Improve issue search (#2387)

* Improve issue indexer

* Fix new issue sqlite bug

* Different test indexer paths for each db

* Add integration indexer paths to make clean
This commit is contained in:
Ethan Koenig 2017-09-16 13:16:21 -07:00 committed by Lauris BH
parent 52e11b24bf
commit b0f7457d9e
122 changed files with 15280 additions and 1458 deletions

View file

@ -48,6 +48,8 @@ type Index interface {
Advanced() (store.KVStore, error)
}
type DocumentFieldTermVisitor func(field string, term []byte)
type IndexReader interface {
TermFieldReader(term []byte, field string, includeFreq, includeNorm, includeTermVectors bool) (TermFieldReader, error)
@ -64,7 +66,7 @@ type IndexReader interface {
FieldDictPrefix(field string, termPrefix []byte) (FieldDict, error)
Document(id string) (*document.Document, error)
DocumentFieldTerms(id IndexInternalID, fields []string) (FieldTerms, error)
DocumentVisitFieldTerms(id IndexInternalID, fields []string, visitor DocumentFieldTermVisitor) error
Fields() ([]string, error)

View file

@ -90,7 +90,7 @@ func (udc *UpsideDownCouch) Analyze(d *document.Document) *index.AnalysisResult
rv.Rows = append(make([]index.IndexRow, 0, rowsCapNeeded), rv.Rows...)
backIndexTermEntries := make([]*BackIndexTermEntry, 0, rowsCapNeeded)
backIndexTermsEntries := make([]*BackIndexTermsEntry, 0, len(fieldTermFreqs))
// walk through the collated information and process
// once for each indexed field (unique name)
@ -99,11 +99,11 @@ func (udc *UpsideDownCouch) Analyze(d *document.Document) *index.AnalysisResult
includeTermVectors := fieldIncludeTermVectors[fieldIndex]
// encode this field
rv.Rows, backIndexTermEntries = udc.indexField(docIDBytes, includeTermVectors, fieldIndex, fieldLength, tokenFreqs, rv.Rows, backIndexTermEntries)
rv.Rows, backIndexTermsEntries = udc.indexField(docIDBytes, includeTermVectors, fieldIndex, fieldLength, tokenFreqs, rv.Rows, backIndexTermsEntries)
}
// build the back index row
backIndexRow := NewBackIndexRow(docIDBytes, backIndexTermEntries, backIndexStoredEntries)
backIndexRow := NewBackIndexRow(docIDBytes, backIndexTermsEntries, backIndexStoredEntries)
rv.Rows = append(rv.Rows, backIndexRow)
return rv

View file

@ -127,10 +127,12 @@ func (i *IndexReader) DumpDoc(id string) chan interface{} {
}
// build sorted list of term keys
keys := make(keyset, 0)
for _, entry := range back.termEntries {
tfr := NewTermFrequencyRow([]byte(*entry.Term), uint16(*entry.Field), idBytes, 0, 0)
key := tfr.Key()
keys = append(keys, key)
for _, entry := range back.termsEntries {
for i := range entry.Terms {
tfr := NewTermFrequencyRow([]byte(entry.Terms[i]), uint16(*entry.Field), idBytes, 0, 0)
key := tfr.Key()
keys = append(keys, key)
}
}
sort.Sort(keys)

View file

@ -101,15 +101,7 @@ func (i *IndexReader) Document(id string) (doc *document.Document, err error) {
return
}
func (i *IndexReader) DocumentFieldTerms(id index.IndexInternalID, fields []string) (index.FieldTerms, error) {
back, err := backIndexRowForDoc(i.kvreader, id)
if err != nil {
return nil, err
}
if back == nil {
return nil, nil
}
rv := make(index.FieldTerms, len(fields))
func (i *IndexReader) DocumentVisitFieldTerms(id index.IndexInternalID, fields []string, visitor index.DocumentFieldTermVisitor) error {
fieldsMap := make(map[uint16]string, len(fields))
for _, f := range fields {
id, ok := i.index.fieldCache.FieldNamed(f, false)
@ -117,12 +109,34 @@ func (i *IndexReader) DocumentFieldTerms(id index.IndexInternalID, fields []stri
fieldsMap[id] = f
}
}
for _, entry := range back.termEntries {
if field, ok := fieldsMap[uint16(*entry.Field)]; ok {
rv[field] = append(rv[field], *entry.Term)
}
tempRow := BackIndexRow{
doc: id,
}
return rv, nil
keyBuf := GetRowBuffer()
if tempRow.KeySize() > len(keyBuf) {
keyBuf = make([]byte, 2*tempRow.KeySize())
}
defer PutRowBuffer(keyBuf)
keySize, err := tempRow.KeyTo(keyBuf)
if err != nil {
return err
}
value, err := i.kvreader.Get(keyBuf[:keySize])
if err != nil {
return err
}
if value == nil {
return nil
}
return visitBackIndexRow(value, func(field uint32, term []byte) {
if field, ok := fieldsMap[uint16(field)]; ok {
visitor(field, term)
}
})
}
func (i *IndexReader) Fields() (fields []string, err error) {

View file

@ -24,46 +24,57 @@ import (
)
type UpsideDownCouchTermFieldReader struct {
count uint64
indexReader *IndexReader
iterator store.KVIterator
term []byte
tfrNext *TermFrequencyRow
keyBuf []byte
field uint16
count uint64
indexReader *IndexReader
iterator store.KVIterator
term []byte
tfrNext *TermFrequencyRow
tfrPrealloc TermFrequencyRow
keyBuf []byte
field uint16
includeTermVectors bool
}
func newUpsideDownCouchTermFieldReader(indexReader *IndexReader, term []byte, field uint16, includeFreq, includeNorm, includeTermVectors bool) (*UpsideDownCouchTermFieldReader, error) {
dictionaryRow := NewDictionaryRow(term, field, 0)
val, err := indexReader.kvreader.Get(dictionaryRow.Key())
bufNeeded := termFrequencyRowKeySize(term, nil)
if bufNeeded < dictionaryRowKeySize(term) {
bufNeeded = dictionaryRowKeySize(term)
}
buf := make([]byte, bufNeeded)
bufUsed := dictionaryRowKeyTo(buf, field, term)
val, err := indexReader.kvreader.Get(buf[:bufUsed])
if err != nil {
return nil, err
}
if val == nil {
atomic.AddUint64(&indexReader.index.stats.termSearchersStarted, uint64(1))
return &UpsideDownCouchTermFieldReader{
count: 0,
term: term,
tfrNext: &TermFrequencyRow{},
field: field,
}, nil
rv := &UpsideDownCouchTermFieldReader{
count: 0,
term: term,
field: field,
includeTermVectors: includeTermVectors,
}
rv.tfrNext = &rv.tfrPrealloc
return rv, nil
}
err = dictionaryRow.parseDictionaryV(val)
count, err := dictionaryRowParseV(val)
if err != nil {
return nil, err
}
tfr := NewTermFrequencyRow(term, field, []byte{}, 0, 0)
it := indexReader.kvreader.PrefixIterator(tfr.Key())
bufUsed = termFrequencyRowKeyTo(buf, field, term, nil)
it := indexReader.kvreader.PrefixIterator(buf[:bufUsed])
atomic.AddUint64(&indexReader.index.stats.termSearchersStarted, uint64(1))
return &UpsideDownCouchTermFieldReader{
indexReader: indexReader,
iterator: it,
count: dictionaryRow.count,
term: term,
field: field,
indexReader: indexReader,
iterator: it,
count: count,
term: term,
field: field,
includeTermVectors: includeTermVectors,
}, nil
}
@ -79,7 +90,7 @@ func (r *UpsideDownCouchTermFieldReader) Next(preAlloced *index.TermFieldDoc) (*
if r.tfrNext != nil {
r.iterator.Next()
} else {
r.tfrNext = &TermFrequencyRow{}
r.tfrNext = &r.tfrPrealloc
}
key, val, valid := r.iterator.Current()
if valid {
@ -88,7 +99,7 @@ func (r *UpsideDownCouchTermFieldReader) Next(preAlloced *index.TermFieldDoc) (*
if err != nil {
return nil, err
}
err = tfr.parseV(val)
err = tfr.parseV(val, r.includeTermVectors)
if err != nil {
return nil, err
}
@ -125,7 +136,7 @@ func (r *UpsideDownCouchTermFieldReader) Advance(docID index.IndexInternalID, pr
if err != nil {
return nil, err
}
err = tfr.parseV(val)
err = tfr.parseV(val, r.includeTermVectors)
if err != nil {
return nil, err
}

View file

@ -254,14 +254,22 @@ func (dr *DictionaryRow) Key() []byte {
}
func (dr *DictionaryRow) KeySize() int {
return len(dr.term) + 3
return dictionaryRowKeySize(dr.term)
}
func dictionaryRowKeySize(term []byte) int {
return len(term) + 3
}
func (dr *DictionaryRow) KeyTo(buf []byte) (int, error) {
return dictionaryRowKeyTo(buf, dr.field, dr.term), nil
}
func dictionaryRowKeyTo(buf []byte, field uint16, term []byte) int {
buf[0] = 'd'
binary.LittleEndian.PutUint16(buf[1:3], dr.field)
size := copy(buf[3:], dr.term)
return size + 3, nil
binary.LittleEndian.PutUint16(buf[1:3], field)
size := copy(buf[3:], term)
return size + 3
}
func (dr *DictionaryRow) Value() []byte {
@ -324,14 +332,22 @@ func (dr *DictionaryRow) parseDictionaryK(key []byte) error {
}
func (dr *DictionaryRow) parseDictionaryV(value []byte) error {
count, nread := binary.Uvarint(value)
if nread <= 0 {
return fmt.Errorf("DictionaryRow parse Uvarint error, nread: %d", nread)
count, err := dictionaryRowParseV(value)
if err != nil {
return err
}
dr.count = count
return nil
}
func dictionaryRowParseV(value []byte) (uint64, error) {
count, nread := binary.Uvarint(value)
if nread <= 0 {
return 0, fmt.Errorf("DictionaryRow parse Uvarint error, nread: %d", nread)
}
return count, nil
}
// TERM FIELD FREQUENCY
type TermVector struct {
@ -394,16 +410,24 @@ func (tfr *TermFrequencyRow) Key() []byte {
}
func (tfr *TermFrequencyRow) KeySize() int {
return 3 + len(tfr.term) + 1 + len(tfr.doc)
return termFrequencyRowKeySize(tfr.term, tfr.doc)
}
func termFrequencyRowKeySize(term, doc []byte) int {
return 3 + len(term) + 1 + len(doc)
}
func (tfr *TermFrequencyRow) KeyTo(buf []byte) (int, error) {
return termFrequencyRowKeyTo(buf, tfr.field, tfr.term, tfr.doc), nil
}
func termFrequencyRowKeyTo(buf []byte, field uint16, term, doc []byte) int {
buf[0] = 't'
binary.LittleEndian.PutUint16(buf[1:3], tfr.field)
termLen := copy(buf[3:], tfr.term)
binary.LittleEndian.PutUint16(buf[1:3], field)
termLen := copy(buf[3:], term)
buf[3+termLen] = ByteSeparator
docLen := copy(buf[3+termLen+1:], tfr.doc)
return 3 + termLen + 1 + docLen, nil
docLen := copy(buf[3+termLen+1:], doc)
return 3 + termLen + 1 + docLen
}
func (tfr *TermFrequencyRow) KeyAppendTo(buf []byte) ([]byte, error) {
@ -538,7 +562,7 @@ func (tfr *TermFrequencyRow) parseKDoc(key []byte, term []byte) error {
return nil
}
func (tfr *TermFrequencyRow) parseV(value []byte) error {
func (tfr *TermFrequencyRow) parseV(value []byte, includeTermVectors bool) error {
var bytesRead int
tfr.freq, bytesRead = binary.Uvarint(value)
if bytesRead <= 0 {
@ -556,6 +580,10 @@ func (tfr *TermFrequencyRow) parseV(value []byte) error {
tfr.norm = math.Float32frombits(uint32(norm))
tfr.vectors = nil
if !includeTermVectors {
return nil
}
var field uint64
field, bytesRead = binary.Uvarint(value[currOffset:])
for bytesRead > 0 {
@ -620,7 +648,7 @@ func NewTermFrequencyRowKV(key, value []byte) (*TermFrequencyRow, error) {
return nil, err
}
err = rv.parseV(value)
err = rv.parseV(value, true)
if err != nil {
return nil, err
}
@ -630,7 +658,7 @@ func NewTermFrequencyRowKV(key, value []byte) (*TermFrequencyRow, error) {
type BackIndexRow struct {
doc []byte
termEntries []*BackIndexTermEntry
termsEntries []*BackIndexTermsEntry
storedEntries []*BackIndexStoreEntry
}
@ -638,10 +666,12 @@ func (br *BackIndexRow) AllTermKeys() [][]byte {
if br == nil {
return nil
}
rv := make([][]byte, len(br.termEntries))
for i, termEntry := range br.termEntries {
termRow := NewTermFrequencyRow([]byte(termEntry.GetTerm()), uint16(termEntry.GetField()), br.doc, 0, 0)
rv[i] = termRow.Key()
rv := make([][]byte, 0, len(br.termsEntries)) // FIXME this underestimates severely
for _, termsEntry := range br.termsEntries {
for i := range termsEntry.Terms {
termRow := NewTermFrequencyRow([]byte(termsEntry.Terms[i]), uint16(termsEntry.GetField()), br.doc, 0, 0)
rv = append(rv, termRow.Key())
}
}
return rv
}
@ -682,7 +712,7 @@ func (br *BackIndexRow) Value() []byte {
func (br *BackIndexRow) ValueSize() int {
birv := &BackIndexRowValue{
TermEntries: br.termEntries,
TermsEntries: br.termsEntries,
StoredEntries: br.storedEntries,
}
return birv.Size()
@ -690,20 +720,20 @@ func (br *BackIndexRow) ValueSize() int {
func (br *BackIndexRow) ValueTo(buf []byte) (int, error) {
birv := &BackIndexRowValue{
TermEntries: br.termEntries,
TermsEntries: br.termsEntries,
StoredEntries: br.storedEntries,
}
return birv.MarshalTo(buf)
}
func (br *BackIndexRow) String() string {
return fmt.Sprintf("Backindex DocId: `%s` Term Entries: %v, Stored Entries: %v", string(br.doc), br.termEntries, br.storedEntries)
return fmt.Sprintf("Backindex DocId: `%s` Terms Entries: %v, Stored Entries: %v", string(br.doc), br.termsEntries, br.storedEntries)
}
func NewBackIndexRow(docID []byte, entries []*BackIndexTermEntry, storedFields []*BackIndexStoreEntry) *BackIndexRow {
func NewBackIndexRow(docID []byte, entries []*BackIndexTermsEntry, storedFields []*BackIndexStoreEntry) *BackIndexRow {
return &BackIndexRow{
doc: docID,
termEntries: entries,
termsEntries: entries,
storedEntries: storedFields,
}
}
@ -732,7 +762,7 @@ func NewBackIndexRowKV(key, value []byte) (*BackIndexRow, error) {
if err != nil {
return nil, err
}
rv.termEntries = birv.TermEntries
rv.termsEntries = birv.TermsEntries
rv.storedEntries = birv.StoredEntries
return &rv, nil
@ -851,3 +881,232 @@ func NewStoredRowKV(key, value []byte) (*StoredRow, error) {
rv.value = value[1:]
return rv, nil
}
type backIndexFieldTermVisitor func(field uint32, term []byte)
// visitBackIndexRow is designed to process a protobuf encoded
// value, without creating unnecessary garbage. Instead values are passed
// to a callback, inspected first, and only copied if necessary.
// Due to the fact that this borrows from generated code, it must be marnually
// updated if the protobuf definition changes.
//
// This code originates from:
// func (m *BackIndexRowValue) Unmarshal(data []byte) error
// the sections which create garbage or parse unintersting sections
// have been commented out. This was done by design to allow for easier
// merging in the future if that original function is regenerated
func visitBackIndexRow(data []byte, callback backIndexFieldTermVisitor) error {
l := len(data)
iNdEx := 0
for iNdEx < l {
var wire uint64
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
wire |= (uint64(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
fieldNum := int32(wire >> 3)
wireType := int(wire & 0x7)
switch fieldNum {
case 1:
if wireType != 2 {
return fmt.Errorf("proto: wrong wireType = %d for field TermsEntries", wireType)
}
var msglen int
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
msglen |= (int(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
postIndex := iNdEx + msglen
if msglen < 0 {
return ErrInvalidLengthUpsidedown
}
if postIndex > l {
return io.ErrUnexpectedEOF
}
// dont parse term entries
// m.TermsEntries = append(m.TermsEntries, &BackIndexTermsEntry{})
// if err := m.TermsEntries[len(m.TermsEntries)-1].Unmarshal(data[iNdEx:postIndex]); err != nil {
// return err
// }
// instead, inspect them
if err := visitBackIndexRowFieldTerms(data[iNdEx:postIndex], callback); err != nil {
return err
}
iNdEx = postIndex
case 2:
if wireType != 2 {
return fmt.Errorf("proto: wrong wireType = %d for field StoredEntries", wireType)
}
var msglen int
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
msglen |= (int(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
postIndex := iNdEx + msglen
if msglen < 0 {
return ErrInvalidLengthUpsidedown
}
if postIndex > l {
return io.ErrUnexpectedEOF
}
// don't parse stored entries
// m.StoredEntries = append(m.StoredEntries, &BackIndexStoreEntry{})
// if err := m.StoredEntries[len(m.StoredEntries)-1].Unmarshal(data[iNdEx:postIndex]); err != nil {
// return err
// }
iNdEx = postIndex
default:
var sizeOfWire int
for {
sizeOfWire++
wire >>= 7
if wire == 0 {
break
}
}
iNdEx -= sizeOfWire
skippy, err := skipUpsidedown(data[iNdEx:])
if err != nil {
return err
}
if skippy < 0 {
return ErrInvalidLengthUpsidedown
}
if (iNdEx + skippy) > l {
return io.ErrUnexpectedEOF
}
// don't track unrecognized data
//m.XXX_unrecognized = append(m.XXX_unrecognized, data[iNdEx:iNdEx+skippy]...)
iNdEx += skippy
}
}
return nil
}
// visitBackIndexRowFieldTerms is designed to process a protobuf encoded
// sub-value within the BackIndexRowValue, without creating unnecessary garbage.
// Instead values are passed to a callback, inspected first, and only copied if
// necessary. Due to the fact that this borrows from generated code, it must
// be marnually updated if the protobuf definition changes.
//
// This code originates from:
// func (m *BackIndexTermsEntry) Unmarshal(data []byte) error {
// the sections which create garbage or parse uninteresting sections
// have been commented out. This was done by design to allow for easier
// merging in the future if that original function is regenerated
func visitBackIndexRowFieldTerms(data []byte, callback backIndexFieldTermVisitor) error {
var theField uint32
var hasFields [1]uint64
l := len(data)
iNdEx := 0
for iNdEx < l {
var wire uint64
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
wire |= (uint64(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
fieldNum := int32(wire >> 3)
wireType := int(wire & 0x7)
switch fieldNum {
case 1:
if wireType != 0 {
return fmt.Errorf("proto: wrong wireType = %d for field Field", wireType)
}
var v uint32
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
v |= (uint32(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
// m.Field = &v
theField = v
hasFields[0] |= uint64(0x00000001)
case 2:
if wireType != 2 {
return fmt.Errorf("proto: wrong wireType = %d for field Terms", wireType)
}
var stringLen uint64
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
stringLen |= (uint64(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
postIndex := iNdEx + int(stringLen)
if postIndex > l {
return io.ErrUnexpectedEOF
}
//m.Terms = append(m.Terms, string(data[iNdEx:postIndex]))
callback(theField, data[iNdEx:postIndex])
iNdEx = postIndex
default:
var sizeOfWire int
for {
sizeOfWire++
wire >>= 7
if wire == 0 {
break
}
}
iNdEx -= sizeOfWire
skippy, err := skipUpsidedown(data[iNdEx:])
if err != nil {
return err
}
if skippy < 0 {
return ErrInvalidLengthUpsidedown
}
if (iNdEx + skippy) > l {
return io.ErrUnexpectedEOF
}
//m.XXX_unrecognized = append(m.XXX_unrecognized, data[iNdEx:iNdEx+skippy]...)
iNdEx += skippy
}
}
// if hasFields[0]&uint64(0x00000001) == 0 {
// return new(github_com_golang_protobuf_proto.RequiredNotSetError)
// }
return nil
}

View file

@ -45,7 +45,7 @@ const RowBufferSize = 4 * 1024
var VersionKey = []byte{'v'}
const Version uint8 = 5
const Version uint8 = 7
var IncompatibleVersion = fmt.Errorf("incompatible version, %d is supported", Version)
@ -499,44 +499,65 @@ func (udc *UpsideDownCouch) Update(doc *document.Document) (err error) {
func (udc *UpsideDownCouch) mergeOldAndNew(backIndexRow *BackIndexRow, rows []index.IndexRow) (addRows []UpsideDownCouchRow, updateRows []UpsideDownCouchRow, deleteRows []UpsideDownCouchRow) {
addRows = make([]UpsideDownCouchRow, 0, len(rows))
if backIndexRow == nil {
addRows = addRows[0:len(rows)]
for i, row := range rows {
addRows[i] = row
}
return addRows, nil, nil
}
updateRows = make([]UpsideDownCouchRow, 0, len(rows))
deleteRows = make([]UpsideDownCouchRow, 0, len(rows))
existingTermKeys := make(map[string]bool)
for _, key := range backIndexRow.AllTermKeys() {
existingTermKeys[string(key)] = true
var existingTermKeys map[string]struct{}
backIndexTermKeys := backIndexRow.AllTermKeys()
if len(backIndexTermKeys) > 0 {
existingTermKeys = make(map[string]struct{}, len(backIndexTermKeys))
for _, key := range backIndexTermKeys {
existingTermKeys[string(key)] = struct{}{}
}
}
existingStoredKeys := make(map[string]bool)
for _, key := range backIndexRow.AllStoredKeys() {
existingStoredKeys[string(key)] = true
var existingStoredKeys map[string]struct{}
backIndexStoredKeys := backIndexRow.AllStoredKeys()
if len(backIndexStoredKeys) > 0 {
existingStoredKeys = make(map[string]struct{}, len(backIndexStoredKeys))
for _, key := range backIndexStoredKeys {
existingStoredKeys[string(key)] = struct{}{}
}
}
keyBuf := GetRowBuffer()
for _, row := range rows {
switch row := row.(type) {
case *TermFrequencyRow:
if row.KeySize() > len(keyBuf) {
keyBuf = make([]byte, row.KeySize())
}
keySize, _ := row.KeyTo(keyBuf)
if _, ok := existingTermKeys[string(keyBuf[:keySize])]; ok {
updateRows = append(updateRows, row)
delete(existingTermKeys, string(keyBuf[:keySize]))
} else {
addRows = append(addRows, row)
if existingTermKeys != nil {
if row.KeySize() > len(keyBuf) {
keyBuf = make([]byte, row.KeySize())
}
keySize, _ := row.KeyTo(keyBuf)
if _, ok := existingTermKeys[string(keyBuf[:keySize])]; ok {
updateRows = append(updateRows, row)
delete(existingTermKeys, string(keyBuf[:keySize]))
continue
}
}
addRows = append(addRows, row)
case *StoredRow:
if row.KeySize() > len(keyBuf) {
keyBuf = make([]byte, row.KeySize())
}
keySize, _ := row.KeyTo(keyBuf)
if _, ok := existingStoredKeys[string(keyBuf[:keySize])]; ok {
updateRows = append(updateRows, row)
delete(existingStoredKeys, string(keyBuf[:keySize]))
} else {
addRows = append(addRows, row)
if existingStoredKeys != nil {
if row.KeySize() > len(keyBuf) {
keyBuf = make([]byte, row.KeySize())
}
keySize, _ := row.KeyTo(keyBuf)
if _, ok := existingStoredKeys[string(keyBuf[:keySize])]; ok {
updateRows = append(updateRows, row)
delete(existingStoredKeys, string(keyBuf[:keySize]))
continue
}
}
addRows = append(addRows, row)
default:
updateRows = append(updateRows, row)
}
@ -583,33 +604,41 @@ func encodeFieldType(f document.Field) byte {
fieldType = 'd'
case *document.BooleanField:
fieldType = 'b'
case *document.GeoPointField:
fieldType = 'g'
case *document.CompositeField:
fieldType = 'c'
}
return fieldType
}
func (udc *UpsideDownCouch) indexField(docID []byte, includeTermVectors bool, fieldIndex uint16, fieldLength int, tokenFreqs analysis.TokenFrequencies, rows []index.IndexRow, backIndexTermEntries []*BackIndexTermEntry) ([]index.IndexRow, []*BackIndexTermEntry) {
func (udc *UpsideDownCouch) indexField(docID []byte, includeTermVectors bool, fieldIndex uint16, fieldLength int, tokenFreqs analysis.TokenFrequencies, rows []index.IndexRow, backIndexTermsEntries []*BackIndexTermsEntry) ([]index.IndexRow, []*BackIndexTermsEntry) {
fieldNorm := float32(1.0 / math.Sqrt(float64(fieldLength)))
termFreqRows := make([]TermFrequencyRow, len(tokenFreqs))
termFreqRowsUsed := 0
terms := make([]string, 0, len(tokenFreqs))
for k, tf := range tokenFreqs {
var termFreqRow *TermFrequencyRow
termFreqRow := &termFreqRows[termFreqRowsUsed]
termFreqRowsUsed++
InitTermFrequencyRow(termFreqRow, tf.Term, fieldIndex, docID,
uint64(frequencyFromTokenFreq(tf)), fieldNorm)
if includeTermVectors {
var tv []*TermVector
tv, rows = udc.termVectorsFromTokenFreq(fieldIndex, tf, rows)
termFreqRow = NewTermFrequencyRowWithTermVectors(tf.Term, fieldIndex, docID, uint64(frequencyFromTokenFreq(tf)), fieldNorm, tv)
} else {
termFreqRow = NewTermFrequencyRow(tf.Term, fieldIndex, docID, uint64(frequencyFromTokenFreq(tf)), fieldNorm)
termFreqRow.vectors, rows = udc.termVectorsFromTokenFreq(fieldIndex, tf, rows)
}
// record the back index entry
backIndexTermEntry := BackIndexTermEntry{Term: proto.String(k), Field: proto.Uint32(uint32(fieldIndex))}
backIndexTermEntries = append(backIndexTermEntries, &backIndexTermEntry)
terms = append(terms, k)
rows = append(rows, termFreqRow)
}
backIndexTermsEntry := BackIndexTermsEntry{Field: proto.Uint32(uint32(fieldIndex)), Terms: terms}
backIndexTermsEntries = append(backIndexTermsEntries, &backIndexTermsEntry)
return rows, backIndexTermEntries
return rows, backIndexTermsEntries
}
func (udc *UpsideDownCouch) Delete(id string) (err error) {
@ -682,9 +711,11 @@ func (udc *UpsideDownCouch) Delete(id string) (err error) {
func (udc *UpsideDownCouch) deleteSingle(id string, backIndexRow *BackIndexRow, deleteRows []UpsideDownCouchRow) []UpsideDownCouchRow {
idBytes := []byte(id)
for _, backIndexEntry := range backIndexRow.termEntries {
tfr := NewTermFrequencyRow([]byte(*backIndexEntry.Term), uint16(*backIndexEntry.Field), idBytes, 0, 0)
deleteRows = append(deleteRows, tfr)
for _, backIndexEntry := range backIndexRow.termsEntries {
for i := range backIndexEntry.Terms {
tfr := NewTermFrequencyRow([]byte(backIndexEntry.Terms[i]), uint16(*backIndexEntry.Field), idBytes, 0, 0)
deleteRows = append(deleteRows, tfr)
}
}
for _, se := range backIndexRow.storedEntries {
sf := NewStoredRow(idBytes, uint16(*se.Field), se.ArrayPositions, 'x', nil)
@ -706,6 +737,8 @@ func decodeFieldType(typ byte, name string, pos []uint64, value []byte) document
return document.NewDateTimeFieldFromBytes(name, pos, value)
case 'b':
return document.NewBooleanFieldFromBytes(name, pos, value)
case 'g':
return document.NewGeoPointFieldFromBytes(name, pos, value)
}
return nil
}
@ -715,6 +748,7 @@ func frequencyFromTokenFreq(tf *analysis.TokenFreq) int {
}
func (udc *UpsideDownCouch) termVectorsFromTokenFreq(field uint16, tf *analysis.TokenFreq, rows []index.IndexRow) ([]*TermVector, []index.IndexRow) {
a := make([]TermVector, len(tf.Locations))
rv := make([]*TermVector, len(tf.Locations))
for i, l := range tf.Locations {
@ -727,14 +761,14 @@ func (udc *UpsideDownCouch) termVectorsFromTokenFreq(field uint16, tf *analysis.
rows = append(rows, newFieldRow)
}
}
tv := TermVector{
a[i] = TermVector{
field: fieldIndex,
arrayPositions: l.ArrayPositions,
pos: uint64(l.Position),
start: uint64(l.Start),
end: uint64(l.End),
}
rv[i] = &tv
rv[i] = &a[i]
}
return rv, rows
@ -745,18 +779,19 @@ func (udc *UpsideDownCouch) termFieldVectorsFromTermVectors(in []*TermVector) []
return nil
}
a := make([]index.TermFieldVector, len(in))
rv := make([]*index.TermFieldVector, len(in))
for i, tv := range in {
fieldName := udc.fieldCache.FieldIndexed(tv.field)
tfv := index.TermFieldVector{
a[i] = index.TermFieldVector{
Field: fieldName,
ArrayPositions: tv.arrayPositions,
Pos: tv.pos,
Start: tv.start,
End: tv.end,
}
rv[i] = &tfv
rv[i] = &a[i]
}
return rv
}
@ -1008,7 +1043,7 @@ func init() {
func backIndexRowForDoc(kvreader store.KVReader, docID index.IndexInternalID) (*BackIndexRow, error) {
// use a temporary row structure to build key
tempRow := &BackIndexRow{
tempRow := BackIndexRow{
doc: docID,
}

View file

@ -3,15 +3,15 @@
// DO NOT EDIT!
/*
Package upsidedown is a generated protocol buffer package.
Package upsidedown is a generated protocol buffer package.
It is generated from these files:
upsidedown.proto
It is generated from these files:
upsidedown.proto
It has these top-level messages:
BackIndexTermEntry
BackIndexStoreEntry
BackIndexRowValue
It has these top-level messages:
BackIndexTermsEntry
BackIndexStoreEntry
BackIndexRowValue
*/
package upsidedown
@ -26,30 +26,30 @@ import github_com_golang_protobuf_proto "github.com/golang/protobuf/proto"
var _ = proto.Marshal
var _ = math.Inf
type BackIndexTermEntry struct {
Term *string `protobuf:"bytes,1,req,name=term" json:"term,omitempty"`
Field *uint32 `protobuf:"varint,2,req,name=field" json:"field,omitempty"`
XXX_unrecognized []byte `json:"-"`
type BackIndexTermsEntry struct {
Field *uint32 `protobuf:"varint,1,req,name=field" json:"field,omitempty"`
Terms []string `protobuf:"bytes,2,rep,name=terms" json:"terms,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
func (m *BackIndexTermEntry) Reset() { *m = BackIndexTermEntry{} }
func (m *BackIndexTermEntry) String() string { return proto.CompactTextString(m) }
func (*BackIndexTermEntry) ProtoMessage() {}
func (m *BackIndexTermsEntry) Reset() { *m = BackIndexTermsEntry{} }
func (m *BackIndexTermsEntry) String() string { return proto.CompactTextString(m) }
func (*BackIndexTermsEntry) ProtoMessage() {}
func (m *BackIndexTermEntry) GetTerm() string {
if m != nil && m.Term != nil {
return *m.Term
}
return ""
}
func (m *BackIndexTermEntry) GetField() uint32 {
func (m *BackIndexTermsEntry) GetField() uint32 {
if m != nil && m.Field != nil {
return *m.Field
}
return 0
}
func (m *BackIndexTermsEntry) GetTerms() []string {
if m != nil {
return m.Terms
}
return nil
}
type BackIndexStoreEntry struct {
Field *uint32 `protobuf:"varint,1,req,name=field" json:"field,omitempty"`
ArrayPositions []uint64 `protobuf:"varint,2,rep,name=arrayPositions" json:"arrayPositions,omitempty"`
@ -75,7 +75,7 @@ func (m *BackIndexStoreEntry) GetArrayPositions() []uint64 {
}
type BackIndexRowValue struct {
TermEntries []*BackIndexTermEntry `protobuf:"bytes,1,rep,name=termEntries" json:"termEntries,omitempty"`
TermsEntries []*BackIndexTermsEntry `protobuf:"bytes,1,rep,name=termsEntries" json:"termsEntries,omitempty"`
StoredEntries []*BackIndexStoreEntry `protobuf:"bytes,2,rep,name=storedEntries" json:"storedEntries,omitempty"`
XXX_unrecognized []byte `json:"-"`
}
@ -84,9 +84,9 @@ func (m *BackIndexRowValue) Reset() { *m = BackIndexRowValue{} }
func (m *BackIndexRowValue) String() string { return proto.CompactTextString(m) }
func (*BackIndexRowValue) ProtoMessage() {}
func (m *BackIndexRowValue) GetTermEntries() []*BackIndexTermEntry {
func (m *BackIndexRowValue) GetTermsEntries() []*BackIndexTermsEntry {
if m != nil {
return m.TermEntries
return m.TermsEntries
}
return nil
}
@ -98,7 +98,7 @@ func (m *BackIndexRowValue) GetStoredEntries() []*BackIndexStoreEntry {
return nil
}
func (m *BackIndexTermEntry) Unmarshal(data []byte) error {
func (m *BackIndexTermsEntry) Unmarshal(data []byte) error {
var hasFields [1]uint64
l := len(data)
iNdEx := 0
@ -119,8 +119,26 @@ func (m *BackIndexTermEntry) Unmarshal(data []byte) error {
wireType := int(wire & 0x7)
switch fieldNum {
case 1:
if wireType != 0 {
return fmt.Errorf("proto: wrong wireType = %d for field Field", wireType)
}
var v uint32
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
v |= (uint32(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
m.Field = &v
hasFields[0] |= uint64(0x00000001)
case 2:
if wireType != 2 {
return fmt.Errorf("proto: wrong wireType = %d for field Term", wireType)
return fmt.Errorf("proto: wrong wireType = %d for field Terms", wireType)
}
var stringLen uint64
for shift := uint(0); ; shift += 7 {
@ -138,28 +156,8 @@ func (m *BackIndexTermEntry) Unmarshal(data []byte) error {
if postIndex > l {
return io.ErrUnexpectedEOF
}
s := string(data[iNdEx:postIndex])
m.Term = &s
m.Terms = append(m.Terms, string(data[iNdEx:postIndex]))
iNdEx = postIndex
hasFields[0] |= uint64(0x00000001)
case 2:
if wireType != 0 {
return fmt.Errorf("proto: wrong wireType = %d for field Field", wireType)
}
var v uint32
for shift := uint(0); ; shift += 7 {
if iNdEx >= l {
return io.ErrUnexpectedEOF
}
b := data[iNdEx]
iNdEx++
v |= (uint32(b) & 0x7F) << shift
if b < 0x80 {
break
}
}
m.Field = &v
hasFields[0] |= uint64(0x00000002)
default:
var sizeOfWire int
for {
@ -187,9 +185,6 @@ func (m *BackIndexTermEntry) Unmarshal(data []byte) error {
if hasFields[0]&uint64(0x00000001) == 0 {
return new(github_com_golang_protobuf_proto.RequiredNotSetError)
}
if hasFields[0]&uint64(0x00000002) == 0 {
return new(github_com_golang_protobuf_proto.RequiredNotSetError)
}
return nil
}
@ -299,7 +294,7 @@ func (m *BackIndexRowValue) Unmarshal(data []byte) error {
switch fieldNum {
case 1:
if wireType != 2 {
return fmt.Errorf("proto: wrong wireType = %d for field TermEntries", wireType)
return fmt.Errorf("proto: wrong wireType = %d for field TermsEntries", wireType)
}
var msglen int
for shift := uint(0); ; shift += 7 {
@ -320,8 +315,8 @@ func (m *BackIndexRowValue) Unmarshal(data []byte) error {
if postIndex > l {
return io.ErrUnexpectedEOF
}
m.TermEntries = append(m.TermEntries, &BackIndexTermEntry{})
if err := m.TermEntries[len(m.TermEntries)-1].Unmarshal(data[iNdEx:postIndex]); err != nil {
m.TermsEntries = append(m.TermsEntries, &BackIndexTermsEntry{})
if err := m.TermsEntries[len(m.TermsEntries)-1].Unmarshal(data[iNdEx:postIndex]); err != nil {
return err
}
iNdEx = postIndex
@ -472,16 +467,18 @@ var (
ErrInvalidLengthUpsidedown = fmt.Errorf("proto: negative length found during unmarshaling")
)
func (m *BackIndexTermEntry) Size() (n int) {
func (m *BackIndexTermsEntry) Size() (n int) {
var l int
_ = l
if m.Term != nil {
l = len(*m.Term)
n += 1 + l + sovUpsidedown(uint64(l))
}
if m.Field != nil {
n += 1 + sovUpsidedown(uint64(*m.Field))
}
if len(m.Terms) > 0 {
for _, s := range m.Terms {
l = len(s)
n += 1 + l + sovUpsidedown(uint64(l))
}
}
if m.XXX_unrecognized != nil {
n += len(m.XXX_unrecognized)
}
@ -508,8 +505,8 @@ func (m *BackIndexStoreEntry) Size() (n int) {
func (m *BackIndexRowValue) Size() (n int) {
var l int
_ = l
if len(m.TermEntries) > 0 {
for _, e := range m.TermEntries {
if len(m.TermsEntries) > 0 {
for _, e := range m.TermsEntries {
l = e.Size()
n += 1 + l + sovUpsidedown(uint64(l))
}
@ -539,7 +536,7 @@ func sovUpsidedown(x uint64) (n int) {
func sozUpsidedown(x uint64) (n int) {
return sovUpsidedown(uint64((x << 1) ^ uint64((int64(x) >> 63))))
}
func (m *BackIndexTermEntry) Marshal() (data []byte, err error) {
func (m *BackIndexTermsEntry) Marshal() (data []byte, err error) {
size := m.Size()
data = make([]byte, size)
n, err := m.MarshalTo(data)
@ -549,26 +546,33 @@ func (m *BackIndexTermEntry) Marshal() (data []byte, err error) {
return data[:n], nil
}
func (m *BackIndexTermEntry) MarshalTo(data []byte) (n int, err error) {
func (m *BackIndexTermsEntry) MarshalTo(data []byte) (n int, err error) {
var i int
_ = i
var l int
_ = l
if m.Term == nil {
return 0, new(github_com_golang_protobuf_proto.RequiredNotSetError)
} else {
data[i] = 0xa
i++
i = encodeVarintUpsidedown(data, i, uint64(len(*m.Term)))
i += copy(data[i:], *m.Term)
}
if m.Field == nil {
return 0, new(github_com_golang_protobuf_proto.RequiredNotSetError)
} else {
data[i] = 0x10
data[i] = 0x8
i++
i = encodeVarintUpsidedown(data, i, uint64(*m.Field))
}
if len(m.Terms) > 0 {
for _, s := range m.Terms {
data[i] = 0x12
i++
l = len(s)
for l >= 1<<7 {
data[i] = uint8(uint64(l)&0x7f | 0x80)
l >>= 7
i++
}
data[i] = uint8(l)
i++
i += copy(data[i:], s)
}
}
if m.XXX_unrecognized != nil {
i += copy(data[i:], m.XXX_unrecognized)
}
@ -625,8 +629,8 @@ func (m *BackIndexRowValue) MarshalTo(data []byte) (n int, err error) {
_ = i
var l int
_ = l
if len(m.TermEntries) > 0 {
for _, msg := range m.TermEntries {
if len(m.TermsEntries) > 0 {
for _, msg := range m.TermsEntries {
data[i] = 0xa
i++
i = encodeVarintUpsidedown(data, i, uint64(msg.Size()))

View file

@ -1,6 +1,6 @@
message BackIndexTermEntry {
required string term = 1;
required uint32 field = 2;
message BackIndexTermsEntry {
required uint32 field = 1;
repeated string terms = 2;
}
message BackIndexStoreEntry {
@ -9,6 +9,6 @@ message BackIndexStoreEntry {
}
message BackIndexRowValue {
repeated BackIndexTermEntry termEntries = 1;
repeated BackIndexTermsEntry termsEntries = 1;
repeated BackIndexStoreEntry storedEntries = 2;
}