remove support for prefix replacement

Remove `PrefixReplacement` in favor of `SyntheticPrefix`. The general
prefix replacement turns out to be too difficult so we will instead
strip the prefix when writing out the backup files.

compaction.go

@@ -2444,14 +2444,14 @@ func (d *DB) runCopyCompaction(
 	// a new FileNum. This has the potential of making the block cache less
 	// effective, however.
 	newMeta := &fileMetadata{
-		Size:              inputMeta.Size,
-		CreationTime:      inputMeta.CreationTime,
-		SmallestSeqNum:    inputMeta.SmallestSeqNum,
-		LargestSeqNum:     inputMeta.LargestSeqNum,
-		Stats:             inputMeta.Stats,
-		PrefixReplacement: inputMeta.PrefixReplacement,
-		Virtual:           inputMeta.Virtual,
-		SyntheticSuffix:   inputMeta.SyntheticSuffix,
+		Size:            inputMeta.Size,
+		CreationTime:    inputMeta.CreationTime,
+		SmallestSeqNum:  inputMeta.SmallestSeqNum,
+		LargestSeqNum:   inputMeta.LargestSeqNum,
+		Stats:           inputMeta.Stats,
+		Virtual:         inputMeta.Virtual,
+		SyntheticPrefix: inputMeta.SyntheticPrefix,
+		SyntheticSuffix: inputMeta.SyntheticSuffix,
 	}
 	if inputMeta.HasPointKeys {
 		newMeta.ExtendPointKeyBounds(c.cmp, inputMeta.SmallestPointKey, inputMeta.LargestPointKey)

data_test.go

@@ -1294,7 +1294,7 @@ func runIngestExternalCmd(
 		usageErr := func(info interface{}) {
 			t.Helper()
 			td.Fatalf(t, "error parsing %q: %v; "+
-				"usage: obj bounds=(smallest,largest) [size=x] [prefix-replace=(from,to)] [synthetic-prefix=prefix] [synthetic-suffix=suffix]",
+				"usage: obj bounds=(smallest,largest) [size=x] [synthetic-prefix=prefix] [synthetic-suffix=suffix]",
 				line, info,
 			)
 		}
@@ -1328,11 +1328,6 @@ func runIngestExternalCmd(
 				nArgs(1)
 				arg.Scan(t, 0, &ef.Size)
 
-			case "prefix-replace":
-				nArgs(2)
-				ef.ContentPrefix = []byte(arg.Vals[0])
-				ef.SyntheticPrefix = []byte(arg.Vals[1])
-
 			case "synthetic-prefix":
 				nArgs(1)
 				ef.SyntheticPrefix = []byte(arg.Vals[0])

ingest.go

@@ -232,12 +232,7 @@ func ingestLoad1External(
 		)
 	}
 
-	if len(e.SyntheticPrefix) != 0 {
-		meta.PrefixReplacement = &sstable.PrefixReplacement{
-			ContentPrefix:   e.ContentPrefix,
-			SyntheticPrefix: e.SyntheticPrefix,
-		}
-	}
+	meta.SyntheticPrefix = e.SyntheticPrefix
 	meta.SyntheticSuffix = e.SyntheticSuffix
 
 	if err := meta.Validate(opts.Comparer.Compare, opts.Comparer.FormatKey); err != nil {
@@ -1153,15 +1148,11 @@ type ExternalFile struct {
 	// ingestion.
 	HasPointKey, HasRangeKey bool
 
-	// ContentPrefix and SyntheticPrefix denote a prefix replacement rule causing
-	// a file, in which all keys have prefix ContentPrefix, to appear whenever it
-	// is accessed as if those keys all instead have prefix SyntheticPrefix.
+	// SyntheticPrefix will prepend this suffix to all keys in the file during
+	// iteration. Note that the backing file itself is not modified.
 	//
 	// SyntheticPrefix must be a prefix of both Bounds.Start and Bounds.End.
-	//
-	// NB: If the SyntheticPrefix is non-empty and the ContentPrefix is empty,
-	// then the read path will conduct block level prefix synthesis.
-	ContentPrefix, SyntheticPrefix []byte
+	SyntheticPrefix []byte
 
 	// SyntheticSuffix will replace the suffix of every key in the file during
 	// iteration. Note that the file itself is not modified, rather, every key

internal/manifest/version.go

@@ -270,10 +270,11 @@ type FileMetadata struct {
 	// Virtual is true if the FileMetadata belongs to a virtual sstable.
 	Virtual bool
 
-	// PrefixReplacement is used for virtual files where the backing file has a
-	// different prefix on its keys than the span in which it is being exposed.
-	PrefixReplacement *sstable.PrefixReplacement
+	// SyntheticPrefix is used to prepend a prefix to all keys; used for some virtual
+	// tables.
+	SyntheticPrefix sstable.SyntheticPrefix
 
+	// SyntheticSuffix overrides all suffixes in a table; used for some virtual tables.
 	SyntheticSuffix sstable.SyntheticSuffix
 }
 
@@ -293,14 +294,10 @@ func (m *FileMetadata) SyntheticSeqNum() sstable.SyntheticSeqNum {
 
 // IterTransforms returns an sstable.IterTransforms that has SyntheticSeqNum set as needed.
 func (m *FileMetadata) IterTransforms() sstable.IterTransforms {
-	var syntheticPrefix []byte
-	if m.PrefixReplacement != nil && !m.PrefixReplacement.UsePrefixReplacementIterator() {
-		syntheticPrefix = m.PrefixReplacement.SyntheticPrefix
-	}
 	return sstable.IterTransforms{
 		SyntheticSeqNum: m.SyntheticSeqNum(),
 		SyntheticSuffix: m.SyntheticSuffix,
-		SyntheticPrefix: syntheticPrefix,
+		SyntheticPrefix: m.SyntheticPrefix,
 	}
 }
 
@@ -353,13 +350,12 @@ type VirtualFileMeta struct {
 // sstable reader.
 func (m VirtualFileMeta) VirtualReaderParams(isShared bool) sstable.VirtualReaderParams {
 	return sstable.VirtualReaderParams{
-		Lower:             m.Smallest,
-		Upper:             m.Largest,
-		FileNum:           m.FileNum,
-		IsSharedIngested:  isShared && m.SyntheticSeqNum() != 0,
-		Size:              m.Size,
-		BackingSize:       m.FileBacking.Size,
-		PrefixReplacement: m.PrefixReplacement,
+		Lower:            m.Smallest,
+		Upper:            m.Largest,
+		FileNum:          m.FileNum,
+		IsSharedIngested: isShared && m.SyntheticSeqNum() != 0,
+		Size:             m.Size,
+		BackingSize:      m.FileBacking.Size,
 	}
 }
 
@@ -865,14 +861,14 @@ func (m *FileMetadata) Validate(cmp Compare, formatKey base.FormatKey) error {
 		return base.CorruptionErrorf("file metadata FileBacking not set")
 	}
 
-	if m.PrefixReplacement != nil {
+	if m.SyntheticPrefix.IsSet() {
 		if !m.Virtual {
 			return base.CorruptionErrorf("prefix replacement rule set with non-virtual file")
 		}
-		if !bytes.HasPrefix(m.Smallest.UserKey, m.PrefixReplacement.SyntheticPrefix) {
+		if !bytes.HasPrefix(m.Smallest.UserKey, m.SyntheticPrefix) {
 			return base.CorruptionErrorf("virtual file with prefix replacement rules has smallest key with a different prefix: %s", m.Smallest.Pretty(formatKey))
 		}
-		if !bytes.HasPrefix(m.Largest.UserKey, m.PrefixReplacement.SyntheticPrefix) {
+		if !bytes.HasPrefix(m.Largest.UserKey, m.SyntheticPrefix) {
 			return base.CorruptionErrorf("virtual file with prefix replacement rules has largest key with a different prefix: %s", m.Largest.Pretty(formatKey))
 		}
 	}

internal/manifest/version_edit.go

@@ -342,8 +342,8 @@ func (v *VersionEdit) Decode(r io.Reader) error {
 				virtual        bool
 				backingFileNum uint64
 			}{}
-			var virtualPrefix *sstable.PrefixReplacement
-			var syntheticSuffix []byte
+			var syntheticPrefix sstable.SyntheticPrefix
+			var syntheticSuffix sstable.SyntheticSuffix
 			if tag == tagNewFile4 || tag == tagNewFile5 {
 				for {
 					customTag, err := d.readUvarint()
@@ -385,18 +385,19 @@ func (v *VersionEdit) Decode(r io.Reader) error {
 						}
 
 					case customTagPrefixRewrite:
+						// We used to have a content prefix; we no longer use it.
 						content, err := d.readBytes()
 						if err != nil {
 							return err
 						}
+						if len(content) > 0 {
+							return base.CorruptionErrorf("content prefix not supported")
+						}
 						synthetic, err := d.readBytes()
 						if err != nil {
 							return err
 						}
-						virtualPrefix = &sstable.PrefixReplacement{
-							ContentPrefix:   content,
-							SyntheticPrefix: synthetic,
-						}
+						syntheticPrefix = synthetic
 
 					case customTagSuffixRewrite:
 						if syntheticSuffix, err = d.readBytes(); err != nil {
@@ -421,7 +422,7 @@ func (v *VersionEdit) Decode(r io.Reader) error {
 				LargestSeqNum:       largestSeqNum,
 				MarkedForCompaction: markedForCompaction,
 				Virtual:             virtualState.virtual,
-				PrefixReplacement:   virtualPrefix,
+				SyntheticPrefix:     syntheticPrefix,
 				SyntheticSuffix:     syntheticSuffix,
 			}
 			if tag != tagNewFile5 { // no range keys present
@@ -705,10 +706,11 @@ func (v *VersionEdit) Encode(w io.Writer) error {
 				e.writeUvarint(customTagVirtual)
 				e.writeUvarint(uint64(x.Meta.FileBacking.DiskFileNum))
 			}
-			if x.Meta.PrefixReplacement != nil {
+			if x.Meta.SyntheticPrefix != nil {
 				e.writeUvarint(customTagPrefixRewrite)
-				e.writeBytes(x.Meta.PrefixReplacement.ContentPrefix)
-				e.writeBytes(x.Meta.PrefixReplacement.SyntheticPrefix)
+				// We used to have a content prefix; we no longer use it.
+				e.writeBytes(nil)
+				e.writeBytes(x.Meta.SyntheticPrefix)
 			}
 			if x.Meta.SyntheticSuffix != nil {
 				e.writeUvarint(customTagSuffixRewrite)

internal/manifest/version_edit_test.go

@@ -18,7 +18,6 @@ import (
 	"github.com/cockroachdb/errors"
 	"github.com/cockroachdb/pebble/internal/base"
 	"github.com/cockroachdb/pebble/record"
-	"github.com/cockroachdb/pebble/sstable"
 	"github.com/kr/pretty"
 	"github.com/stretchr/testify/require"
 )
@@ -45,15 +44,12 @@ func checkRoundTrip(e0 VersionEdit) error {
 func TestVERoundTripAndAccumulate(t *testing.T) {
 	cmp := base.DefaultComparer.Compare
 	m1 := (&FileMetadata{
-		FileNum:        810,
-		Size:           8090,
-		CreationTime:   809060,
-		SmallestSeqNum: 9,
-		LargestSeqNum:  11,
-		PrefixReplacement: &sstable.PrefixReplacement{
-			ContentPrefix:   []byte("before"),
-			SyntheticPrefix: []byte("after"),
-		},
+		FileNum:         810,
+		Size:            8090,
+		CreationTime:    809060,
+		SmallestSeqNum:  9,
+		LargestSeqNum:   11,
+		SyntheticPrefix: []byte("after"),
 		SyntheticSuffix: []byte("foo"),
 	}).ExtendPointKeyBounds(
 		cmp,

lsm_view.go

@@ -181,10 +181,11 @@ func (b *lsmViewBuilder) tableDetails(
 		outf("virtual; backed by %s (%ssize: %s)", m.FileBacking.DiskFileNum, backingInfo, humanize.Bytes.Uint64(m.FileBacking.Size))
 	}
 	outf("seqnums: %d - %d", m.SmallestSeqNum, m.LargestSeqNum)
-	if p := m.PrefixReplacement; p != nil {
-		outf("prefix replacement: %s -> %s", p.ContentPrefix, p.SyntheticPrefix)
+	if m.SyntheticPrefix.IsSet() {
+		// Note: we are abusing the key formatter by passing just the prefix.
+		outf("synthetic prefix: %s", b.fmtKey(m.SyntheticPrefix))
 	}
-	if len(m.SyntheticSuffix) > 0 {
+	if m.SyntheticSuffix.IsSet() {
 		// Note: we are abusing the key formatter by passing just the suffix.
 		outf("synthetic suffix: %s", b.fmtKey(m.SyntheticSuffix))
 	}

metamorphic/build.go

@@ -30,7 +30,7 @@ func writeSSTForIngestion(
 	rangeKeyIter keyspan.FragmentIterator,
 	uniquePrefixes bool,
 	syntheticSuffix sstable.SyntheticSuffix,
-	prefixChange *sstable.PrefixReplacement,
+	syntheticPrefix sstable.SyntheticPrefix,
 	writable objstorage.Writable,
 	targetFMV pebble.FormatMajorVersion,
 ) (*sstable.WriterMetadata, error) {
@@ -47,11 +47,11 @@ func writeSSTForIngestion(
 	defer rangeKeyIterCloser.Close()
 
 	outputKey := func(key []byte) []byte {
-		if prefixChange == nil && !syntheticSuffix.IsSet() {
+		if !syntheticPrefix.IsSet() && !syntheticSuffix.IsSet() {
 			return slices.Clone(key)
 		}
-		if prefixChange != nil {
-			key = prefixChange.Apply(key)
+		if syntheticPrefix.IsSet() {
+			key = syntheticPrefix.Apply(key)
 		}
 		if syntheticSuffix.IsSet() {
 			n := t.opts.Comparer.Split(key)
@@ -179,7 +179,7 @@ func buildForIngest(
 		iter, rangeDelIter, rangeKeyIter,
 		false, /* uniquePrefixes */
 		nil,   /* syntheticSuffix */
-		nil,   /* prefixChange */
+		nil,   /* syntheticPrefix */
 		writable,
 		db.FormatMajorVersion(),
 	)
@@ -195,14 +195,14 @@ func buildForIngestExternalEmulation(
 	externalObjID objID,
 	bounds pebble.KeyRange,
 	syntheticSuffix sstable.SyntheticSuffix,
-	prefixChange *sstable.PrefixReplacement,
+	syntheticPrefix sstable.SyntheticPrefix,
 	i int,
 ) (path string, _ *sstable.WriterMetadata) {
 	path = t.opts.FS.PathJoin(t.tmpDir, fmt.Sprintf("ext%d-%d", dbID.slot(), i))
 	f, err := t.opts.FS.Create(path)
 	panicIfErr(err)
 
-	reader, pointIter, rangeDelIter, rangeKeyIter := openExternalObj(t, externalObjID, bounds, prefixChange)
+	reader, pointIter, rangeDelIter, rangeKeyIter := openExternalObj(t, externalObjID, bounds, syntheticPrefix)
 	defer reader.Close()
 
 	writable := objstorageprovider.NewFileWritable(f)
@@ -215,7 +215,7 @@ func buildForIngestExternalEmulation(
 		pointIter, rangeDelIter, rangeKeyIter,
 		uniquePrefixes,
 		syntheticSuffix,
-		prefixChange,
+		syntheticPrefix,
 		writable,
 		t.minFMV(),
 	)
@@ -226,7 +226,7 @@ func buildForIngestExternalEmulation(
 }
 
 func openExternalObj(
-	t *Test, externalObjID objID, bounds pebble.KeyRange, prefixChange *sstable.PrefixReplacement,
+	t *Test, externalObjID objID, bounds pebble.KeyRange, syntheticPrefix sstable.SyntheticPrefix,
 ) (
 	reader *sstable.Reader,
 	pointIter base.InternalIterator,
@@ -243,9 +243,9 @@ func openExternalObj(
 
 	start := bounds.Start
 	end := bounds.End
-	if prefixChange != nil {
-		start = prefixChange.Invert(start)
-		end = prefixChange.Invert(end)
+	if syntheticPrefix.IsSet() {
+		start = syntheticPrefix.Invert(start)
+		end = syntheticPrefix.Invert(end)
 	}
 	pointIter, err = reader.NewIter(sstable.NoTransforms, start, end)
 	panicIfErr(err)
@@ -277,9 +277,9 @@ func openExternalObj(
 // externalObjIsEmpty returns true if the given external object has no point or
 // range keys withing the given bounds.
 func externalObjIsEmpty(
-	t *Test, externalObjID objID, bounds pebble.KeyRange, prefixChange *sstable.PrefixReplacement,
+	t *Test, externalObjID objID, bounds pebble.KeyRange, syntheticPrefix sstable.SyntheticPrefix,
 ) bool {
-	reader, pointIter, rangeDelIter, rangeKeyIter := openExternalObj(t, externalObjID, bounds, prefixChange)
+	reader, pointIter, rangeDelIter, rangeKeyIter := openExternalObj(t, externalObjID, bounds, syntheticPrefix)
 	defer reader.Close()
 	defer closeIters(pointIter, rangeDelIter, rangeKeyIter)
 

metamorphic/generator.go

@@ -1314,28 +1314,14 @@ func (g *generator) writerIngestExternalFiles() {
 		if g.cmp(start, end) == 0 {
 			end = objEnd
 		}
-		// Randomly set up prefix change.
-		var prefixChange *sstable.PrefixReplacement
+		// Randomly set up synthetic prefix.
+		var syntheticPrefix sstable.SyntheticPrefix
 		// We can only use a synthetic prefix if we don't have range dels.
 		// TODO(radu): we will want to support this at some point.
 		if !g.keyManager.objKeyMeta(id).hasRangeDels && g.rng.Intn(2) == 0 {
-			prefixChange = &sstable.PrefixReplacement{
-				SyntheticPrefix: randBytes(g.rng, 1, 5),
-			}
-			// TODO(radu): fix prefix replacement implementation or remove
-			// ContentPrefix altogether.
-			if false {
-				prefixLen := 0
-				limit := min(len(start), len(end))
-				for ; prefixLen < limit && start[prefixLen] == end[prefixLen]; prefixLen++ {
-				}
-				prefixLen = g.rng.Intn(prefixLen + 1)
-				if prefixLen > 0 {
-					prefixChange.ContentPrefix = start[:prefixLen]
-				}
-			}
-			start = prefixChange.Apply(start)
-			end = prefixChange.Apply(end)
+			syntheticPrefix = randBytes(g.rng, 1, 5)
+			start = syntheticPrefix.Apply(start)
+			end = syntheticPrefix.Apply(end)
 		}
 
 		objs[i] = externalObjWithBounds{
@@ -1344,7 +1330,7 @@ func (g *generator) writerIngestExternalFiles() {
 				Start: start,
 				End:   end,
 			},
-			prefixChange: prefixChange,
+			syntheticPrefix: syntheticPrefix,
 		}
 	}
 

metamorphic/key_manager.go

@@ -302,8 +302,8 @@ func (k *keyManager) KeysForExternalIngest(obj externalObjWithBounds) []keyMeta
 	var res []keyMeta
 	for _, km := range k.SortedKeysForObj(obj.externalObjID) {
 		// Apply prefix and suffix changes, then check the bounds.
-		if obj.prefixChange != nil {
-			km.key = obj.prefixChange.Apply(km.key)
+		if obj.syntheticPrefix.IsSet() {
+			km.key = obj.syntheticPrefix.Apply(km.key)
 		}
 		if obj.syntheticSuffix.IsSet() {
 			n := k.comparer.Split(km.key)