gibheer
/
bleve
0
0
Fork 0
Code

Merge pull request #815 from blevesearch/loadchunk_minor 

minor optimisation to loadChunk method
This commit is contained in:
Sreekanth Sivasankaran 2018-03-16 08:15:37 +05:30 committed by GitHub
parent d1b84d4578 d1155c223a
commit 23cebae5a8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
7 changed files with 279 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
index/scorch/segment/zap/build.go
View File

@ -22,7 +22,7 @@ import (
"github.com/Smerity/govarint"
)
const version uint32 = 4
const version uint32 = 5
const fieldNotUninverted = math.MaxUint64

8
index/scorch/segment/zap/contentcoder.go
View File

@ -156,9 +156,11 @@ func (c *chunkedContentCoder) Write(w io.Writer) (int, error) {
if err != nil {
return tw, err
}
// write out the chunk lens
for _, chunkLen := range c.chunkLens {
n := binary.PutUvarint(buf, uint64(chunkLen))
chunkOffsets := modifyLengthsToEndOffsets(c.chunkLens)
// write out the chunk offsets
for _, chunkOffset := range chunkOffsets {
n := binary.PutUvarint(buf, chunkOffset)
nw, err = w.Write(buf[:n])
tw += nw
if err != nil {

2
index/scorch/segment/zap/contentcoder_test.go
View File

@ -46,7 +46,7 @@ func TestChunkContentCoder(t *testing.T) {
[]byte("scorch"),
},
expected: string([]byte{0x02, 0x0c, 0x0c, 0x01, 0x00, 0x00, 0x06, 0x06, 0x14,
expected: string([]byte{0x02, 0x0c, 0x18, 0x01, 0x00, 0x00, 0x06, 0x06, 0x14,
0x75, 0x70, 0x73, 0x69, 0x64, 0x65, 0x01, 0x01, 0x00, 0x06, 0x06,
0x14, 0x73, 0x63, 0x6f, 0x72, 0x63, 0x68}),
},

29
index/scorch/segment/zap/docvalues.go
View File

@ -38,7 +38,7 @@ type docValueIterator struct {
field string
curChunkNum uint64
numChunks uint64
chunkLens []uint64
chunkOffsets []uint64
dvDataLoc uint64
curChunkHeader []MetaData
curChunkData []byte // compressed data cache
@ -47,7 +47,7 @@ type docValueIterator struct {
func (di *docValueIterator) size() int {
return reflectStaticSizedocValueIterator + size.SizeOfPtr +
len(di.field) +
len(di.chunkLens)*size.SizeOfUint64 +
len(di.chunkOffsets)*size.SizeOfUint64 +
len(di.curChunkHeader)*reflectStaticSizeMetaData +
len(di.curChunkData)
}
@ -69,7 +69,7 @@ func (s *SegmentBase) loadFieldDocValueIterator(field string,
}
// read the number of chunks, chunk lengths
var offset, clen uint64
var offset, loc uint64
numChunks, read := binary.Uvarint(s.mem[fieldDvLoc : fieldDvLoc+binary.MaxVarintLen64])
if read <= 0 {
return nil, fmt.Errorf("failed to read the field "+
@ -78,16 +78,16 @@ func (s *SegmentBase) loadFieldDocValueIterator(field string,
offset += uint64(read)
fdvIter := &docValueIterator{
curChunkNum: math.MaxUint64,
field: field,
chunkLens: make([]uint64, int(numChunks)),
curChunkNum: math.MaxUint64,
field: field,
chunkOffsets: make([]uint64, int(numChunks)),
}
for i := 0; i < int(numChunks); i++ {
clen, read = binary.Uvarint(s.mem[fieldDvLoc+offset : fieldDvLoc+offset+binary.MaxVarintLen64])
loc, read = binary.Uvarint(s.mem[fieldDvLoc+offset : fieldDvLoc+offset+binary.MaxVarintLen64])
if read <= 0 {
return nil, fmt.Errorf("corrupted chunk length during segment load")
return nil, fmt.Errorf("corrupted chunk offset during segment load")
}
fdvIter.chunkLens[i] = clen
fdvIter.chunkOffsets[i] = loc
offset += uint64(read)
}
@ -99,12 +99,11 @@ func (di *docValueIterator) loadDvChunk(chunkNumber,
localDocNum uint64, s *SegmentBase) error {
// advance to the chunk where the docValues
// reside for the given docNum
destChunkDataLoc := di.dvDataLoc
for i := 0; i < int(chunkNumber); i++ {
destChunkDataLoc += di.chunkLens[i]
}
destChunkDataLoc, curChunkEnd := di.dvDataLoc, di.dvDataLoc
start, end := readChunkBoundary(int(chunkNumber), di.chunkOffsets)
destChunkDataLoc += start
curChunkEnd += end
curChunkSize := di.chunkLens[chunkNumber]
// read the number of docs reside in the chunk
numDocs, read := binary.Uvarint(s.mem[destChunkDataLoc : destChunkDataLoc+binary.MaxVarintLen64])
if read <= 0 {
@ -124,7 +123,7 @@ func (di *docValueIterator) loadDvChunk(chunkNumber,
}
compressedDataLoc := chunkMetaLoc + offset
dataLength := destChunkDataLoc + curChunkSize - compressedDataLoc
dataLength := curChunkEnd - compressedDataLoc
di.curChunkData = s.mem[compressedDataLoc : compressedDataLoc+dataLength]
di.curChunkNum = chunkNumber
return nil

44
index/scorch/segment/zap/intcoder.go
View File

@ -111,10 +111,13 @@ func (c *chunkedIntCoder) Write(w io.Writer) (int, error) {
}
buf := c.buf
// write out the number of chunks & each chunkLen
n := binary.PutUvarint(buf, uint64(len(c.chunkLens)))
for _, chunkLen := range c.chunkLens {
n += binary.PutUvarint(buf[n:], uint64(chunkLen))
// convert the chunk lengths into chunk offsets
chunkOffsets := modifyLengthsToEndOffsets(c.chunkLens)
// write out the number of chunks & each chunk offsets
n := binary.PutUvarint(buf, uint64(len(chunkOffsets)))
for _, chunkOffset := range chunkOffsets {
n += binary.PutUvarint(buf[n:], chunkOffset)
}
tw, err := w.Write(buf[:n])
@ -134,3 +137,36 @@ func (c *chunkedIntCoder) Write(w io.Writer) (int, error) {
func (c *chunkedIntCoder) FinalSize() int {
return len(c.final)
}
// modifyLengthsToEndOffsets converts the chunk length array
// to a chunk offset array. The readChunkBoundary
// will figure out the start and end of every chunk from
// these offsets. Starting offset of i'th index is stored
// in i-1'th position except for 0'th index and ending offset
// is stored at i'th index position.
// For 0'th element, starting position is always zero.
// eg:
// Lens -> 5 5 5 5 => 5 10 15 20
// Lens -> 0 5 0 5 => 0 5 5 10
// Lens -> 0 0 0 5 => 0 0 0 5
// Lens -> 5 0 0 0 => 5 5 5 5
// Lens -> 0 5 0 0 => 0 5 5 5
// Lens -> 0 0 5 0 => 0 0 5 5
func modifyLengthsToEndOffsets(lengths []uint64) []uint64 {
var runningOffset uint64
var index, i int
for i = 1; i <= len(lengths); i++ {
runningOffset += lengths[i-1]
lengths[index] = runningOffset
index++
}
return lengths
}
func readChunkBoundary(chunk int, offsets []uint64) (uint64, uint64) {
var start uint64
if chunk > 0 {
start = offsets[chunk-1]
}
return start, offsets[chunk]
}

200
index/scorch/segment/zap/intcoder_test.go
View File

@ -46,8 +46,8 @@ func TestChunkIntCoder(t *testing.T) {
[]uint64{3},
[]uint64{7},
},
// 2 chunks, chunk-0 length 1, chunk-1 length 1, value 3, value 7
expected: []byte{0x2, 0x1, 0x1, 0x3, 0x7},
// 2 chunks, chunk-0 offset 1, chunk-1 offset 2, value 3, value 7
expected: []byte{0x2, 0x1, 0x2, 0x3, 0x7},
},
}
@ -71,3 +71,199 @@ func TestChunkIntCoder(t *testing.T) {
}
}
}
func TestChunkLengthToOffsets(t *testing.T) {
tests := []struct {
lengths []uint64
expectedOffsets []uint64
}{
{
lengths: []uint64{5, 5, 5, 5, 5},
expectedOffsets: []uint64{5, 10, 15, 20, 25},
},
{
lengths: []uint64{0, 5, 0, 5, 0},
expectedOffsets: []uint64{0, 5, 5, 10, 10},
},
{
lengths: []uint64{0, 0, 0, 0, 5},
expectedOffsets: []uint64{0, 0, 0, 0, 5},
},
{
lengths: []uint64{5, 0, 0, 0, 0},
expectedOffsets: []uint64{5, 5, 5, 5, 5},
},
{
lengths: []uint64{0, 5, 0, 0, 0},
expectedOffsets: []uint64{0, 5, 5, 5, 5},
},
{
lengths: []uint64{0, 0, 0, 5, 0},
expectedOffsets: []uint64{0, 0, 0, 5, 5},
},
{
lengths: []uint64{0, 0, 0, 5, 5},
expectedOffsets: []uint64{0, 0, 0, 5, 10},
},
{
lengths: []uint64{5, 5, 5, 0, 0},
expectedOffsets: []uint64{5, 10, 15, 15, 15},
},
{
lengths: []uint64{5},
expectedOffsets: []uint64{5},
},
{
lengths: []uint64{5, 5},
expectedOffsets: []uint64{5, 10},
},
}
for i, test := range tests {
modifyLengthsToEndOffsets(test.lengths)
if !reflect.DeepEqual(test.expectedOffsets, test.lengths) {
t.Errorf("Test: %d failed, got %+v, expected %+v", i, test.lengths, test.expectedOffsets)
}
}
}
func TestChunkReadBoundaryFromOffsets(t *testing.T) {
tests := []struct {
chunkNumber int
offsets []uint64
expectedStart uint64
expectedEnd uint64
}{
{
offsets: []uint64{5, 10, 15, 20, 25},
chunkNumber: 4,
expectedStart: 20,
expectedEnd: 25,
},
{
offsets: []uint64{5, 10, 15, 20, 25},
chunkNumber: 0,
expectedStart: 0,
expectedEnd: 5,
},
{
offsets: []uint64{5, 10, 15, 20, 25},
chunkNumber: 2,
expectedStart: 10,
expectedEnd: 15,
},
{
offsets: []uint64{0, 5, 5, 10, 10},
chunkNumber: 4,
expectedStart: 10,
expectedEnd: 10,
},
{
offsets: []uint64{0, 5, 5, 10, 10},
chunkNumber: 1,
expectedStart: 0,
expectedEnd: 5,
},
{
offsets: []uint64{5, 5, 5, 5, 5},
chunkNumber: 0,
expectedStart: 0,
expectedEnd: 5,
},
{
offsets: []uint64{5, 5, 5, 5, 5},
chunkNumber: 4,
expectedStart: 5,
expectedEnd: 5,
},
{
offsets: []uint64{5, 5, 5, 5, 5},
chunkNumber: 1,
expectedStart: 5,
expectedEnd: 5,
},
{
offsets: []uint64{0, 5, 5, 5, 5},
chunkNumber: 1,
expectedStart: 0,
expectedEnd: 5,
},
{
offsets: []uint64{0, 5, 5, 5, 5},
chunkNumber: 0,
expectedStart: 0,
expectedEnd: 0,
},
{
offsets: []uint64{0, 0, 0, 5, 5},
chunkNumber: 2,
expectedStart: 0,
expectedEnd: 0,
},
{
offsets: []uint64{0, 0, 0, 5, 5},
chunkNumber: 1,
expectedStart: 0,
expectedEnd: 0,
},
{
offsets: []uint64{0, 0, 0, 0, 5},
chunkNumber: 4,
expectedStart: 0,
expectedEnd: 5,
},
{
offsets: []uint64{0, 0, 0, 0, 5},
chunkNumber: 2,
expectedStart: 0,
expectedEnd: 0,
},
{
offsets: []uint64{5, 10, 15, 15, 15},
chunkNumber: 0,
expectedStart: 0,
expectedEnd: 5,
},
{
offsets: []uint64{5, 10, 15, 15, 15},
chunkNumber: 1,
expectedStart: 5,
expectedEnd: 10,
},
{
offsets: []uint64{5, 10, 15, 15, 15},
chunkNumber: 2,
expectedStart: 10,
expectedEnd: 15,
},
{
offsets: []uint64{5, 10, 15, 15, 15},
chunkNumber: 3,
expectedStart: 15,
expectedEnd: 15,
},
{
offsets: []uint64{5, 10, 15, 15, 15},
chunkNumber: 4,
expectedStart: 15,
expectedEnd: 15,
},
{
offsets: []uint64{5},
chunkNumber: 0,
expectedStart: 0,
expectedEnd: 5,
},
}
for i, test := range tests {
s, e := readChunkBoundary(test.chunkNumber, test.offsets)
if test.expectedStart != s || test.expectedEnd != e {
t.Errorf("Test: %d failed for chunkNumber: %d got start: %d end: %d,"+
" expected start: %d end: %d", i, test.chunkNumber, s, e,
test.expectedStart, test.expectedEnd)
}
}
}

44
index/scorch/segment/zap/posting.go
View File

@ -189,9 +189,9 @@ func (p *PostingsList) iterator(rv *PostingsIterator) *PostingsIterator {
var numFreqChunks uint64
numFreqChunks, read = binary.Uvarint(p.sb.mem[p.freqOffset+n : p.freqOffset+n+binary.MaxVarintLen64])
n += uint64(read)
rv.freqChunkLens = make([]uint64, int(numFreqChunks))
rv.freqChunkOffsets = make([]uint64, int(numFreqChunks))
for i := 0; i < int(numFreqChunks); i++ {
rv.freqChunkLens[i], read = binary.Uvarint(p.sb.mem[p.freqOffset+n : p.freqOffset+n+binary.MaxVarintLen64])
rv.freqChunkOffsets[i], read = binary.Uvarint(p.sb.mem[p.freqOffset+n : p.freqOffset+n+binary.MaxVarintLen64])
n += uint64(read)
}
rv.freqChunkStart = p.freqOffset + n
@ -201,9 +201,9 @@ func (p *PostingsList) iterator(rv *PostingsIterator) *PostingsIterator {
var numLocChunks uint64
numLocChunks, read = binary.Uvarint(p.sb.mem[p.locOffset+n : p.locOffset+n+binary.MaxVarintLen64])
n += uint64(read)
rv.locChunkLens = make([]uint64, int(numLocChunks))
rv.locChunkOffsets = make([]uint64, int(numLocChunks))
for i := 0; i < int(numLocChunks); i++ {
rv.locChunkLens[i], read = binary.Uvarint(p.sb.mem[p.locOffset+n : p.locOffset+n+binary.MaxVarintLen64])
rv.locChunkOffsets[i], read = binary.Uvarint(p.sb.mem[p.locOffset+n : p.locOffset+n+binary.MaxVarintLen64])
n += uint64(read)
}
rv.locChunkStart = p.locOffset + n
@ -316,11 +316,11 @@ type PostingsIterator struct {
locDecoder *govarint.Base128Decoder
locReader *bytes.Reader
freqChunkLens []uint64
freqChunkStart uint64
freqChunkOffsets []uint64
freqChunkStart uint64
locChunkLens []uint64
locChunkStart uint64
locChunkOffsets []uint64
locChunkStart uint64
locBitmap *roaring.Bitmap
@ -337,8 +337,8 @@ func (i *PostingsIterator) Size() int {
sizeInBytes := reflectStaticSizePostingsIterator + size.SizeOfPtr +
len(i.currChunkFreqNorm) +
len(i.currChunkLoc) +
len(i.freqChunkLens)*size.SizeOfUint64 +
len(i.locChunkLens)*size.SizeOfUint64 +
len(i.freqChunkOffsets)*size.SizeOfUint64 +
len(i.locChunkOffsets)*size.SizeOfUint64 +
i.next.Size()
if i.locBitmap != nil {
@ -353,16 +353,14 @@ func (i *PostingsIterator) Size() int {
}
func (i *PostingsIterator) loadChunk(chunk int) error {
if chunk >= len(i.freqChunkLens) || chunk >= len(i.locChunkLens) {
return fmt.Errorf("tried to load chunk that doesn't exist %d/(%d %d)", chunk, len(i.freqChunkLens), len(i.locChunkLens))
if chunk >= len(i.freqChunkOffsets) || chunk >= len(i.locChunkOffsets) {
return fmt.Errorf("tried to load chunk that doesn't exist %d/(%d %d)", chunk, len(i.freqChunkOffsets), len(i.locChunkOffsets))
}
// load freq chunk bytes
start := i.freqChunkStart
for j := 0; j < chunk; j++ {
start += i.freqChunkLens[j]
}
end := start + i.freqChunkLens[chunk]
end, start := i.freqChunkStart, i.freqChunkStart
s, e := readChunkBoundary(chunk, i.freqChunkOffsets)
start += s
end += e
i.currChunkFreqNorm = i.postings.sb.mem[start:end]
if i.freqNormReader == nil {
i.freqNormReader = bytes.NewReader(i.currChunkFreqNorm)
@ -371,12 +369,10 @@ func (i *PostingsIterator) loadChunk(chunk int) error {
i.freqNormReader.Reset(i.currChunkFreqNorm)
}
// load loc chunk bytes
start = i.locChunkStart
for j := 0; j < chunk; j++ {
start += i.locChunkLens[j]
}
end = start + i.locChunkLens[chunk]
end, start = i.locChunkStart, i.locChunkStart
s, e = readChunkBoundary(chunk, i.locChunkOffsets)
start += s
end += e
i.currChunkLoc = i.postings.sb.mem[start:end]
if i.locReader == nil {
i.locReader = bytes.NewReader(i.currChunkLoc)