Rate Limit Scan Statistics

This pull request uses a `TokenBucket` to limit the number of keys that
read from `ScanStatistics` within a certain period of time.

Fixes: cockroachdb#2778

db.go

@@ -29,6 +29,7 @@ import (
 	"github.com/cockroachdb/pebble/sstable"
 	"github.com/cockroachdb/pebble/vfs"
 	"github.com/cockroachdb/pebble/vfs/atomicfs"
+	"github.com/cockroachdb/tokenbucket"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
@@ -1183,6 +1184,7 @@ func (d *DB) ScanInternal(
 	visitRangeKey func(start, end []byte, keys []rangekey.Key) error,
 	visitSharedFile func(sst *SharedSSTMeta) error,
 	includeObsoleteKeys bool,
+	rateLimitFunc func(key *InternalKey, val LazyValue),
 ) error {
 	scanInternalOpts := &scanInternalOptions{
 		visitPointKey:       visitPointKey,
@@ -1191,6 +1193,7 @@ func (d *DB) ScanInternal(
 		visitSharedFile:     visitSharedFile,
 		skipSharedLevels:    visitSharedFile != nil,
 		includeObsoleteKeys: includeObsoleteKeys,
+		rateLimitFunc:       rateLimitFunc,
 		IterOptions: IterOptions{
 			KeyTypes:   IterKeyTypePointsAndRanges,
 			LowerBound: lower,
@@ -2670,11 +2673,37 @@ type LSMKeyStatistics struct {
 	bytesRead   uint64
 }
 
+// ScanStatisticsOptions is used by DB.ScanStatistics.
+type ScanStatisticsOptions struct {
+	// LimitBytesPerSecond indicates the number of bytes that are able to be read
+	// per second using ScanInternal.
+	LimitBytesPerSecond int64
+}
+
 // ScanStatistics returns the count of different key kinds within the lsm for a
 // key span [lower, upper) as well as the number of snapshot keys.
-func (d *DB) ScanStatistics(ctx context.Context, lower, upper []byte) (LSMKeyStatistics, error) {
+func (d *DB) ScanStatistics(
+	ctx context.Context, lower, upper []byte, opts ScanStatisticsOptions,
+) (LSMKeyStatistics, error) {
 	stats := LSMKeyStatistics{}
 	var prevKey InternalKey
+	var rateLimitFunc func(key *InternalKey, val LazyValue)
+	tb := tokenbucket.TokenBucket{}
+
+	if opts.LimitBytesPerSecond != 0 {
+		// Each "token" roughly corresponds to a byte that was read.
+		tb.Init(tokenbucket.TokensPerSecond(opts.LimitBytesPerSecond), tokenbucket.Tokens(1024))
+		rateLimitFunc = func(key *InternalKey, val LazyValue) {
+			for {
+				fulfilled, tryAgainAfter := tb.TryToFulfill(tokenbucket.Tokens(key.Size() + val.Len()))
+
+				if fulfilled {
+					break
+				}
+				time.Sleep(tryAgainAfter)
+			}
+		}
+	}
 
 	err := d.ScanInternal(ctx, lower, upper,
 		func(key *InternalKey, value LazyValue, iterInfo IteratorLevel) error {
@@ -2712,6 +2741,7 @@ func (d *DB) ScanStatistics(ctx context.Context, lower, upper []byte) (LSMKeySta
 		},
 		nil,
 		true,
+		rateLimitFunc,
 	)
 
 	if err != nil {

ingest_test.go

@@ -932,7 +932,9 @@ func TestIngestShared(t *testing.T) {
 					sharedSSTs = append(sharedSSTs, *sst)
 					return nil
 				},
-				false)
+				false,
+				nil, /* rateLimitFunc */
+			)
 			require.NoError(t, err)
 			require.NoError(t, w.Close())
 

options.go

@@ -259,6 +259,9 @@ type scanInternalOptions struct {
 	// includeObsoleteKeys specifies whether keys shadowed by newer internal keys
 	// are exposed. If false, only one internal key per user key is exposed.
 	includeObsoleteKeys bool
+
+	// rateLimitFunc is used to limit the amount of bytes read per second.
+	rateLimitFunc func(key *InternalKey, value LazyValue)
 }
 
 // RangeKeyMasking configures automatic hiding of point keys by range keys. A

scan_internal.go

@@ -354,7 +354,6 @@ func (p *pointCollapsingIterator) String() string {
 
 var _ internalIterator = &pointCollapsingIterator{}
 
-<<<<<<< HEAD
 type iteratorLevelKind int8
 
 const (
@@ -374,14 +373,6 @@ type IteratorLevel struct {
 	Level int
 	// Sublevel is only valid if kind == iteratorLevelLSM and Level == 0.
 	Sublevel int
-=======
-// This is used with scanInternalIterator to surface additional iterator-specific info where possible.
-// Note: this is struct is only provided for point keys.
-type iterInfo struct {
-	// level indicates the level of point key called with visitPointKey (level == -1 indicates an invalid level).
-	// Note: level may be inaccurate if scanInternalOptions.includeObsoleteKeys is False.
-	level int
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 }
 
 // scanInternalIterator is an iterator that returns all internal keys, including
@@ -410,11 +401,7 @@ type scanInternalIterator struct {
 	newIters        tableNewIters
 	newIterRangeKey keyspan.TableNewSpanIter
 	seqNum          uint64
-<<<<<<< HEAD
 	iterLevels      []IteratorLevel
-=======
-	iterInfo        []iterInfo
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 	mergingIter     *mergingIter
 
 	// boundsBuf holds two buffers used to store the lower and upper bounds.
@@ -647,6 +634,11 @@ func scanInternalImpl(
 
 	for valid := iter.seekGE(lower); valid && iter.error() == nil; valid = iter.next() {
 		key := iter.unsafeKey()
+
+		if opts.rateLimitFunc != nil {
+			opts.rateLimitFunc(key, iter.lazyValue())
+		}
+
 		switch key.Kind() {
 		case InternalKeyKindRangeKeyDelete, InternalKeyKindRangeKeyUnset, InternalKeyKindRangeKeySet:
 			if opts.visitRangeKey != nil {
@@ -664,24 +656,12 @@ func scanInternalImpl(
 			}
 		default:
 			if opts.visitPointKey != nil {
-<<<<<<< HEAD
 				var info IteratorLevel
 				if len(iter.mergingIter.heap.items) > 0 {
 					mergingIterIdx := iter.mergingIter.heap.items[0].index
 					info = iter.iterLevels[mergingIterIdx]
 				} else {
 					info = IteratorLevel{kind: iteratorLevelUnknown}
-=======
-				var info iterInfo
-				if len(iter.mergingIter.heap.items) > 0 {
-					mergingIterIdx := iter.mergingIter.heap.items[0].index
-					info = iter.iterInfo[mergingIterIdx]
-				} else {
-					// Point key does not have a valid level (mergingIter heap is empty).
-					info = iterInfo{
-						level: -1,
-					}
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 				}
 				val := iter.lazyValue()
 				if err := opts.visitPointKey(key, val, info); err != nil {
@@ -733,11 +713,7 @@ func (i *scanInternalIterator) constructPointIter(memtables flushableList, buf *
 	rangeDelIters := make([]keyspan.FragmentIterator, 0, numMergingLevels)
 	rangeDelLevels := make([]keyspan.LevelIter, 0, numLevelIters)
 
-<<<<<<< HEAD
 	i.iterLevels = make([]IteratorLevel, numMergingLevels)
-=======
-	i.iterInfo = make([]iterInfo, numMergingLevels)
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 	mlevelsIndex := 0
 
 	// Next are the memtables.
@@ -746,14 +722,10 @@ func (i *scanInternalIterator) constructPointIter(memtables flushableList, buf *
 		mlevels = append(mlevels, mergingIterLevel{
 			iter: mem.newIter(&i.opts.IterOptions),
 		})
-<<<<<<< HEAD
 		i.iterLevels[mlevelsIndex] = IteratorLevel{
 			kind:           iteratorLevelFlushable,
 			FlushableIndex: j,
 		}
-=======
-		i.iterInfo[mlevelsIndex] = iterInfo{level: -1}
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 		mlevelsIndex++
 		if rdi := mem.newRangeDelIter(&i.opts.IterOptions); rdi != nil {
 			rangeDelIters = append(rangeDelIters, rdi)
@@ -785,15 +757,11 @@ func (i *scanInternalIterator) constructPointIter(memtables flushableList, buf *
 	}
 
 	for j := len(current.L0SublevelFiles) - 1; j >= 0; j-- {
-<<<<<<< HEAD
 		i.iterLevels[mlevelsIndex] = IteratorLevel{
 			kind:     iteratorLevelLSM,
 			Level:    0,
 			Sublevel: j,
 		}
-=======
-		i.iterInfo[mlevelsIndex] = iterInfo{level: 0}
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 		addLevelIterForFiles(current.L0SublevelFiles[j].Iter(), manifest.L0Sublevel(j))
 	}
 	// Add level iterators for the non-empty non-L0 levels.
@@ -804,11 +772,7 @@ func (i *scanInternalIterator) constructPointIter(memtables flushableList, buf *
 		if i.opts.skipSharedLevels && level >= sharedLevelsStart {
 			continue
 		}
-<<<<<<< HEAD
 		i.iterLevels[mlevelsIndex] = IteratorLevel{kind: iteratorLevelLSM, Level: level}
-=======
-		i.iterInfo[mlevelsIndex] = iterInfo{level: level}
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 		addLevelIterForFiles(current.Levels[level].Iter(), manifest.Level(level))
 	}
 

scan_internal_test.go

@@ -31,6 +31,7 @@ func TestScanStatistics(t *testing.T) {
 		ScanStatistics(
 			ctx context.Context,
 			lower, upper []byte,
+			opts ScanStatisticsOptions,
 		) (LSMKeyStatistics, error)
 	}
 	batches := map[string]*Batch{}
@@ -163,7 +164,7 @@ func TestScanStatistics(t *testing.T) {
 				default:
 				}
 			}
-			stats, err := reader.ScanStatistics(ctx, lower, upper)
+			stats, err := reader.ScanStatistics(ctx, lower, upper, ScanStatisticsOptions{})
 			if err != nil {
 				return err.Error()
 			}
@@ -208,6 +209,7 @@ func TestScanInternal(t *testing.T) {
 			visitRangeKey func(start, end []byte, keys []rangekey.Key) error,
 			visitSharedFile func(sst *SharedSSTMeta) error,
 			includeObsoleteKeys bool,
+			rateLimitFunc func(key *InternalKey, val LazyValue),
 		) error
 	}
 	batches := map[string]*Batch{}
@@ -413,11 +415,7 @@ func TestScanInternal(t *testing.T) {
 				}
 			}
 			err := reader.ScanInternal(context.TODO(), lower, upper,
-<<<<<<< HEAD
 				func(key *InternalKey, value LazyValue, _ IteratorLevel) error {
-=======
-				func(key *InternalKey, value LazyValue, _ iterInfo) error {
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 					v := value.InPlaceValue()
 					fmt.Fprintf(&b, "%s (%s)\n", key, v)
 					return nil
@@ -433,6 +431,7 @@ func TestScanInternal(t *testing.T) {
 				},
 				fileVisitor,
 				false,
+				nil, /* rateLimitFunc */
 			)
 			if err != nil {
 				return err.Error()

snapshot.go

@@ -71,6 +71,7 @@ func (s *Snapshot) ScanInternal(
 	visitRangeKey func(start, end []byte, keys []rangekey.Key) error,
 	visitSharedFile func(sst *SharedSSTMeta) error,
 	includeObsoleteKeys bool,
+	rateLimitFunc func(key *InternalKey, value LazyValue),
 ) error {
 	if s.db == nil {
 		panic(ErrClosed)
@@ -82,6 +83,7 @@ func (s *Snapshot) ScanInternal(
 		visitSharedFile:     visitSharedFile,
 		skipSharedLevels:    visitSharedFile != nil,
 		includeObsoleteKeys: includeObsoleteKeys,
+		rateLimitFunc:       rateLimitFunc,
 		IterOptions: IterOptions{
 			KeyTypes:   IterKeyTypePointsAndRanges,
 			LowerBound: lower,

testdata/scan_statistics

@@ -8,17 +8,10 @@ set e foo
 ----
 committed 2 keys
 
-<<<<<<< HEAD
 scan-statistics lower=b upper=f keys=(SET)
 ----
 Aggregate:
   SET key count: 2
-=======
-scan-statistics lower=b upper=e keys=(SET)
-----
-Aggregate:
-  SET key count: 0
->>>>>>> ba2f425f (pebble: Collect Pebble Key Statistics)
 
 flush
 ----