0
0
Fork 0
bleve/search/sort.go

712 lines
17 KiB
Go
Raw Normal View History

// Copyright (c) 2014 Couchbase, Inc.
2016-10-02 16:13:14 +02:00
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package search
import (
"encoding/json"
"fmt"
"math"
"sort"
"strings"
"github.com/blevesearch/bleve/geo"
"github.com/blevesearch/bleve/numeric"
)
var HighTerm = strings.Repeat(string([]byte{0xff}), 10)
var LowTerm = string([]byte{0x00})
type SearchSort interface {
UpdateVisitor(field string, term []byte)
improved implementation to address perf regressions primary change is going back to sort values be []string and not []interface{}, this avoid allocatiosn converting into the interface{} that sounds obvious, so why didn't we just do that first? because a common (default) sort is score, which is naturally a number, not a string (like terms). converting into the number was also expensive, and the common case. so, this solution also makes the change to NOT put the score into the sort value list. instead you see the dummy value "_score". this is just a placeholder, the actual sort impl knows that field of the sort is the score, and will sort using the actual score. also, several other aspets of the benchmark were cleaned up so that unnecessary allocations do not pollute the cpu profiles Here are the updated benchmarks: $ go test -run=xxx -bench=. -benchmem -cpuprofile=cpu.out BenchmarkTop10of100000Scores-4 3000 465809 ns/op 2548 B/op 33 allocs/op BenchmarkTop100of100000Scores-4 2000 626488 ns/op 21484 B/op 213 allocs/op BenchmarkTop10of1000000Scores-4 300 5107658 ns/op 2560 B/op 33 allocs/op BenchmarkTop100of1000000Scores-4 300 5275403 ns/op 21624 B/op 213 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.188s Prior to this PR, master reported: $ go test -run=xxx -bench=. -benchmem BenchmarkTop10of100000Scores-4 3000 453269 ns/op 360161 B/op 42 allocs/op BenchmarkTop100of100000Scores-4 2000 519131 ns/op 388275 B/op 219 allocs/op BenchmarkTop10of1000000Scores-4 200 7459004 ns/op 4628236 B/op 52 allocs/op BenchmarkTop100of1000000Scores-4 200 8064864 ns/op 4656596 B/op 232 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.385s So, we're pretty close on the smaller datasets, and we scale better on the larger datasets. We also show fewer allocations and bytes in all cases (some of this is artificial due to test cleanup).
2016-08-25 21:47:07 +02:00
Value(a *DocumentMatch) string
Descending() bool
RequiresDocID() bool
RequiresScoring() bool
RequiresFields() []string
Copy() SearchSort
}
func ParseSearchSortObj(input map[string]interface{}) (SearchSort, error) {
descending, ok := input["desc"].(bool)
by, ok := input["by"].(string)
if !ok {
return nil, fmt.Errorf("search sort must specify by")
}
switch by {
case "id":
return &SortDocID{
Desc: descending,
}, nil
case "score":
return &SortScore{
Desc: descending,
}, nil
case "geo_distance":
field, ok := input["field"].(string)
if !ok {
return nil, fmt.Errorf("search sort mode geo_distance must specify field")
}
lon, lat, foundLocation := geo.ExtractGeoPoint(input["location"])
if !foundLocation {
return nil, fmt.Errorf("unable to parse geo_distance location")
}
rvd := &SortGeoDistance{
Field: field,
Desc: descending,
Lon: lon,
Lat: lat,
unitMult: 1.0,
}
if distUnit, ok := input["unit"].(string); ok {
var err error
rvd.unitMult, err = geo.ParseDistanceUnit(distUnit)
if err != nil {
return nil, err
}
rvd.Unit = distUnit
}
return rvd, nil
case "field":
field, ok := input["field"].(string)
if !ok {
return nil, fmt.Errorf("search sort mode field must specify field")
}
rv := &SortField{
Field: field,
Desc: descending,
}
typ, ok := input["type"].(string)
if ok {
switch typ {
case "auto":
rv.Type = SortFieldAuto
case "string":
rv.Type = SortFieldAsString
case "number":
rv.Type = SortFieldAsNumber
case "date":
rv.Type = SortFieldAsDate
default:
2016-10-02 18:11:15 +02:00
return nil, fmt.Errorf("unknown sort field type: %s", typ)
}
}
mode, ok := input["mode"].(string)
if ok {
switch mode {
case "default":
rv.Mode = SortFieldDefault
case "min":
rv.Mode = SortFieldMin
case "max":
rv.Mode = SortFieldMax
default:
return nil, fmt.Errorf("unknown sort field mode: %s", mode)
}
}
missing, ok := input["missing"].(string)
if ok {
switch missing {
case "first":
rv.Missing = SortFieldMissingFirst
case "last":
rv.Missing = SortFieldMissingLast
default:
return nil, fmt.Errorf("unknown sort field missing: %s", missing)
}
}
return rv, nil
}
return nil, fmt.Errorf("unknown search sort by: %s", by)
}
func ParseSearchSortString(input string) SearchSort {
descending := false
if strings.HasPrefix(input, "-") {
descending = true
input = input[1:]
} else if strings.HasPrefix(input, "+") {
input = input[1:]
}
if input == "_id" {
return &SortDocID{
Desc: descending,
}
} else if input == "_score" {
return &SortScore{
Desc: descending,
}
}
return &SortField{
Field: input,
Desc: descending,
}
}
func ParseSearchSortJSON(input json.RawMessage) (SearchSort, error) {
// first try to parse it as string
var sortString string
err := json.Unmarshal(input, &sortString)
if err != nil {
var sortObj map[string]interface{}
err = json.Unmarshal(input, &sortObj)
if err != nil {
return nil, err
}
return ParseSearchSortObj(sortObj)
}
return ParseSearchSortString(sortString), nil
}
func ParseSortOrderStrings(in []string) SortOrder {
rv := make(SortOrder, 0, len(in))
for _, i := range in {
ss := ParseSearchSortString(i)
rv = append(rv, ss)
}
return rv
}
func ParseSortOrderJSON(in []json.RawMessage) (SortOrder, error) {
rv := make(SortOrder, 0, len(in))
for _, i := range in {
ss, err := ParseSearchSortJSON(i)
if err != nil {
return nil, err
}
rv = append(rv, ss)
}
return rv, nil
}
type SortOrder []SearchSort
func (so SortOrder) Value(doc *DocumentMatch) {
for _, soi := range so {
doc.Sort = append(doc.Sort, soi.Value(doc))
}
}
func (so SortOrder) UpdateVisitor(field string, term []byte) {
for _, soi := range so {
soi.UpdateVisitor(field, term)
}
}
func (so SortOrder) Copy() SortOrder {
rv := make(SortOrder, len(so))
for i, soi := range so {
rv[i] = soi.Copy()
}
return rv
}
// Compare will compare two document matches using the specified sort order
// if both are numbers, we avoid converting back to term
func (so SortOrder) Compare(cachedScoring, cachedDesc []bool, i, j *DocumentMatch) int {
// compare the documents on all search sorts until a differences is found
for x := range so {
c := 0
if cachedScoring[x] {
improved implementation to address perf regressions primary change is going back to sort values be []string and not []interface{}, this avoid allocatiosn converting into the interface{} that sounds obvious, so why didn't we just do that first? because a common (default) sort is score, which is naturally a number, not a string (like terms). converting into the number was also expensive, and the common case. so, this solution also makes the change to NOT put the score into the sort value list. instead you see the dummy value "_score". this is just a placeholder, the actual sort impl knows that field of the sort is the score, and will sort using the actual score. also, several other aspets of the benchmark were cleaned up so that unnecessary allocations do not pollute the cpu profiles Here are the updated benchmarks: $ go test -run=xxx -bench=. -benchmem -cpuprofile=cpu.out BenchmarkTop10of100000Scores-4 3000 465809 ns/op 2548 B/op 33 allocs/op BenchmarkTop100of100000Scores-4 2000 626488 ns/op 21484 B/op 213 allocs/op BenchmarkTop10of1000000Scores-4 300 5107658 ns/op 2560 B/op 33 allocs/op BenchmarkTop100of1000000Scores-4 300 5275403 ns/op 21624 B/op 213 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.188s Prior to this PR, master reported: $ go test -run=xxx -bench=. -benchmem BenchmarkTop10of100000Scores-4 3000 453269 ns/op 360161 B/op 42 allocs/op BenchmarkTop100of100000Scores-4 2000 519131 ns/op 388275 B/op 219 allocs/op BenchmarkTop10of1000000Scores-4 200 7459004 ns/op 4628236 B/op 52 allocs/op BenchmarkTop100of1000000Scores-4 200 8064864 ns/op 4656596 B/op 232 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.385s So, we're pretty close on the smaller datasets, and we scale better on the larger datasets. We also show fewer allocations and bytes in all cases (some of this is artificial due to test cleanup).
2016-08-25 21:47:07 +02:00
if i.Score < j.Score {
c = -1
} else if i.Score > j.Score {
c = 1
}
improved implementation to address perf regressions primary change is going back to sort values be []string and not []interface{}, this avoid allocatiosn converting into the interface{} that sounds obvious, so why didn't we just do that first? because a common (default) sort is score, which is naturally a number, not a string (like terms). converting into the number was also expensive, and the common case. so, this solution also makes the change to NOT put the score into the sort value list. instead you see the dummy value "_score". this is just a placeholder, the actual sort impl knows that field of the sort is the score, and will sort using the actual score. also, several other aspets of the benchmark were cleaned up so that unnecessary allocations do not pollute the cpu profiles Here are the updated benchmarks: $ go test -run=xxx -bench=. -benchmem -cpuprofile=cpu.out BenchmarkTop10of100000Scores-4 3000 465809 ns/op 2548 B/op 33 allocs/op BenchmarkTop100of100000Scores-4 2000 626488 ns/op 21484 B/op 213 allocs/op BenchmarkTop10of1000000Scores-4 300 5107658 ns/op 2560 B/op 33 allocs/op BenchmarkTop100of1000000Scores-4 300 5275403 ns/op 21624 B/op 213 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.188s Prior to this PR, master reported: $ go test -run=xxx -bench=. -benchmem BenchmarkTop10of100000Scores-4 3000 453269 ns/op 360161 B/op 42 allocs/op BenchmarkTop100of100000Scores-4 2000 519131 ns/op 388275 B/op 219 allocs/op BenchmarkTop10of1000000Scores-4 200 7459004 ns/op 4628236 B/op 52 allocs/op BenchmarkTop100of1000000Scores-4 200 8064864 ns/op 4656596 B/op 232 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.385s So, we're pretty close on the smaller datasets, and we scale better on the larger datasets. We also show fewer allocations and bytes in all cases (some of this is artificial due to test cleanup).
2016-08-25 21:47:07 +02:00
} else {
iVal := i.Sort[x]
jVal := j.Sort[x]
c = strings.Compare(iVal, jVal)
}
if c == 0 {
continue
}
if cachedDesc[x] {
c = -c
}
return c
}
// if they are the same at this point, impose order based on index natural sort order
if i.HitNumber == j.HitNumber {
return 0
} else if i.HitNumber > j.HitNumber {
return 1
}
return -1
}
func (so SortOrder) RequiresScore() bool {
rv := false
for _, soi := range so {
if soi.RequiresScoring() {
rv = true
}
}
return rv
}
func (so SortOrder) RequiresDocID() bool {
rv := false
for _, soi := range so {
if soi.RequiresDocID() {
rv = true
}
}
return rv
}
func (so SortOrder) RequiredFields() []string {
var rv []string
for _, soi := range so {
rv = append(rv, soi.RequiresFields()...)
}
return rv
}
func (so SortOrder) CacheIsScore() []bool {
var rv []bool
for _, soi := range so {
rv = append(rv, soi.RequiresScoring())
}
return rv
}
func (so SortOrder) CacheDescending() []bool {
var rv []bool
for _, soi := range so {
rv = append(rv, soi.Descending())
}
return rv
}
// SortFieldType lets you control some internal sort behavior
// normally leaving this to the zero-value of SortFieldAuto is fine
type SortFieldType int
const (
// SortFieldAuto applies heuristics attempt to automatically sort correctly
SortFieldAuto SortFieldType = iota
// SortFieldAsString forces sort as string (no prefix coded terms removed)
SortFieldAsString
// SortFieldAsNumber forces sort as string (prefix coded terms with shift > 0 removed)
SortFieldAsNumber
// SortFieldAsDate forces sort as string (prefix coded terms with shift > 0 removed)
SortFieldAsDate
)
// SortFieldMode describes the behavior if the field has multiple values
type SortFieldMode int
const (
// SortFieldDefault uses the first (or only) value, this is the default zero-value
SortFieldDefault SortFieldMode = iota // FIXME name is confusing
// SortFieldMin uses the minimum value
SortFieldMin
// SortFieldMax uses the maximum value
SortFieldMax
)
// SortFieldMissing controls where documents missing a field value should be sorted
type SortFieldMissing int
const (
// SortFieldMissingLast sorts documents missing a field at the end
SortFieldMissingLast SortFieldMissing = iota
// SortFieldMissingFirst sorts documents missing a field at the beginning
SortFieldMissingFirst
)
// SortField will sort results by the value of a stored field
// Field is the name of the field
// Descending reverse the sort order (default false)
// Type allows forcing of string/number/date behavior (default auto)
// Mode controls behavior for multi-values fields (default first)
// Missing controls behavior of missing values (default last)
type SortField struct {
Field string
Desc bool
Type SortFieldType
Mode SortFieldMode
Missing SortFieldMissing
values []string
}
// UpdateVisitor notifies this sort field that in this document
// this field has the specified term
func (s *SortField) UpdateVisitor(field string, term []byte) {
if field == s.Field {
s.values = append(s.values, string(term))
}
}
// Value returns the sort value of the DocumentMatch
// it also resets the state of this SortField for
// processing the next document
improved implementation to address perf regressions primary change is going back to sort values be []string and not []interface{}, this avoid allocatiosn converting into the interface{} that sounds obvious, so why didn't we just do that first? because a common (default) sort is score, which is naturally a number, not a string (like terms). converting into the number was also expensive, and the common case. so, this solution also makes the change to NOT put the score into the sort value list. instead you see the dummy value "_score". this is just a placeholder, the actual sort impl knows that field of the sort is the score, and will sort using the actual score. also, several other aspets of the benchmark were cleaned up so that unnecessary allocations do not pollute the cpu profiles Here are the updated benchmarks: $ go test -run=xxx -bench=. -benchmem -cpuprofile=cpu.out BenchmarkTop10of100000Scores-4 3000 465809 ns/op 2548 B/op 33 allocs/op BenchmarkTop100of100000Scores-4 2000 626488 ns/op 21484 B/op 213 allocs/op BenchmarkTop10of1000000Scores-4 300 5107658 ns/op 2560 B/op 33 allocs/op BenchmarkTop100of1000000Scores-4 300 5275403 ns/op 21624 B/op 213 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.188s Prior to this PR, master reported: $ go test -run=xxx -bench=. -benchmem BenchmarkTop10of100000Scores-4 3000 453269 ns/op 360161 B/op 42 allocs/op BenchmarkTop100of100000Scores-4 2000 519131 ns/op 388275 B/op 219 allocs/op BenchmarkTop10of1000000Scores-4 200 7459004 ns/op 4628236 B/op 52 allocs/op BenchmarkTop100of1000000Scores-4 200 8064864 ns/op 4656596 B/op 232 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.385s So, we're pretty close on the smaller datasets, and we scale better on the larger datasets. We also show fewer allocations and bytes in all cases (some of this is artificial due to test cleanup).
2016-08-25 21:47:07 +02:00
func (s *SortField) Value(i *DocumentMatch) string {
iTerms := s.filterTermsByType(s.values)
iTerm := s.filterTermsByMode(iTerms)
s.values = nil
return iTerm
}
// Descending determines the order of the sort
func (s *SortField) Descending() bool {
return s.Desc
}
func (s *SortField) filterTermsByMode(terms []string) string {
if len(terms) == 1 || (len(terms) > 1 && s.Mode == SortFieldDefault) {
return terms[0]
} else if len(terms) > 1 {
switch s.Mode {
case SortFieldMin:
sort.Strings(terms)
return terms[0]
case SortFieldMax:
sort.Strings(terms)
return terms[len(terms)-1]
}
}
// handle missing terms
if s.Missing == SortFieldMissingLast {
if s.Desc {
return LowTerm
}
return HighTerm
}
if s.Desc {
return HighTerm
}
return LowTerm
}
// filterTermsByType attempts to make one pass on the terms
// if we are in auto-mode AND all the terms look like prefix-coded numbers
// return only the terms which had shift of 0
// if we are in explicit number or date mode, return only valid
// prefix coded numbers with shift of 0
func (s *SortField) filterTermsByType(terms []string) []string {
stype := s.Type
if stype == SortFieldAuto {
allTermsPrefixCoded := true
var termsWithShiftZero []string
for _, term := range terms {
valid, shift := numeric.ValidPrefixCodedTerm(term)
if valid && shift == 0 {
termsWithShiftZero = append(termsWithShiftZero, term)
} else if !valid {
allTermsPrefixCoded = false
}
}
if allTermsPrefixCoded {
terms = termsWithShiftZero
}
} else if stype == SortFieldAsNumber || stype == SortFieldAsDate {
var termsWithShiftZero []string
for _, term := range terms {
valid, shift := numeric.ValidPrefixCodedTerm(term)
if valid && shift == 0 {
termsWithShiftZero = append(termsWithShiftZero, term)
}
}
terms = termsWithShiftZero
}
return terms
}
// RequiresDocID says this SearchSort does not require the DocID be loaded
func (s *SortField) RequiresDocID() bool { return false }
// RequiresScoring says this SearchStore does not require scoring
func (s *SortField) RequiresScoring() bool { return false }
// RequiresFields says this SearchStore requires the specified stored field
func (s *SortField) RequiresFields() []string { return []string{s.Field} }
func (s *SortField) MarshalJSON() ([]byte, error) {
// see if simple format can be used
if s.Missing == SortFieldMissingLast &&
s.Mode == SortFieldDefault &&
s.Type == SortFieldAuto {
if s.Desc {
return json.Marshal("-" + s.Field)
}
return json.Marshal(s.Field)
}
sfm := map[string]interface{}{
"by": "field",
"field": s.Field,
}
if s.Desc {
sfm["desc"] = true
}
if s.Missing > SortFieldMissingLast {
switch s.Missing {
case SortFieldMissingFirst:
sfm["missing"] = "first"
}
}
if s.Mode > SortFieldDefault {
switch s.Mode {
case SortFieldMin:
sfm["mode"] = "min"
case SortFieldMax:
sfm["mode"] = "max"
}
}
if s.Type > SortFieldAuto {
switch s.Type {
case SortFieldAsString:
sfm["type"] = "string"
case SortFieldAsNumber:
sfm["type"] = "number"
case SortFieldAsDate:
sfm["type"] = "date"
}
}
return json.Marshal(sfm)
}
func (s *SortField) Copy() SearchSort {
var rv SortField
rv = *s
return &rv
}
// SortDocID will sort results by the document identifier
type SortDocID struct {
Desc bool
}
// UpdateVisitor is a no-op for SortDocID as it's value
// is not dependent on any field terms
func (s *SortDocID) UpdateVisitor(field string, term []byte) {
}
// Value returns the sort value of the DocumentMatch
improved implementation to address perf regressions primary change is going back to sort values be []string and not []interface{}, this avoid allocatiosn converting into the interface{} that sounds obvious, so why didn't we just do that first? because a common (default) sort is score, which is naturally a number, not a string (like terms). converting into the number was also expensive, and the common case. so, this solution also makes the change to NOT put the score into the sort value list. instead you see the dummy value "_score". this is just a placeholder, the actual sort impl knows that field of the sort is the score, and will sort using the actual score. also, several other aspets of the benchmark were cleaned up so that unnecessary allocations do not pollute the cpu profiles Here are the updated benchmarks: $ go test -run=xxx -bench=. -benchmem -cpuprofile=cpu.out BenchmarkTop10of100000Scores-4 3000 465809 ns/op 2548 B/op 33 allocs/op BenchmarkTop100of100000Scores-4 2000 626488 ns/op 21484 B/op 213 allocs/op BenchmarkTop10of1000000Scores-4 300 5107658 ns/op 2560 B/op 33 allocs/op BenchmarkTop100of1000000Scores-4 300 5275403 ns/op 21624 B/op 213 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.188s Prior to this PR, master reported: $ go test -run=xxx -bench=. -benchmem BenchmarkTop10of100000Scores-4 3000 453269 ns/op 360161 B/op 42 allocs/op BenchmarkTop100of100000Scores-4 2000 519131 ns/op 388275 B/op 219 allocs/op BenchmarkTop10of1000000Scores-4 200 7459004 ns/op 4628236 B/op 52 allocs/op BenchmarkTop100of1000000Scores-4 200 8064864 ns/op 4656596 B/op 232 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.385s So, we're pretty close on the smaller datasets, and we scale better on the larger datasets. We also show fewer allocations and bytes in all cases (some of this is artificial due to test cleanup).
2016-08-25 21:47:07 +02:00
func (s *SortDocID) Value(i *DocumentMatch) string {
return i.ID
}
// Descending determines the order of the sort
func (s *SortDocID) Descending() bool {
return s.Desc
}
// RequiresDocID says this SearchSort does require the DocID be loaded
func (s *SortDocID) RequiresDocID() bool { return true }
// RequiresScoring says this SearchStore does not require scoring
func (s *SortDocID) RequiresScoring() bool { return false }
// RequiresFields says this SearchStore does not require any stored fields
func (s *SortDocID) RequiresFields() []string { return nil }
func (s *SortDocID) MarshalJSON() ([]byte, error) {
if s.Desc {
return json.Marshal("-_id")
}
return json.Marshal("_id")
}
func (s *SortDocID) Copy() SearchSort {
var rv SortDocID
rv = *s
return &rv
}
// SortScore will sort results by the document match score
type SortScore struct {
Desc bool
}
// UpdateVisitor is a no-op for SortScore as it's value
// is not dependent on any field terms
func (s *SortScore) UpdateVisitor(field string, term []byte) {
}
// Value returns the sort value of the DocumentMatch
improved implementation to address perf regressions primary change is going back to sort values be []string and not []interface{}, this avoid allocatiosn converting into the interface{} that sounds obvious, so why didn't we just do that first? because a common (default) sort is score, which is naturally a number, not a string (like terms). converting into the number was also expensive, and the common case. so, this solution also makes the change to NOT put the score into the sort value list. instead you see the dummy value "_score". this is just a placeholder, the actual sort impl knows that field of the sort is the score, and will sort using the actual score. also, several other aspets of the benchmark were cleaned up so that unnecessary allocations do not pollute the cpu profiles Here are the updated benchmarks: $ go test -run=xxx -bench=. -benchmem -cpuprofile=cpu.out BenchmarkTop10of100000Scores-4 3000 465809 ns/op 2548 B/op 33 allocs/op BenchmarkTop100of100000Scores-4 2000 626488 ns/op 21484 B/op 213 allocs/op BenchmarkTop10of1000000Scores-4 300 5107658 ns/op 2560 B/op 33 allocs/op BenchmarkTop100of1000000Scores-4 300 5275403 ns/op 21624 B/op 213 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.188s Prior to this PR, master reported: $ go test -run=xxx -bench=. -benchmem BenchmarkTop10of100000Scores-4 3000 453269 ns/op 360161 B/op 42 allocs/op BenchmarkTop100of100000Scores-4 2000 519131 ns/op 388275 B/op 219 allocs/op BenchmarkTop10of1000000Scores-4 200 7459004 ns/op 4628236 B/op 52 allocs/op BenchmarkTop100of1000000Scores-4 200 8064864 ns/op 4656596 B/op 232 allocs/op PASS ok github.com/blevesearch/bleve/search/collectors 7.385s So, we're pretty close on the smaller datasets, and we scale better on the larger datasets. We also show fewer allocations and bytes in all cases (some of this is artificial due to test cleanup).
2016-08-25 21:47:07 +02:00
func (s *SortScore) Value(i *DocumentMatch) string {
return "_score"
}
// Descending determines the order of the sort
func (s *SortScore) Descending() bool {
return s.Desc
}
// RequiresDocID says this SearchSort does not require the DocID be loaded
func (s *SortScore) RequiresDocID() bool { return false }
// RequiresScoring says this SearchStore does require scoring
func (s *SortScore) RequiresScoring() bool { return true }
// RequiresFields says this SearchStore does not require any store fields
func (s *SortScore) RequiresFields() []string { return nil }
func (s *SortScore) MarshalJSON() ([]byte, error) {
if s.Desc {
return json.Marshal("-_score")
}
return json.Marshal("_score")
}
func (s *SortScore) Copy() SearchSort {
var rv SortScore
rv = *s
return &rv
}
var maxDistance = string(numeric.MustNewPrefixCodedInt64(math.MaxInt64, 0))
// NewSortGeoDistance creates SearchSort instance for sorting documents by
// their distance from the specified point.
func NewSortGeoDistance(field, unit string, lon, lat float64, desc bool) (
*SortGeoDistance, error) {
rv := &SortGeoDistance{
Field: field,
Desc: desc,
Unit: unit,
Lon: lon,
Lat: lat,
}
var err error
rv.unitMult, err = geo.ParseDistanceUnit(unit)
if err != nil {
return nil, err
}
return rv, nil
}
// SortGeoDistance will sort results by the distance of an
// indexed geo point, from the provided location.
// Field is the name of the field
// Descending reverse the sort order (default false)
type SortGeoDistance struct {
Field string
Desc bool
Unit string
values []string
Lon float64
Lat float64
unitMult float64
}
// UpdateVisitor notifies this sort field that in this document
// this field has the specified term
func (s *SortGeoDistance) UpdateVisitor(field string, term []byte) {
if field == s.Field {
s.values = append(s.values, string(term))
}
}
// Value returns the sort value of the DocumentMatch
// it also resets the state of this SortField for
// processing the next document
func (s *SortGeoDistance) Value(i *DocumentMatch) string {
iTerms := s.filterTermsByType(s.values)
iTerm := s.filterTermsByMode(iTerms)
s.values = nil
if iTerm == "" {
return maxDistance
}
i64, err := numeric.PrefixCoded(iTerm).Int64()
if err != nil {
return maxDistance
}
docLon := geo.MortonUnhashLon(uint64(i64))
docLat := geo.MortonUnhashLat(uint64(i64))
dist := geo.Haversin(s.Lon, s.Lat, docLon, docLat)
// dist is returned in km, so convert to m
dist *= 1000
if s.unitMult != 0 {
dist /= s.unitMult
}
distInt64 := numeric.Float64ToInt64(dist)
return string(numeric.MustNewPrefixCodedInt64(distInt64, 0))
}
// Descending determines the order of the sort
func (s *SortGeoDistance) Descending() bool {
return s.Desc
}
func (s *SortGeoDistance) filterTermsByMode(terms []string) string {
if len(terms) >= 1 {
return terms[0]
}
return ""
}
// filterTermsByType attempts to make one pass on the terms
// return only valid prefix coded numbers with shift of 0
func (s *SortGeoDistance) filterTermsByType(terms []string) []string {
var termsWithShiftZero []string
for _, term := range terms {
valid, shift := numeric.ValidPrefixCodedTerm(term)
if valid && shift == 0 {
termsWithShiftZero = append(termsWithShiftZero, term)
}
}
return termsWithShiftZero
}
// RequiresDocID says this SearchSort does not require the DocID be loaded
func (s *SortGeoDistance) RequiresDocID() bool { return false }
// RequiresScoring says this SearchStore does not require scoring
func (s *SortGeoDistance) RequiresScoring() bool { return false }
// RequiresFields says this SearchStore requires the specified stored field
func (s *SortGeoDistance) RequiresFields() []string { return []string{s.Field} }
func (s *SortGeoDistance) MarshalJSON() ([]byte, error) {
sfm := map[string]interface{}{
"by": "geo_distance",
"field": s.Field,
"location": map[string]interface{}{
"lon": s.Lon,
"lat": s.Lat,
},
}
if s.Unit != "" {
sfm["unit"] = s.Unit
}
if s.Desc {
sfm["desc"] = true
}
return json.Marshal(sfm)
}
func (s *SortGeoDistance) Copy() SearchSort {
var rv SortGeoDistance
rv = *s
return &rv
}