0
0
Fork 0
bleve/index_alias_impl_test.go

1312 lines
30 KiB
Go
Raw Normal View History

package bleve
import (
"fmt"
2014-11-25 19:50:15 +01:00
"reflect"
"testing"
2014-11-25 19:50:15 +01:00
"time"
"golang.org/x/net/context"
"github.com/blevesearch/bleve/document"
"github.com/blevesearch/bleve/index"
"github.com/blevesearch/bleve/index/store"
"github.com/blevesearch/bleve/mapping"
"github.com/blevesearch/bleve/numeric"
2014-11-25 19:50:15 +01:00
"github.com/blevesearch/bleve/search"
)
func TestIndexAliasSingle(t *testing.T) {
expectedError := fmt.Errorf("expected")
ei1 := &stubIndex{
err: expectedError,
}
alias := NewIndexAlias(ei1)
err := alias.Index("a", "a")
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
err = alias.Delete("a")
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
batch := alias.NewBatch()
err = alias.Batch(batch)
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
_, err = alias.Document("a")
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
_, err = alias.Fields()
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
_, err = alias.GetInternal([]byte("a"))
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
err = alias.SetInternal([]byte("a"), []byte("a"))
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
err = alias.DeleteInternal([]byte("a"))
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
mapping := alias.Mapping()
if mapping != nil {
t.Errorf("expected nil, got %v", mapping)
}
indexStat := alias.Stats()
if indexStat != nil {
t.Errorf("expected nil, got %v", indexStat)
}
// now a few things that should work
sr := NewSearchRequest(NewTermQuery("test"))
_, err = alias.Search(sr)
if err != expectedError {
t.Errorf("expected %v, got %v", expectedError, err)
}
2016-02-23 19:05:16 +01:00
count, err := alias.DocCount()
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if count != 0 {
t.Errorf("expected count 0, got %d", count)
}
// now change the def using add/remove
expectedError2 := fmt.Errorf("expected2")
ei2 := &stubIndex{
err: expectedError2,
}
alias.Add(ei2)
alias.Remove(ei1)
err = alias.Index("a", "a")
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
err = alias.Delete("a")
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
err = alias.Batch(batch)
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
_, err = alias.Document("a")
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
_, err = alias.Fields()
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
_, err = alias.GetInternal([]byte("a"))
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
err = alias.SetInternal([]byte("a"), []byte("a"))
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
err = alias.DeleteInternal([]byte("a"))
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
mapping = alias.Mapping()
if mapping != nil {
t.Errorf("expected nil, got %v", mapping)
}
indexStat = alias.Stats()
if indexStat != nil {
t.Errorf("expected nil, got %v", indexStat)
}
// now a few things that should work
_, err = alias.Search(sr)
if err != expectedError2 {
t.Errorf("expected %v, got %v", expectedError2, err)
}
2016-02-23 19:05:16 +01:00
count, err = alias.DocCount()
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if count != 0 {
t.Errorf("expected count 0, got %d", count)
}
// now change the def using swap
expectedError3 := fmt.Errorf("expected3")
ei3 := &stubIndex{
err: expectedError3,
}
alias.Swap([]Index{ei3}, []Index{ei2})
err = alias.Index("a", "a")
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
err = alias.Delete("a")
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
err = alias.Batch(batch)
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
_, err = alias.Document("a")
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
_, err = alias.Fields()
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
_, err = alias.GetInternal([]byte("a"))
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
err = alias.SetInternal([]byte("a"), []byte("a"))
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
err = alias.DeleteInternal([]byte("a"))
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
mapping = alias.Mapping()
if mapping != nil {
t.Errorf("expected nil, got %v", mapping)
}
indexStat = alias.Stats()
if indexStat != nil {
t.Errorf("expected nil, got %v", indexStat)
}
// now a few things that should work
_, err = alias.Search(sr)
if err != expectedError3 {
t.Errorf("expected %v, got %v", expectedError3, err)
}
2016-02-23 19:05:16 +01:00
count, err = alias.DocCount()
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if count != 0 {
t.Errorf("expected count 0, got %d", count)
}
}
func TestIndexAliasClosed(t *testing.T) {
alias := NewIndexAlias()
err := alias.Close()
if err != nil {
t.Fatal(err)
}
err = alias.Index("a", "a")
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
err = alias.Delete("a")
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
batch := alias.NewBatch()
err = alias.Batch(batch)
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
_, err = alias.Document("a")
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
_, err = alias.Fields()
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
_, err = alias.GetInternal([]byte("a"))
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
err = alias.SetInternal([]byte("a"), []byte("a"))
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
err = alias.DeleteInternal([]byte("a"))
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
mapping := alias.Mapping()
if mapping != nil {
t.Errorf("expected nil, got %v", mapping)
}
indexStat := alias.Stats()
if indexStat != nil {
t.Errorf("expected nil, got %v", indexStat)
}
// now a few things that should work
sr := NewSearchRequest(NewTermQuery("test"))
_, err = alias.Search(sr)
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
_, err = alias.DocCount()
if err != ErrorIndexClosed {
t.Errorf("expected %v, got %v", ErrorIndexClosed, err)
}
}
func TestIndexAliasEmpty(t *testing.T) {
alias := NewIndexAlias()
err := alias.Index("a", "a")
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
err = alias.Delete("a")
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
batch := alias.NewBatch()
err = alias.Batch(batch)
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
_, err = alias.Document("a")
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
_, err = alias.Fields()
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
_, err = alias.GetInternal([]byte("a"))
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
err = alias.SetInternal([]byte("a"), []byte("a"))
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
err = alias.DeleteInternal([]byte("a"))
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
mapping := alias.Mapping()
if mapping != nil {
t.Errorf("expected nil, got %v", mapping)
}
indexStat := alias.Stats()
if indexStat != nil {
t.Errorf("expected nil, got %v", indexStat)
}
// now a few things that should work
sr := NewSearchRequest(NewTermQuery("test"))
_, err = alias.Search(sr)
if err != ErrorAliasEmpty {
t.Errorf("expected %v, got %v", ErrorAliasEmpty, err)
}
count, err := alias.DocCount()
2016-04-03 04:42:56 +02:00
if err != nil {
t.Errorf("error getting alias doc count: %v", err)
}
if count != 0 {
t.Errorf("expected %d, got %d", 0, count)
}
}
func TestIndexAliasMulti(t *testing.T) {
score1, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(1.0), 0)
score2, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(2.0), 0)
ei1Count := uint64(7)
ei1 := &stubIndex{
err: nil,
docCountResult: &ei1Count,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 1.0,
}}
ei2Count := uint64(8)
ei2 := &stubIndex{
err: nil,
docCountResult: &ei2Count,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
},
},
MaxScore: 2.0,
}}
alias := NewIndexAlias(ei1, ei2)
err := alias.Index("a", "a")
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
err = alias.Delete("a")
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
batch := alias.NewBatch()
err = alias.Batch(batch)
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
_, err = alias.Document("a")
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
_, err = alias.Fields()
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
_, err = alias.GetInternal([]byte("a"))
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
err = alias.SetInternal([]byte("a"), []byte("a"))
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
err = alias.DeleteInternal([]byte("a"))
if err != ErrorAliasMulti {
t.Errorf("expected %v, got %v", ErrorAliasMulti, err)
}
mapping := alias.Mapping()
if mapping != nil {
t.Errorf("expected nil, got %v", mapping)
}
indexStat := alias.Stats()
if indexStat != nil {
t.Errorf("expected nil, got %v", indexStat)
}
// now a few things that should work
sr := NewSearchRequest(NewTermQuery("test"))
expected := &SearchResult{
Status: &SearchStatus{
Total: 2,
Successful: 2,
Errors: make(map[string]error),
},
Request: sr,
Total: 2,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
},
2016-04-03 03:54:33 +02:00
{
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 2.0,
}
results, err := alias.Search(sr)
if err != nil {
t.Error(err)
}
// cheat and ensure that Took field matches since it invovles time
expected.Took = results.Took
if !reflect.DeepEqual(results, expected) {
t.Errorf("expected %#v, got %#v", expected, results)
}
count, err := alias.DocCount()
2016-04-03 04:42:56 +02:00
if err != nil {
t.Errorf("error getting alias doc count: %v", err)
}
if count != (*ei1.docCountResult + *ei2.docCountResult) {
t.Errorf("expected %d, got %d", (*ei1.docCountResult + *ei2.docCountResult), count)
}
}
2014-11-25 19:50:15 +01:00
// TestMultiSearchNoError
func TestMultiSearchNoError(t *testing.T) {
score1, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(1.0), 0)
score2, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(2.0), 0)
2014-11-25 19:50:15 +01:00
ei1 := &stubIndex{err: nil, searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
2014-11-25 19:50:15 +01:00
Total: 1,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
Index: "1",
2014-11-25 19:50:15 +01:00
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
2014-11-25 19:50:15 +01:00
},
},
MaxScore: 1.0,
}}
ei2 := &stubIndex{err: nil, searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
2014-11-25 19:50:15 +01:00
Total: 1,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
Index: "2",
2014-11-25 19:50:15 +01:00
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
2014-11-25 19:50:15 +01:00
},
},
MaxScore: 2.0,
}}
sr := NewSearchRequest(NewTermQuery("test"))
2014-11-25 19:50:15 +01:00
expected := &SearchResult{
Status: &SearchStatus{
Total: 2,
Successful: 2,
Errors: make(map[string]error),
},
Request: sr,
Total: 2,
2014-11-25 19:50:15 +01:00
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
Index: "2",
2014-11-25 19:50:15 +01:00
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
2014-11-25 19:50:15 +01:00
},
2016-04-03 03:54:33 +02:00
{
Index: "1",
2014-11-25 19:50:15 +01:00
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
2014-11-25 19:50:15 +01:00
},
},
MaxScore: 2.0,
}
results, err := MultiSearch(context.Background(), sr, ei1, ei2)
2014-11-25 19:50:15 +01:00
if err != nil {
t.Error(err)
}
// cheat and ensure that Took field matches since it invovles time
expected.Took = results.Took
2014-11-25 19:50:15 +01:00
if !reflect.DeepEqual(results, expected) {
t.Errorf("expected %#v, got %#v", expected, results)
}
}
// TestMultiSearchSomeError
func TestMultiSearchSomeError(t *testing.T) {
ei1 := &stubIndex{name: "ei1", err: nil, searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
2014-11-25 19:50:15 +01:00
Total: 1,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
2014-11-25 19:50:15 +01:00
ID: "a",
Score: 1.0,
},
},
Took: 1 * time.Second,
MaxScore: 1.0,
}}
ei2 := &stubIndex{name: "ei2", err: fmt.Errorf("deliberate error")}
2014-11-25 19:50:15 +01:00
sr := NewSearchRequest(NewTermQuery("test"))
res, err := MultiSearch(context.Background(), sr, ei1, ei2)
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if res.Status.Total != 2 {
t.Errorf("expected 2 indexes to be queried, got %d", res.Status.Total)
}
if res.Status.Failed != 1 {
t.Errorf("expected 1 index to fail, got %d", res.Status.Failed)
}
if res.Status.Successful != 1 {
t.Errorf("expected 1 index to be successful, got %d", res.Status.Successful)
}
if len(res.Status.Errors) != 1 {
t.Fatalf("expected 1 status error message, got %d", len(res.Status.Errors))
}
if res.Status.Errors["ei2"].Error() != "deliberate error" {
t.Errorf("expected ei2 index error message 'deliberate error', got '%s'", res.Status.Errors["ei2"])
2014-11-25 19:50:15 +01:00
}
}
// TestMultiSearchAllError
// reproduces https://github.com/blevesearch/bleve/issues/126
func TestMultiSearchAllError(t *testing.T) {
ei1 := &stubIndex{name: "ei1", err: fmt.Errorf("deliberate error")}
ei2 := &stubIndex{name: "ei2", err: fmt.Errorf("deliberate error")}
sr := NewSearchRequest(NewTermQuery("test"))
res, err := MultiSearch(context.Background(), sr, ei1, ei2)
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if res.Status.Total != 2 {
t.Errorf("expected 2 indexes to be queried, got %d", res.Status.Total)
}
if res.Status.Failed != 2 {
t.Errorf("expected 2 indexes to fail, got %d", res.Status.Failed)
}
if res.Status.Successful != 0 {
t.Errorf("expected 0 indexes to be successful, got %d", res.Status.Successful)
}
if len(res.Status.Errors) != 2 {
t.Fatalf("expected 2 status error messages, got %d", len(res.Status.Errors))
}
if res.Status.Errors["ei1"].Error() != "deliberate error" {
t.Errorf("expected ei1 index error message 'deliberate error', got '%s'", res.Status.Errors["ei1"])
}
if res.Status.Errors["ei2"].Error() != "deliberate error" {
t.Errorf("expected ei2 index error message 'deliberate error', got '%s'", res.Status.Errors["ei2"])
}
}
func TestMultiSearchSecondPage(t *testing.T) {
checkRequest := func(sr *SearchRequest) error {
if sr.From != 0 {
return fmt.Errorf("child request from should be 0")
}
if sr.Size != 20 {
return fmt.Errorf("child request size should be 20")
}
return nil
}
ei1 := &stubIndex{
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
},
checkRequest: checkRequest,
}
ei2 := &stubIndex{
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
},
checkRequest: checkRequest,
}
sr := NewSearchRequestOptions(NewTermQuery("test"), 10, 10, false)
_, err := MultiSearch(context.Background(), sr, ei1, ei2)
if err != nil {
t.Errorf("unexpected error %v", err)
}
}
// TestMultiSearchTimeout tests simple timeout cases
// 1. all searches finish successfully before timeout
// 2. no searchers finish before the timeout
// 3. no searches finish before cancellation
func TestMultiSearchTimeout(t *testing.T) {
score1, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(1.0), 0)
score2, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(2.0), 0)
ei1 := &stubIndex{
name: "ei1",
checkRequest: func(req *SearchRequest) error {
time.Sleep(50 * time.Millisecond)
return nil
},
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 1.0,
}}
ei2 := &stubIndex{
name: "ei2",
checkRequest: func(req *SearchRequest) error {
time.Sleep(50 * time.Millisecond)
return nil
},
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "2",
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
},
},
MaxScore: 2.0,
}}
// first run with absurdly long time out, should succeed
ctx, _ := context.WithTimeout(context.Background(), 10*time.Second)
query := NewTermQuery("test")
sr := NewSearchRequest(query)
res, err := MultiSearch(ctx, sr, ei1, ei2)
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if res.Status.Total != 2 {
t.Errorf("expected 2 total, got %d", res.Status.Failed)
}
if res.Status.Successful != 2 {
t.Errorf("expected 0 success, got %d", res.Status.Successful)
}
if res.Status.Failed != 0 {
t.Errorf("expected 2 failed, got %d", res.Status.Failed)
}
if len(res.Status.Errors) != 0 {
t.Errorf("expected 0 errors, got %v", res.Status.Errors)
}
// now run a search again with an absurdly low timeout (should timeout)
ctx, _ = context.WithTimeout(context.Background(), 1*time.Microsecond)
res, err = MultiSearch(ctx, sr, ei1, ei2)
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if res.Status.Total != 2 {
t.Errorf("expected 2 failed, got %d", res.Status.Failed)
}
if res.Status.Successful != 0 {
t.Errorf("expected 0 success, got %d", res.Status.Successful)
}
if res.Status.Failed != 2 {
t.Errorf("expected 2 failed, got %d", res.Status.Failed)
}
if len(res.Status.Errors) != 2 {
t.Errorf("expected 2 errors, got %v", res.Status.Errors)
} else {
if res.Status.Errors["ei1"].Error() != context.DeadlineExceeded.Error() {
t.Errorf("expected err for 'ei1' to be '%s' got '%s'", context.DeadlineExceeded.Error(), res.Status.Errors["ei1"])
}
if res.Status.Errors["ei2"].Error() != context.DeadlineExceeded.Error() {
t.Errorf("expected err for 'ei2' to be '%s' got '%s'", context.DeadlineExceeded.Error(), res.Status.Errors["ei2"])
}
}
// now run a search again with a normal timeout, but cancel it first
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
cancel()
res, err = MultiSearch(ctx, sr, ei1, ei2)
if err != nil {
t.Errorf("expected no error, got %v", err)
}
if res.Status.Total != 2 {
t.Errorf("expected 2 failed, got %d", res.Status.Failed)
}
if res.Status.Successful != 0 {
t.Errorf("expected 0 success, got %d", res.Status.Successful)
}
if res.Status.Failed != 2 {
t.Errorf("expected 2 failed, got %d", res.Status.Failed)
}
if len(res.Status.Errors) != 2 {
t.Errorf("expected 2 errors, got %v", res.Status.Errors)
} else {
if res.Status.Errors["ei1"].Error() != context.Canceled.Error() {
t.Errorf("expected err for 'ei1' to be '%s' got '%s'", context.Canceled.Error(), res.Status.Errors["ei1"])
}
if res.Status.Errors["ei2"].Error() != context.Canceled.Error() {
t.Errorf("expected err for 'ei2' to be '%s' got '%s'", context.Canceled.Error(), res.Status.Errors["ei2"])
}
}
}
// TestMultiSearchTimeoutPartial tests the case where some indexes exceed
// the timeout, while others complete successfully
func TestMultiSearchTimeoutPartial(t *testing.T) {
score1, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(1.0), 0)
score2, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(2.0), 0)
score3, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(3.0), 0)
ei1 := &stubIndex{
name: "ei1",
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 1.0,
}}
ei2 := &stubIndex{
name: "ei2",
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "2",
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
},
},
MaxScore: 2.0,
}}
ei3 := &stubIndex{
name: "ei3",
checkRequest: func(req *SearchRequest) error {
time.Sleep(50 * time.Millisecond)
return nil
},
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "3",
ID: "c",
Score: 3.0,
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
Sort: []string{string(score3)},
},
},
MaxScore: 3.0,
}}
// ei3 is set to take >50ms, so run search with timeout less than
// this, this should return partial results
ctx, _ := context.WithTimeout(context.Background(), 25*time.Millisecond)
query := NewTermQuery("test")
sr := NewSearchRequest(query)
expected := &SearchResult{
Status: &SearchStatus{
Total: 3,
Successful: 2,
Failed: 1,
Errors: map[string]error{
"ei3": context.DeadlineExceeded,
},
},
Request: sr,
Total: 2,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
Index: "2",
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
},
2016-04-03 03:54:33 +02:00
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 2.0,
}
res, err := MultiSearch(ctx, sr, ei1, ei2, ei3)
if err != nil {
t.Fatalf("expected no err, got %v", err)
}
expected.Took = res.Took
if !reflect.DeepEqual(res, expected) {
t.Errorf("expected %#v, got %#v", expected, res)
}
}
func TestIndexAliasMultipleLayer(t *testing.T) {
score1, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(1.0), 0)
score2, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(2.0), 0)
score3, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(3.0), 0)
score4, _ := numeric.NewPrefixCodedInt64(numeric.Float64ToInt64(4.0), 0)
ei1 := &stubIndex{
name: "ei1",
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 1.0,
}}
ei2 := &stubIndex{
name: "ei2",
checkRequest: func(req *SearchRequest) error {
time.Sleep(50 * time.Millisecond)
return nil
},
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "2",
ID: "b",
Score: 2.0,
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
Sort: []string{string(score2)},
},
},
MaxScore: 2.0,
}}
ei3 := &stubIndex{
name: "ei3",
checkRequest: func(req *SearchRequest) error {
time.Sleep(50 * time.Millisecond)
return nil
},
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "3",
ID: "c",
Score: 3.0,
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
Sort: []string{string(score3)},
},
},
MaxScore: 3.0,
}}
ei4 := &stubIndex{
name: "ei4",
err: nil,
searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 1,
Hits: []*search.DocumentMatch{
2016-04-03 03:54:33 +02:00
{
Index: "4",
ID: "d",
Score: 4.0,
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
Sort: []string{string(score4)},
},
},
MaxScore: 4.0,
}}
alias1 := NewIndexAlias(ei1, ei2)
alias2 := NewIndexAlias(ei3, ei4)
aliasTop := NewIndexAlias(alias1, alias2)
// ei2 and ei3 have 50ms delay
// search across aliasTop should still get results from ei1 and ei4
// total should still be 4
ctx, _ := context.WithTimeout(context.Background(), 25*time.Millisecond)
query := NewTermQuery("test")
sr := NewSearchRequest(query)
expected := &SearchResult{
Status: &SearchStatus{
Total: 4,
Successful: 2,
Failed: 2,
Errors: map[string]error{
"ei2": context.DeadlineExceeded,
"ei3": context.DeadlineExceeded,
},
},
Request: sr,
Total: 2,
Hits: search.DocumentMatchCollection{
2016-04-03 03:54:33 +02:00
{
Index: "4",
ID: "d",
Score: 4.0,
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
Sort: []string{string(score4)},
},
2016-04-03 03:54:33 +02:00
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{string(score1)},
},
},
MaxScore: 4.0,
}
res, err := aliasTop.SearchInContext(ctx, sr)
if err != nil {
t.Fatalf("expected no err, got %v", err)
}
expected.Took = res.Took
if !reflect.DeepEqual(res, expected) {
t.Errorf("expected %#v, got %#v", expected, res)
}
}
// TestMultiSearchNoError
func TestMultiSearchCustomSort(t *testing.T) {
ei1 := &stubIndex{err: nil, searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 2,
Hits: search.DocumentMatchCollection{
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{"albert"},
},
{
Index: "1",
ID: "b",
Score: 2.0,
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
Sort: []string{"crown"},
},
},
MaxScore: 2.0,
}}
ei2 := &stubIndex{err: nil, searchResult: &SearchResult{
Status: &SearchStatus{
Total: 1,
Successful: 1,
Errors: make(map[string]error),
},
Total: 2,
Hits: search.DocumentMatchCollection{
{
Index: "2",
ID: "c",
Score: 2.5,
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
Sort: []string{"frank"},
},
{
Index: "2",
ID: "d",
Score: 3.0,
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
Sort: []string{"zombie"},
},
},
MaxScore: 3.0,
}}
sr := NewSearchRequest(NewTermQuery("test"))
sr.SortBy([]string{"name"})
expected := &SearchResult{
Status: &SearchStatus{
Total: 2,
Successful: 2,
Errors: make(map[string]error),
},
Request: sr,
Total: 4,
Hits: search.DocumentMatchCollection{
{
Index: "1",
ID: "a",
Score: 1.0,
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
Sort: []string{"albert"},
},
{
Index: "1",
ID: "b",
Score: 2.0,
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
Sort: []string{"crown"},
},
{
Index: "2",
ID: "c",
Score: 2.5,
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
Sort: []string{"frank"},
},
{
Index: "2",
ID: "d",
Score: 3.0,
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
Sort: []string{"zombie"},
},
},
MaxScore: 3.0,
}
results, err := MultiSearch(context.Background(), sr, ei1, ei2)
if err != nil {
t.Error(err)
}
// cheat and ensure that Took field matches since it invovles time
expected.Took = results.Took
if !reflect.DeepEqual(results, expected) {
t.Errorf("expected %v, got %v", expected, results)
}
}
2014-11-25 19:50:15 +01:00
// stubIndex is an Index impl for which all operations
// return the configured error value, unless the
// corresponding operation result value has been
// set, in which case that is returned instead
type stubIndex struct {
name string
2014-11-25 19:50:15 +01:00
err error
searchResult *SearchResult
documentResult *document.Document
docCountResult *uint64
checkRequest func(*SearchRequest) error
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Index(id string, data interface{}) error {
return i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Delete(id string) error {
return i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Batch(b *Batch) error {
return i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Document(id string) (*document.Document, error) {
if i.documentResult != nil {
return i.documentResult, nil
}
return nil, i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) DocCount() (uint64, error) {
if i.docCountResult != nil {
return *i.docCountResult, nil
}
return 0, i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Search(req *SearchRequest) (*SearchResult, error) {
return i.SearchInContext(context.Background(), req)
}
func (i *stubIndex) SearchInContext(ctx context.Context, req *SearchRequest) (*SearchResult, error) {
if i.checkRequest != nil {
err := i.checkRequest(req)
if err != nil {
return nil, err
}
}
2014-11-25 19:50:15 +01:00
if i.searchResult != nil {
return i.searchResult, nil
}
return nil, i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Fields() ([]string, error) {
return nil, i.err
}
func (i *stubIndex) FieldDict(field string) (index.FieldDict, error) {
return nil, i.err
}
func (i *stubIndex) FieldDictRange(field string, startTerm []byte, endTerm []byte) (index.FieldDict, error) {
return nil, i.err
}
func (i *stubIndex) FieldDictPrefix(field string, termPrefix []byte) (index.FieldDict, error) {
return nil, i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Close() error {
return i.err
}
func (i *stubIndex) Mapping() mapping.IndexMapping {
return nil
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) Stats() *IndexStat {
return nil
}
func (i *stubIndex) StatsMap() map[string]interface{} {
return nil
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) GetInternal(key []byte) ([]byte, error) {
return nil, i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) SetInternal(key, val []byte) error {
return i.err
}
2014-11-25 19:50:15 +01:00
func (i *stubIndex) DeleteInternal(key []byte) error {
return i.err
}
func (i *stubIndex) Advanced() (index.Index, store.KVStore, error) {
return nil, nil, nil
}
func (i *stubIndex) NewBatch() *Batch {
return &Batch{}
}
func (i *stubIndex) Name() string {
return i.name
}
func (i *stubIndex) SetName(name string) {
i.name = name
}