code
1 : // Copyright 2020 The LevelDB-Go and Pebble Authors. All rights reserved. Use
2 : // of this source code is governed by a BSD-style license that can be found in
3 : // the LICENSE file.
4 :
5 : package pebble
6 :
7 : import (
8 : "bytes"
9 : "context"
10 : "fmt"
11 : "io"
12 : "runtime/debug"
13 : "sync"
14 : "sync/atomic"
15 : "unsafe"
16 :
17 : "github.com/cockroachdb/errors"
18 : "github.com/cockroachdb/pebble/internal/base"
19 : "github.com/cockroachdb/pebble/internal/cache"
20 : "github.com/cockroachdb/pebble/internal/genericcache"
21 : "github.com/cockroachdb/pebble/internal/invariants"
22 : "github.com/cockroachdb/pebble/internal/keyspan"
23 : "github.com/cockroachdb/pebble/internal/keyspan/keyspanimpl"
24 : "github.com/cockroachdb/pebble/internal/manifest"
25 : "github.com/cockroachdb/pebble/internal/sstableinternal"
26 : "github.com/cockroachdb/pebble/objstorage"
27 : "github.com/cockroachdb/pebble/objstorage/objstorageprovider/objiotracing"
28 : "github.com/cockroachdb/pebble/sstable"
29 : "github.com/cockroachdb/pebble/sstable/block"
30 : "github.com/cockroachdb/pebble/sstable/valblk"
31 : "github.com/cockroachdb/redact"
32 : )
33 :
34 : var emptyIter = &errorIter{err: nil}
35 : var emptyKeyspanIter = &errorKeyspanIter{err: nil}
36 :
37 : // tableNewIters creates new iterators (point, range deletion and/or range key)
38 : // for the given table metadata. Which of the various iterator kinds the user is
39 : // requesting is specified with the iterKinds bitmap.
40 : //
41 : // On success, the requested subset of iters.{point,rangeDel,rangeKey} are
42 : // populated with iterators.
43 : //
44 : // If a point iterator is requested and the operation was successful,
45 : // iters.point is guaranteed to be non-nil and must be closed when the caller is
46 : // finished.
47 : //
48 : // If a range deletion or range key iterator is requested, the corresponding
49 : // iterator may be nil if the table does not contain any keys of the
50 : // corresponding kind. The returned iterSet type provides RangeDeletion() and
51 : // RangeKey() convenience methods that return non-nil empty iterators that may
52 : // be used if the caller requires a non-nil iterator.
53 : //
54 : // On error, all iterators are nil.
55 : //
56 : // The only (non-test) implementation of tableNewIters is
57 : // fileCacheHandle.newIters().
58 : type tableNewIters func(
59 : ctx context.Context,
60 : file *manifest.TableMetadata,
61 : opts *IterOptions,
62 : internalOpts internalIterOpts,
63 : kinds iterKinds,
64 : ) (iterSet, error)
65 :
66 : // tableNewRangeDelIter takes a tableNewIters and returns a TableNewSpanIter
67 : // for the rangedel iterator returned by tableNewIters.
68 1 : func tableNewRangeDelIter(newIters tableNewIters) keyspanimpl.TableNewSpanIter {
69 1 : return func(ctx context.Context, file *manifest.TableMetadata, iterOptions keyspan.SpanIterOptions) (keyspan.FragmentIterator, error) {
70 1 : iters, err := newIters(ctx, file, nil, internalIterOpts{}, iterRangeDeletions)
71 1 : if err != nil {
72 0 : return nil, err
73 0 : }
74 1 : return iters.RangeDeletion(), nil
75 : }
76 : }
77 :
78 : // tableNewRangeKeyIter takes a tableNewIters and returns a TableNewSpanIter
79 : // for the range key iterator returned by tableNewIters.
80 1 : func tableNewRangeKeyIter(newIters tableNewIters) keyspanimpl.TableNewSpanIter {
81 1 : return func(ctx context.Context, file *manifest.TableMetadata, iterOptions keyspan.SpanIterOptions) (keyspan.FragmentIterator, error) {
82 1 : iters, err := newIters(ctx, file, nil, internalIterOpts{}, iterRangeKeys)
83 1 : if err != nil {
84 0 : return nil, err
85 0 : }
86 1 : return iters.RangeKey(), nil
87 : }
88 : }
89 :
90 : // fileCacheHandle is used to access the file cache. Each DB has its own handle.
91 : type fileCacheHandle struct {
92 : fileCache *FileCache
93 :
94 : // The handle contains fields which are unique to each DB. Note that these get
95 : // accessed from all shards, so keep read-only fields separate for read-write
96 : // fields.
97 : loggerAndTracer LoggerAndTracer
98 : blockCacheHandle *cache.Handle
99 : objProvider objstorage.Provider
100 : readerOpts sstable.ReaderOptions
101 :
102 : // iterCount keeps track of how many iterators are open. It is used to keep
103 : // track of leaked iterators on a per-db level.
104 : iterCount atomic.Int32
105 : sstStatsCollector block.CategoryStatsCollector
106 :
107 : // reportCorruptionFn is used for block.ReadEnv.ReportCorruptionFn. It expects
108 : // the first argument to be a `*TableMetadata`.
109 : reportCorruptionFn func(any, error)
110 :
111 : // This struct is only populated in race builds.
112 : raceMu struct {
113 : sync.Mutex
114 : // nextRefID is the next ID to allocate for a new reference.
115 : nextRefID uint64
116 : // openRefs maps reference IDs to the stack trace recorded at creation
117 : // time. It's used to track which call paths leaked open references to
118 : // files.
119 : openRefs map[uint64][]byte
120 : }
121 : }
122 :
123 : // newHandle creates a handle for the FileCache which has its own options. Each
124 : // handle has its own set of files in the cache, separate from those of other
125 : // handles.
126 : func (c *FileCache) newHandle(
127 : cacheHandle *cache.Handle,
128 : objProvider objstorage.Provider,
129 : loggerAndTracer LoggerAndTracer,
130 : readerOpts sstable.ReaderOptions,
131 : reportSSTableCorruptionFn func(*manifest.TableMetadata, error),
132 1 : ) *fileCacheHandle {
133 1 : c.Ref()
134 1 :
135 1 : t := &fileCacheHandle{
136 1 : fileCache: c,
137 1 : loggerAndTracer: loggerAndTracer,
138 1 : blockCacheHandle: cacheHandle,
139 1 : objProvider: objProvider,
140 1 : }
141 1 : t.readerOpts = readerOpts
142 1 : t.readerOpts.FilterMetricsTracker = &sstable.FilterMetricsTracker{}
143 1 : if reportSSTableCorruptionFn != nil {
144 1 : t.reportCorruptionFn = func(arg any, err error) {
145 0 : reportSSTableCorruptionFn(arg.(*manifest.TableMetadata), err)
146 0 : }
147 : }
148 1 : if invariants.RaceEnabled {
149 0 : t.raceMu.openRefs = make(map[uint64][]byte)
150 0 : }
151 1 : return t
152 : }
153 :
154 : // Close the handle, make sure that there will be no further need
155 : // to access any of the files associated with the store.
156 1 : func (h *fileCacheHandle) Close() error {
157 1 : // We want to do some cleanup work here. Check for leaked iterators
158 1 : // by the DB using this container. Note that we'll still perform cleanup
159 1 : // below in the case that there are leaked iterators.
160 1 : var err error
161 1 : if v := h.iterCount.Load(); v > 0 {
162 0 : if !invariants.RaceEnabled {
163 0 : err = errors.Errorf("leaked iterators: %d", errors.Safe(v))
164 0 : } else {
165 0 : var buf bytes.Buffer
166 0 : for _, stack := range h.raceMu.openRefs {
167 0 : fmt.Fprintf(&buf, "%s\n", stack)
168 0 : }
169 0 : err = errors.Errorf("leaked iterators: %d\n%s", errors.Safe(v), buf.String())
170 : }
171 : }
172 :
173 : // EvictAll would panic if there are still outstanding references.
174 1 : if err == nil {
175 1 : keys := h.fileCache.c.EvictAll(func(key fileCacheKey) bool {
176 1 : return key.handle == h
177 1 : })
178 : // Evict any associated blocks in the cache.
179 1 : for i := range keys {
180 1 : h.blockCacheHandle.EvictFile(keys[i].fileNum)
181 1 : }
182 : }
183 :
184 1 : h.fileCache.Unref()
185 1 : // TODO(radu): we have to tolerate metrics() calls after close (see
186 1 : // https://github.com/cockroachdb/cockroach/issues/140454).
187 1 : // *h = fileCacheHandle{}
188 1 : return err
189 : }
190 :
191 : // openFile is called when we insert a new entry in the file cache.
192 : func (h *fileCacheHandle) openFile(
193 : ctx context.Context, fileNum base.DiskFileNum,
194 1 : ) (*sstable.Reader, objstorage.ObjectMetadata, error) {
195 1 : f, err := h.objProvider.OpenForReading(
196 1 : ctx, base.FileTypeTable, fileNum, objstorage.OpenOptions{MustExist: true},
197 1 : )
198 1 : if err != nil {
199 0 : return nil, objstorage.ObjectMetadata{}, err
200 0 : }
201 1 : o := h.readerOpts
202 1 : o.CacheOpts = sstableinternal.CacheOptions{
203 1 : CacheHandle: h.blockCacheHandle,
204 1 : FileNum: fileNum,
205 1 : }
206 1 : r, err := sstable.NewReader(ctx, f, o)
207 1 : if err != nil {
208 0 : return nil, objstorage.ObjectMetadata{}, err
209 0 : }
210 1 : objMeta, err := h.objProvider.Lookup(base.FileTypeTable, fileNum)
211 1 : if err != nil {
212 0 : r.Close()
213 0 : return nil, objstorage.ObjectMetadata{}, err
214 0 : }
215 1 : return r, objMeta, nil
216 : }
217 :
218 : // findOrCreate retrieves an existing reader or creates a new one for the
219 : // backing file of the given table. If a corruption error is encountered,
220 : // reportCorruptionFn() is called.
221 : func (h *fileCacheHandle) findOrCreate(
222 : ctx context.Context, meta *manifest.TableMetadata,
223 1 : ) (genericcache.ValueRef[fileCacheKey, fileCacheValue], error) {
224 1 : key := fileCacheKey{
225 1 : handle: h,
226 1 : fileNum: meta.FileBacking.DiskFileNum,
227 1 : }
228 1 : valRef, err := h.fileCache.c.FindOrCreate(ctx, key)
229 1 : if err != nil && IsCorruptionError(err) {
230 0 : h.reportCorruptionFn(meta, err)
231 0 : }
232 1 : return valRef, err
233 : }
234 :
235 : // Evict the given file from the file cache and the block cache.
236 1 : func (h *fileCacheHandle) Evict(fileNum base.DiskFileNum) {
237 1 : h.fileCache.c.Evict(fileCacheKey{handle: h, fileNum: fileNum})
238 1 : h.blockCacheHandle.EvictFile(fileNum)
239 1 : }
240 :
241 1 : func (h *fileCacheHandle) SSTStatsCollector() *block.CategoryStatsCollector {
242 1 : return &h.sstStatsCollector
243 1 : }
244 :
245 : // Metrics returns metrics for the file cache. Note that the CacheMetrics track
246 : // the global cache which is shared between multiple handles (stores). The
247 : // FilterMetrics are per-handle.
248 1 : func (h *fileCacheHandle) Metrics() (CacheMetrics, FilterMetrics) {
249 1 : m := h.fileCache.c.Metrics()
250 1 : cm := CacheMetrics{
251 1 : Hits: m.Hits,
252 1 : Misses: m.Misses,
253 1 : Count: m.Count,
254 1 : Size: m.Size + m.Count*int64(unsafe.Sizeof(sstable.Reader{})),
255 1 : }
256 1 : fm := h.readerOpts.FilterMetricsTracker.Load()
257 1 : return cm, fm
258 1 : }
259 :
260 : func (h *fileCacheHandle) estimateSize(
261 : meta *tableMetadata, lower, upper []byte,
262 1 : ) (size uint64, err error) {
263 1 : err = h.withCommonReader(context.TODO(), block.NoReadEnv, meta, func(cr sstable.CommonReader, env block.ReadEnv) error {
264 1 : size, err = cr.EstimateDiskUsage(lower, upper)
265 1 : return err
266 1 : })
267 1 : return size, err
268 : }
269 :
270 : // createCommonReader creates a Reader for this file.
271 1 : func createCommonReader(v *fileCacheValue, file *tableMetadata) sstable.CommonReader {
272 1 : // TODO(bananabrick): We suffer an allocation if file is a virtual sstable.
273 1 : r := v.mustSSTableReader()
274 1 : var cr sstable.CommonReader = r
275 1 : if file.Virtual {
276 1 : virtualReader := sstable.MakeVirtualReader(
277 1 : r, file.VirtualMeta().VirtualReaderParams(v.isShared),
278 1 : )
279 1 : cr = &virtualReader
280 1 : }
281 1 : return cr
282 : }
283 :
284 : func (h *fileCacheHandle) withCommonReader(
285 : ctx context.Context,
286 : env block.ReadEnv,
287 : meta *tableMetadata,
288 : fn func(sstable.CommonReader, block.ReadEnv) error,
289 1 : ) error {
290 1 : ref, err := h.findOrCreate(ctx, meta)
291 1 : if err != nil {
292 0 : return err
293 0 : }
294 1 : defer ref.Unref()
295 1 : env.ReportCorruptionFn = h.reportCorruptionFn
296 1 : env.ReportCorruptionArg = meta
297 1 : return fn(createCommonReader(ref.Value(), meta), env)
298 : }
299 :
300 : func (h *fileCacheHandle) withReader(
301 : ctx context.Context,
302 : env block.ReadEnv,
303 : meta physicalMeta,
304 : fn func(*sstable.Reader, block.ReadEnv) error,
305 1 : ) error {
306 1 : ref, err := h.findOrCreate(ctx, meta.TableMetadata)
307 1 : if err != nil {
308 0 : return err
309 0 : }
310 1 : defer ref.Unref()
311 1 : env.ReportCorruptionFn = h.reportCorruptionFn
312 1 : env.ReportCorruptionArg = meta.TableMetadata
313 1 : return fn(ref.Value().reader.(*sstable.Reader), env)
314 : }
315 :
316 : // withVirtualReader fetches a VirtualReader associated with a virtual sstable.
317 : func (h *fileCacheHandle) withVirtualReader(
318 : ctx context.Context,
319 : env block.ReadEnv,
320 : meta virtualMeta,
321 : fn func(sstable.VirtualReader, block.ReadEnv) error,
322 1 : ) error {
323 1 : ref, err := h.findOrCreate(ctx, meta.TableMetadata)
324 1 : if err != nil {
325 0 : return err
326 0 : }
327 1 : defer ref.Unref()
328 1 : v := ref.Value()
329 1 : env.ReportCorruptionFn = h.reportCorruptionFn
330 1 : env.ReportCorruptionArg = &meta.TableMetadata
331 1 : return fn(sstable.MakeVirtualReader(v.mustSSTableReader(), meta.VirtualReaderParams(v.isShared)), env)
332 : }
333 :
334 1 : func (h *fileCacheHandle) IterCount() int64 {
335 1 : return int64(h.iterCount.Load())
336 1 : }
337 :
338 : // FileCache is a shareable cache for open files. Open files are exclusively
339 : // sstable files today.
340 : type FileCache struct {
341 : refs atomic.Int64
342 :
343 : c genericcache.Cache[fileCacheKey, fileCacheValue]
344 : }
345 :
346 : // Ref adds a reference to the file cache. Once a file cache is constructed, the
347 : // cache only remains valid if there is at least one reference to it.
348 1 : func (c *FileCache) Ref() {
349 1 : v := c.refs.Add(1)
350 1 : // We don't want the reference count to ever go from 0 -> 1,
351 1 : // cause a reference count of 0 implies that we've closed the cache.
352 1 : if v <= 1 {
353 0 : panic(fmt.Sprintf("pebble: inconsistent reference count: %d", v))
354 : }
355 : }
356 :
357 : // Unref removes a reference to the file cache.
358 1 : func (c *FileCache) Unref() {
359 1 : v := c.refs.Add(-1)
360 1 : switch {
361 0 : case v < 0:
362 0 : panic(fmt.Sprintf("pebble: inconsistent reference count: %d", v))
363 1 : case v == 0:
364 1 : c.c.Close()
365 1 : c.c = genericcache.Cache[fileCacheKey, fileCacheValue]{}
366 : }
367 : }
368 :
369 : // NewFileCache will create a new file cache with one outstanding reference. It
370 : // is the callers responsibility to call Unref if they will no longer hold a
371 : // reference to the file cache.
372 1 : func NewFileCache(numShards int, size int) *FileCache {
373 1 : if size == 0 {
374 0 : panic("pebble: cannot create a file cache of size 0")
375 1 : } else if numShards == 0 {
376 0 : panic("pebble: cannot create a file cache with 0 shards")
377 : }
378 :
379 1 : c := &FileCache{}
380 1 :
381 1 : // initFn is used whenever a new entry is added to the file cache.
382 1 : initFn := func(ctx context.Context, key fileCacheKey, vRef genericcache.ValueRef[fileCacheKey, fileCacheValue]) error {
383 1 : v := vRef.Value()
384 1 : handle := key.handle
385 1 : v.readerProvider.init(c, key)
386 1 : v.closeHook = func() {
387 1 : // closeHook is called when an iterator is closed; the initialization of
388 1 : // an iterator with this value will happen after a FindOrCreate() call
389 1 : // with returns the same vRef.
390 1 : vRef.Unref()
391 1 : handle.iterCount.Add(-1)
392 1 : }
393 1 : reader, objMeta, err := handle.openFile(ctx, key.fileNum)
394 1 : if err != nil {
395 0 : return errors.Wrapf(err, "pebble: backing file %s error", redact.Safe(key.fileNum))
396 0 : }
397 1 : v.reader = reader
398 1 : v.isShared = objMeta.IsShared()
399 1 : return nil
400 : }
401 :
402 1 : releaseFn := func(v *fileCacheValue) {
403 1 : if v.reader != nil {
404 1 : _ = v.reader.Close()
405 1 : v.reader = nil
406 1 : }
407 : }
408 :
409 1 : c.c.Init(size, numShards, initFn, releaseFn)
410 1 :
411 1 : // Hold a ref to the cache here.
412 1 : c.refs.Store(1)
413 1 :
414 1 : return c
415 : }
416 :
417 : type fileCacheKey struct {
418 : handle *fileCacheHandle
419 : fileNum base.DiskFileNum
420 : }
421 :
422 : // Shard implements the genericcache.Key interface.
423 1 : func (k fileCacheKey) Shard(numShards int) int {
424 1 : // TODO(radu): maybe incorporate a handle ID.
425 1 : return int(uint64(k.fileNum) % uint64(numShards))
426 1 : }
427 :
428 : // checkAndIntersectFilters checks the specific table and block property filters
429 : // for intersection with any available table and block-level properties. Returns
430 : // true for ok if this table should be read by this iterator.
431 : func checkAndIntersectFilters(
432 : r *sstable.Reader,
433 : blockPropertyFilters []BlockPropertyFilter,
434 : boundLimitedFilter sstable.BoundLimitedBlockPropertyFilter,
435 : syntheticSuffix sstable.SyntheticSuffix,
436 1 : ) (ok bool, filterer *sstable.BlockPropertiesFilterer, err error) {
437 1 : if boundLimitedFilter != nil || len(blockPropertyFilters) > 0 {
438 1 : filterer, err = sstable.IntersectsTable(
439 1 : blockPropertyFilters,
440 1 : boundLimitedFilter,
441 1 : r.Properties.UserProperties,
442 1 : syntheticSuffix,
443 1 : )
444 1 : // NB: IntersectsTable will return a nil filterer if the table-level
445 1 : // properties indicate there's no intersection with the provided filters.
446 1 : if filterer == nil || err != nil {
447 1 : return false, nil, err
448 1 : }
449 : }
450 1 : return true, filterer, nil
451 : }
452 :
453 : func (h *fileCacheHandle) newIters(
454 : ctx context.Context,
455 : file *manifest.TableMetadata,
456 : opts *IterOptions,
457 : internalOpts internalIterOpts,
458 : kinds iterKinds,
459 1 : ) (iterSet, error) {
460 1 : // Calling findOrCreate gives us the responsibility of Unref()ing vRef.
461 1 : vRef, err := h.findOrCreate(ctx, file)
462 1 : if err != nil {
463 0 : return iterSet{}, err
464 0 : }
465 :
466 1 : internalOpts.readEnv.ReportCorruptionFn = h.reportCorruptionFn
467 1 : internalOpts.readEnv.ReportCorruptionArg = file
468 1 :
469 1 : v := vRef.Value()
470 1 : r := v.mustSSTableReader()
471 1 : // Note: This suffers an allocation for virtual sstables.
472 1 : cr := createCommonReader(v, file)
473 1 : var iters iterSet
474 1 : if kinds.RangeKey() && file.HasRangeKeys {
475 1 : iters.rangeKey, err = newRangeKeyIter(ctx, r, file, cr, opts.SpanIterOptions(), internalOpts)
476 1 : }
477 1 : if kinds.RangeDeletion() && file.HasPointKeys && err == nil {
478 1 : iters.rangeDeletion, err = newRangeDelIter(ctx, file, cr, h, internalOpts)
479 1 : }
480 1 : if kinds.Point() && err == nil {
481 1 : iters.point, err = h.newPointIter(ctx, v, file, cr, opts, internalOpts, h)
482 1 : }
483 1 : if err != nil {
484 0 : // NB: There's a subtlety here: Because the point iterator is the last
485 0 : // iterator we attempt to create, it's not possible for:
486 0 : // err != nil && iters.point != nil
487 0 : // If it were possible, we'd need to account for it to avoid double
488 0 : // unref-ing here, once during CloseAll and once during `unrefValue`.
489 0 : iters.CloseAll()
490 0 : vRef.Unref()
491 0 : return iterSet{}, err
492 0 : }
493 : // Only point iterators ever require the reader stay pinned in the cache. If
494 : // we're not returning a point iterator to the caller, we need to unref v.
495 : //
496 : // For point iterators, v.closeHook will be called which will release the ref.
497 1 : if iters.point == nil {
498 1 : vRef.Unref()
499 1 : }
500 1 : return iters, nil
501 : }
502 :
503 : // For flushable ingests, we decide whether to use the bloom filter base on
504 : // size.
505 : const filterBlockSizeLimitForFlushableIngests = 64 * 1024
506 :
507 : // newPointIter is an internal helper that constructs a point iterator over a
508 : // sstable. This function is for internal use only, and callers should use
509 : // newIters instead.
510 : func (h *fileCacheHandle) newPointIter(
511 : ctx context.Context,
512 : v *fileCacheValue,
513 : file *manifest.TableMetadata,
514 : cr sstable.CommonReader,
515 : opts *IterOptions,
516 : internalOpts internalIterOpts,
517 : handle *fileCacheHandle,
518 1 : ) (internalIterator, error) {
519 1 : var (
520 1 : hideObsoletePoints bool = false
521 1 : pointKeyFilters []BlockPropertyFilter
522 1 : filterer *sstable.BlockPropertiesFilterer
523 1 : )
524 1 : r := v.mustSSTableReader()
525 1 : if opts != nil {
526 1 : // This code is appending (at most one filter) in-place to
527 1 : // opts.PointKeyFilters even though the slice is shared for iterators in
528 1 : // the same iterator tree. This is acceptable since all the following
529 1 : // properties are true:
530 1 : // - The iterator tree is single threaded, so the shared backing for the
531 1 : // slice is being mutated in a single threaded manner.
532 1 : // - Each shallow copy of the slice has its own notion of length.
533 1 : // - The appended element is always the obsoleteKeyBlockPropertyFilter
534 1 : // struct, which is stateless, so overwriting that struct when creating
535 1 : // one sstable iterator is harmless to other sstable iterators that are
536 1 : // relying on that struct.
537 1 : //
538 1 : // An alternative would be to have different slices for different sstable
539 1 : // iterators, but that requires more work to avoid allocations.
540 1 : //
541 1 : // TODO(bilal): for compaction reads of foreign sstables, we do hide
542 1 : // obsolete points (see sstable.Reader.newCompactionIter) but we don't
543 1 : // apply the obsolete block property filter. We could optimize this by
544 1 : // applying the filter.
545 1 : hideObsoletePoints, pointKeyFilters =
546 1 : r.TryAddBlockPropertyFilterForHideObsoletePoints(
547 1 : opts.snapshotForHideObsoletePoints, file.LargestSeqNum, opts.PointKeyFilters)
548 1 :
549 1 : var ok bool
550 1 : var err error
551 1 : ok, filterer, err = checkAndIntersectFilters(r, pointKeyFilters,
552 1 : internalOpts.boundLimitedFilter, file.SyntheticPrefixAndSuffix.Suffix())
553 1 : if err != nil {
554 0 : return nil, err
555 1 : } else if !ok {
556 1 : // No point keys within the table match the filters.
557 1 : return nil, nil
558 1 : }
559 : }
560 :
561 1 : var iter sstable.Iterator
562 1 : filterBlockSizeLimit := sstable.AlwaysUseFilterBlock
563 1 : if opts != nil {
564 1 : // By default, we don't use block filters for L6 and restrict the size for
565 1 : // flushable ingests, as these blocks can be very big.
566 1 : if !opts.UseL6Filters {
567 1 : if opts.layer == manifest.Level(6) {
568 1 : filterBlockSizeLimit = sstable.NeverUseFilterBlock
569 1 : } else if opts.layer.IsFlushableIngests() {
570 1 : filterBlockSizeLimit = filterBlockSizeLimitForFlushableIngests
571 1 : }
572 : }
573 1 : if opts.layer.IsSet() && !opts.layer.IsFlushableIngests() {
574 1 : ctx = objiotracing.WithLevel(ctx, opts.layer.Level())
575 1 : }
576 : }
577 1 : tableFormat, err := r.TableFormat()
578 1 : if err != nil {
579 0 : return nil, err
580 0 : }
581 :
582 1 : if v.isShared && file.SyntheticSeqNum() != 0 {
583 1 : if tableFormat < sstable.TableFormatPebblev4 {
584 0 : return nil, errors.New("pebble: shared ingested sstable has a lower table format than expected")
585 0 : }
586 : // The table is shared and ingested.
587 1 : hideObsoletePoints = true
588 : }
589 1 : transforms := file.IterTransforms()
590 1 : transforms.HideObsoletePoints = hideObsoletePoints
591 1 : if internalOpts.readEnv.IterStats == nil && opts != nil {
592 1 : internalOpts.readEnv.IterStats = handle.SSTStatsCollector().Accumulator(uint64(uintptr(unsafe.Pointer(r))), opts.Category)
593 1 : }
594 1 : if internalOpts.compaction {
595 1 : iter, err = cr.NewCompactionIter(transforms, internalOpts.readEnv, &v.readerProvider)
596 1 : } else {
597 1 : iter, err = cr.NewPointIter(
598 1 : ctx, transforms, opts.GetLowerBound(), opts.GetUpperBound(), filterer,
599 1 : filterBlockSizeLimit, internalOpts.readEnv, &v.readerProvider)
600 1 : }
601 1 : if err != nil {
602 0 : return nil, err
603 0 : }
604 : // NB: closeHook (v.closeHook) takes responsibility for calling
605 : // unrefValue(v) here. Take care to avoid introducing an allocation here by
606 : // adding a closure.
607 1 : closeHook := h.addReference(v)
608 1 : iter.SetCloseHook(closeHook)
609 1 : return iter, nil
610 : }
611 :
612 1 : func (h *fileCacheHandle) addReference(v *fileCacheValue) (closeHook func()) {
613 1 : h.iterCount.Add(1)
614 1 : closeHook = v.closeHook
615 1 : if invariants.RaceEnabled {
616 0 : stack := debug.Stack()
617 0 : h.raceMu.Lock()
618 0 : refID := h.raceMu.nextRefID
619 0 : h.raceMu.openRefs[refID] = stack
620 0 : h.raceMu.nextRefID++
621 0 : h.raceMu.Unlock()
622 0 : // In race builds, this closeHook closure will force an allocation.
623 0 : // Race builds are already unperformant (and allocate a stack trace), so
624 0 : // we don't care.
625 0 : closeHook = func() {
626 0 : v.closeHook()
627 0 : h.raceMu.Lock()
628 0 : defer h.raceMu.Unlock()
629 0 : delete(h.raceMu.openRefs, refID)
630 0 : }
631 : }
632 1 : return closeHook
633 : }
634 :
635 : // newRangeDelIter is an internal helper that constructs an iterator over a
636 : // sstable's range deletions. This function is for table-cache internal use
637 : // only, and callers should use newIters instead.
638 : func newRangeDelIter(
639 : ctx context.Context,
640 : file *manifest.TableMetadata,
641 : cr sstable.CommonReader,
642 : handle *fileCacheHandle,
643 : internalOpts internalIterOpts,
644 1 : ) (keyspan.FragmentIterator, error) {
645 1 : // NB: range-del iterator does not maintain a reference to the table, nor
646 1 : // does it need to read from it after creation.
647 1 : rangeDelIter, err := cr.NewRawRangeDelIter(ctx, file.FragmentIterTransforms(), internalOpts.readEnv)
648 1 : if err != nil {
649 0 : return nil, err
650 0 : }
651 : // Assert expected bounds in tests.
652 1 : if invariants.Sometimes(50) && rangeDelIter != nil {
653 1 : cmp := base.DefaultComparer.Compare
654 1 : if handle.readerOpts.Comparer != nil {
655 1 : cmp = handle.readerOpts.Comparer.Compare
656 1 : }
657 1 : rangeDelIter = keyspan.AssertBounds(
658 1 : rangeDelIter, file.SmallestPointKey, file.LargestPointKey.UserKey, cmp,
659 1 : )
660 : }
661 1 : return rangeDelIter, nil
662 : }
663 :
664 : // newRangeKeyIter is an internal helper that constructs an iterator over a
665 : // sstable's range keys. This function is for table-cache internal use only, and
666 : // callers should use newIters instead.
667 : func newRangeKeyIter(
668 : ctx context.Context,
669 : r *sstable.Reader,
670 : file *tableMetadata,
671 : cr sstable.CommonReader,
672 : opts keyspan.SpanIterOptions,
673 : internalOpts internalIterOpts,
674 1 : ) (keyspan.FragmentIterator, error) {
675 1 : transforms := file.FragmentIterTransforms()
676 1 : // Don't filter a table's range keys if the file contains RANGEKEYDELs.
677 1 : // The RANGEKEYDELs may delete range keys in other levels. Skipping the
678 1 : // file's range key blocks may surface deleted range keys below. This is
679 1 : // done here, rather than deferring to the block-property collector in order
680 1 : // to maintain parity with point keys and the treatment of RANGEDELs.
681 1 : if r.Properties.NumRangeKeyDels == 0 && len(opts.RangeKeyFilters) > 0 {
682 0 : ok, _, err := checkAndIntersectFilters(r, opts.RangeKeyFilters, nil, transforms.SyntheticSuffix())
683 0 : if err != nil {
684 0 : return nil, err
685 0 : } else if !ok {
686 0 : return nil, nil
687 0 : }
688 : }
689 : // TODO(radu): wrap in an AssertBounds.
690 1 : return cr.NewRawRangeKeyIter(ctx, transforms, internalOpts.readEnv)
691 : }
692 :
693 : // tableCacheShardReaderProvider implements sstable.ReaderProvider for a
694 : // specific table.
695 : type tableCacheShardReaderProvider struct {
696 : c *genericcache.Cache[fileCacheKey, fileCacheValue]
697 : key fileCacheKey
698 :
699 : mu struct {
700 : sync.Mutex
701 : // r is the result of c.FindOrCreate(), only set iff refCount > 0.
702 : r genericcache.ValueRef[fileCacheKey, fileCacheValue]
703 : // refCount is the number of GetReader() calls that have not received a
704 : // corresponding Close().
705 : refCount int
706 : }
707 : }
708 :
709 : var _ valblk.ReaderProvider = &tableCacheShardReaderProvider{}
710 :
711 1 : func (rp *tableCacheShardReaderProvider) init(fc *FileCache, key fileCacheKey) {
712 1 : rp.c = &fc.c
713 1 : rp.key = key
714 1 : rp.mu.r = genericcache.ValueRef[fileCacheKey, fileCacheValue]{}
715 1 : rp.mu.refCount = 0
716 1 : }
717 :
718 : // GetReader implements sstable.ReaderProvider. Note that it is not the
719 : // responsibility of tableCacheShardReaderProvider to ensure that the file
720 : // continues to exist. The ReaderProvider is used in iterators where the
721 : // top-level iterator is pinning the read state and preventing the files from
722 : // being deleted.
723 : //
724 : // The caller must call tableCacheShardReaderProvider.Close.
725 : //
726 : // Note that currently the Reader returned here is only used to read value
727 : // blocks. This reader shouldn't be used for other purposes like reading keys
728 : // outside of virtual sstable bounds.
729 : //
730 : // TODO(bananabrick): We could return a wrapper over the Reader to ensure
731 : // that the reader isn't used for other purposes.
732 : func (rp *tableCacheShardReaderProvider) GetReader(
733 : ctx context.Context,
734 1 : ) (valblk.ExternalBlockReader, error) {
735 1 : rp.mu.Lock()
736 1 : defer rp.mu.Unlock()
737 1 :
738 1 : if rp.mu.refCount > 0 {
739 0 : // We already have a value.
740 0 : rp.mu.refCount++
741 0 : return rp.mu.r.Value().mustSSTableReader(), nil
742 0 : }
743 :
744 : // Calling FindOrCreate gives us the responsibility of Unref()ing r, which
745 : // will happen when rp.mu.refCount reaches 0. Note that if the table is no
746 : // longer in the cache, FindOrCreate will need to do IO (through initFn in
747 : // NewFileCache) to initialize a new Reader. We hold rp.mu during this time so
748 : // that concurrent GetReader calls block until the Reader is created.
749 1 : r, err := rp.c.FindOrCreate(ctx, rp.key)
750 1 : if err != nil {
751 0 : return nil, err
752 0 : }
753 1 : rp.mu.r = r
754 1 : rp.mu.refCount = 1
755 1 : return r.Value().mustSSTableReader(), nil
756 : }
757 :
758 : // Close implements sstable.ReaderProvider.
759 1 : func (rp *tableCacheShardReaderProvider) Close() {
760 1 : rp.mu.Lock()
761 1 : defer rp.mu.Unlock()
762 1 : rp.mu.refCount--
763 1 : if rp.mu.refCount <= 0 {
764 1 : if rp.mu.refCount < 0 {
765 0 : panic("pebble: sstable.ReaderProvider misuse")
766 : }
767 1 : rp.mu.r.Unref()
768 1 : rp.mu.r = genericcache.ValueRef[fileCacheKey, fileCacheValue]{}
769 : }
770 : }
771 :
772 : // getTableProperties returns sst table properties for the backing file.
773 : //
774 : // WARNING! If file is a virtual table, we return the properties of the physical
775 : // table.
776 0 : func (h *fileCacheHandle) getTableProperties(file *tableMetadata) (*sstable.Properties, error) {
777 0 : // Calling findOrCreate gives us the responsibility of decrementing v's refCount here
778 0 : v, err := h.findOrCreate(context.TODO(), file)
779 0 : if err != nil {
780 0 : return nil, err
781 0 : }
782 0 : defer v.Unref()
783 0 :
784 0 : r := v.Value().mustSSTableReader()
785 0 : return &r.Properties, nil
786 : }
787 :
788 : type fileCacheValue struct {
789 : closeHook func()
790 : reader io.Closer // *sstable.Reader
791 : isShared bool
792 :
793 : // readerProvider is embedded here so that we only allocate it once as long as
794 : // the table stays in the cache. Its state is not always logically tied to
795 : // this specific fileCacheShard - if a table goes out of the cache and then
796 : // comes back in, the readerProvider in a now-defunct fileCacheValue can
797 : // still be used and will internally refer to the new fileCacheValue.
798 : readerProvider tableCacheShardReaderProvider
799 : }
800 :
801 : // mustSSTable retrieves the value's *sstable.Reader. It panics if the cached
802 : // file is not a sstable (i.e., it is a blob file).
803 1 : func (v *fileCacheValue) mustSSTableReader() *sstable.Reader {
804 1 : return v.reader.(*sstable.Reader)
805 1 : }
806 :
807 : // iterSet holds a set of iterators of various key kinds, all constructed over
808 : // the same data structure (eg, an sstable). A subset of the fields may be
809 : // populated depending on the `iterKinds` passed to newIters.
810 : type iterSet struct {
811 : point internalIterator
812 : rangeDeletion keyspan.FragmentIterator
813 : rangeKey keyspan.FragmentIterator
814 : }
815 :
816 : // TODO(jackson): Consider adding methods for fast paths that check whether an
817 : // iterator of a particular kind is nil, so that these call sites don't need to
818 : // reach into the struct's fields directly.
819 :
820 : // Point returns the contained point iterator. If there is no point iterator,
821 : // Point returns a non-nil empty point iterator.
822 1 : func (s *iterSet) Point() internalIterator {
823 1 : if s.point == nil {
824 1 : return emptyIter
825 1 : }
826 1 : return s.point
827 : }
828 :
829 : // RangeDeletion returns the contained range deletion iterator. If there is no
830 : // range deletion iterator, RangeDeletion returns a non-nil empty keyspan
831 : // iterator.
832 1 : func (s *iterSet) RangeDeletion() keyspan.FragmentIterator {
833 1 : if s.rangeDeletion == nil {
834 1 : return emptyKeyspanIter
835 1 : }
836 1 : return s.rangeDeletion
837 : }
838 :
839 : // RangeKey returns the contained range key iterator. If there is no range key
840 : // iterator, RangeKey returns a non-nil empty keyspan iterator.
841 1 : func (s *iterSet) RangeKey() keyspan.FragmentIterator {
842 1 : if s.rangeKey == nil {
843 0 : return emptyKeyspanIter
844 0 : }
845 1 : return s.rangeKey
846 : }
847 :
848 : // CloseAll closes all of the held iterators. If CloseAll is called, then Close
849 : // must be not be called on the constituent iterators.
850 1 : func (s *iterSet) CloseAll() error {
851 1 : var err error
852 1 : if s.point != nil {
853 1 : err = s.point.Close()
854 1 : s.point = nil
855 1 : }
856 1 : if s.rangeDeletion != nil {
857 1 : s.rangeDeletion.Close()
858 1 : s.rangeDeletion = nil
859 1 : }
860 1 : if s.rangeKey != nil {
861 1 : s.rangeKey.Close()
862 1 : s.rangeKey = nil
863 1 : }
864 1 : return err
865 : }
866 :
867 : // iterKinds is a bitmap indicating a set of kinds of iterators. Callers may
868 : // bitwise-OR iterPointKeys, iterRangeDeletions and/or iterRangeKeys together to
869 : // represent a set of desired iterator kinds.
870 : type iterKinds uint8
871 :
872 1 : func (t iterKinds) Point() bool { return (t & iterPointKeys) != 0 }
873 1 : func (t iterKinds) RangeDeletion() bool { return (t & iterRangeDeletions) != 0 }
874 1 : func (t iterKinds) RangeKey() bool { return (t & iterRangeKeys) != 0 }
875 :
876 : const (
877 : iterPointKeys iterKinds = 1 << iota
878 : iterRangeDeletions
879 : iterRangeKeys
880 : )
|
