Merge pull request #530 from steveyen/master

optimizations around search / DocumentMatchPool

index/upsidedown/reader.go

@@ -29,41 +29,49 @@ type UpsideDownCouchTermFieldReader struct {
 	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{
+		rv := &UpsideDownCouchTermFieldReader{
 			count:              0,
 			term:               term,
-			tfrNext:            &TermFrequencyRow{},
 			field:              field,
 			includeTermVectors: includeTermVectors,
-		}, nil
+		}
+		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,
+		count:              count,
 		term:               term,
 		field:              field,
 		includeTermVectors: includeTermVectors,
@@ -82,7 +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 {

index/upsidedown/row.go

@@ -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) {

search/pool.go

@@ -37,13 +37,17 @@ func defaultDocumentMatchPoolTooSmall(p *DocumentMatchPool) *DocumentMatch {
 // pre-allocated to accommodate the requested number of DocumentMatch
 // instances
 func NewDocumentMatchPool(size, sortsize int) *DocumentMatchPool {
-	avail := make(DocumentMatchCollection, 0, size)
+	avail := make(DocumentMatchCollection, size)
 	// pre-allocate the expected number of instances
 	startBlock := make([]DocumentMatch, size)
+	startSorts := make([]string, size*sortsize)
 	// make these initial instances available
-	for i := range startBlock {
-		startBlock[i].Sort = make([]string, 0, sortsize)
-		avail = append(avail, &startBlock[i])
+	i, j := 0, 0
+	for i < size {
+		avail[i] = &startBlock[i]
+		avail[i].Sort = startSorts[j:j]
+		i += 1
+		j += sortsize
 	}
 	return &DocumentMatchPool{
 		avail:    avail,