0
0
Fork 0
bleve/test/versus_test.go

498 lines
14 KiB
Go
Raw Normal View History

// Copyright (c) 2014 Couchbase, Inc.
//
// 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 test
import (
"bytes"
"encoding/json"
"fmt"
"math"
"math/rand"
"os"
"reflect"
"strconv"
"strings"
"testing"
"text/template"
"github.com/blevesearch/bleve"
"github.com/blevesearch/bleve/index/scorch"
"github.com/blevesearch/bleve/index/store/boltdb"
"github.com/blevesearch/bleve/index/upsidedown"
"github.com/blevesearch/bleve/mapping"
"github.com/blevesearch/bleve/search"
)
// Tests scorch indexer versus upsidedown/bolt indexer against various
// templated queries. Example usage from the bleve top-level directory...
//
// go test -v -run TestScorchVersusUpsideDownBolt ./test
// VERBOSE=1 FOCUS=Trista go test -v -run TestScorchVersusUpsideDownBolt ./test
//
func TestScorchVersusUpsideDownBoltAll(t *testing.T) {
(&VersusTest{
t: t,
scorch fix disjunction searcher Advance() Found with "versus" test (TestScorchVersusUpsideDownBoltSmallMNSAM), which had a boolean query with a MustNot that was the same as the Must parameters. This replicates a situation found by Aruna/Mihir/testrunner/RQG (MB-27291). Example: "query": { "must_not": {"disjuncts": [ {"field": "body", "match": "hello"} ]}, "must": {"conjuncts": [ {"field": "body", "match": "hello"} ]} } The nested searchers along the MustNot pathway would end up looking roughly like... booleanSearcher MustNot => disjunctionSearcher => disjunctionSearcher => termSearcher On the first Next() call by the collector, the two disjunction searchers would run through their respective Next() method processing, which includes their initSearcher() processing on the first time. This has the effect of driving the leaf termSearcher through two Next() invocations. That is, if there were 3 docs (doc-1, doc-2, doc-3), the leaf termSearcher would at this point have moved to point to doc-3, while the topmost MustNot would have received doc-1. Next, the booleanSearcher's Must searcher would produce doc-2, so the booleanSearcher would try to Advance() the MustNot searcher to doc-2. But, in scorch, the leafmost termSearcher had already gotten past doc-2 and would return its doc-3. In upsidedown, in contrast, the leaf termSearcher would then drive the KVStore iterator with a Seek(doc-2), and the KVStore iterator would perform a backwards seek to reach doc-2. In scorch, however, backwards iteration seeking isn't supported. So, this fix checks the state of the disjunction searcher to see if we already have the necessary state so that we don't have to perform actual Advance()'es on the underlying searchers. This not only fixes the behavior w.r.t. scorch, but also can have an effect of potentially making upsidedown slightly faster as we're avoiding some backwards KVStore iterator seeks.
2017-12-22 02:49:55 +01:00
NumDocs: 1000,
MaxWordsPerDoc: 20,
scorch fix disjunction searcher Advance() Found with "versus" test (TestScorchVersusUpsideDownBoltSmallMNSAM), which had a boolean query with a MustNot that was the same as the Must parameters. This replicates a situation found by Aruna/Mihir/testrunner/RQG (MB-27291). Example: "query": { "must_not": {"disjuncts": [ {"field": "body", "match": "hello"} ]}, "must": {"conjuncts": [ {"field": "body", "match": "hello"} ]} } The nested searchers along the MustNot pathway would end up looking roughly like... booleanSearcher MustNot => disjunctionSearcher => disjunctionSearcher => termSearcher On the first Next() call by the collector, the two disjunction searchers would run through their respective Next() method processing, which includes their initSearcher() processing on the first time. This has the effect of driving the leaf termSearcher through two Next() invocations. That is, if there were 3 docs (doc-1, doc-2, doc-3), the leaf termSearcher would at this point have moved to point to doc-3, while the topmost MustNot would have received doc-1. Next, the booleanSearcher's Must searcher would produce doc-2, so the booleanSearcher would try to Advance() the MustNot searcher to doc-2. But, in scorch, the leafmost termSearcher had already gotten past doc-2 and would return its doc-3. In upsidedown, in contrast, the leaf termSearcher would then drive the KVStore iterator with a Seek(doc-2), and the KVStore iterator would perform a backwards seek to reach doc-2. In scorch, however, backwards iteration seeking isn't supported. So, this fix checks the state of the disjunction searcher to see if we already have the necessary state so that we don't have to perform actual Advance()'es on the underlying searchers. This not only fixes the behavior w.r.t. scorch, but also can have an effect of potentially making upsidedown slightly faster as we're avoiding some backwards KVStore iterator seeks.
2017-12-22 02:49:55 +01:00
NumWords: 10,
BatchSize: 10,
NumAttemptsPerSearch: 100,
}).run(scorch.Name, boltdb.Name, upsidedown.Name, boltdb.Name, nil, nil)
}
func TestScorchVersusUpsideDownBoltSmallMNSAM(t *testing.T) {
(&VersusTest{
t: t,
Focus: "must-not-same-as-must",
NumDocs: 5,
MaxWordsPerDoc: 2,
NumWords: 1,
BatchSize: 1,
NumAttemptsPerSearch: 1,
}).run(scorch.Name, boltdb.Name, upsidedown.Name, boltdb.Name, nil, nil)
}
func TestScorchVersusUpsideDownBoltSmallCMP11(t *testing.T) {
(&VersusTest{
t: t,
Focus: "conjuncts-match-phrase-1-1",
NumDocs: 30,
MaxWordsPerDoc: 8,
NumWords: 2,
BatchSize: 1,
NumAttemptsPerSearch: 1,
}).run(scorch.Name, boltdb.Name, upsidedown.Name, boltdb.Name, nil, nil)
}
// -------------------------------------------------------
// Templates used to compare search results in the "versus" tests.
var testVersusSearchTemplates = []string{
`{
"about": "expected to return zero hits",
"query": {
"query": "title:notARealTitle"
}
}`,
`{
"about": "try straight word()'s",
"query": {
"query": "body:{{word}}"
}
}`,
`{
"about": "conjuncts on same term",
"query": {
"conjuncts": [
{ "field": "body", "term": "{{word}}", "boost": 1.0 },
{ "field": "body", "term": "{{word}}", "boost": 1.0 }
]
}
}`,
`{
"about": "disjuncts on same term",
"query": {
"disjuncts": [
{ "field": "body", "term": "{{word}}", "boost": 1.0 },
{ "field": "body", "term": "{{word}}", "boost": 1.0 }
]
}
}`,
`{
"about": "never-matching-title-conjuncts",
"query": {
"conjuncts": [
{"field": "body", "match": "{{word}}"},
{"field": "body", "match": "{{word}}"},
{"field": "title", "match": "notAnActualTitle"}
]
}
}`,
`{
"about": "never-matching-title-disjuncts",
"query": {
"disjuncts": [
{"field": "body", "match": "{{word}}"},
{"field": "body", "match": "{{word}}"},
{"field": "title", "match": "notAnActualTitle"}
]
}
}`,
`{
"about": "must-not-never-matches",
"query": {
"must_not": {"disjuncts": [
{"field": "title", "match": "notAnActualTitle"}
]},
"should": {"disjuncts": [
{"field": "body", "match": "{{word}}"}
]}
}
}`,
`{
"about": "must-not-only",
"query": {
"must_not": {"disjuncts": [
{"field": "body", "term": "{{word}}"}
]}
}
}`,
`{
scorch fix disjunction searcher Advance() Found with "versus" test (TestScorchVersusUpsideDownBoltSmallMNSAM), which had a boolean query with a MustNot that was the same as the Must parameters. This replicates a situation found by Aruna/Mihir/testrunner/RQG (MB-27291). Example: "query": { "must_not": {"disjuncts": [ {"field": "body", "match": "hello"} ]}, "must": {"conjuncts": [ {"field": "body", "match": "hello"} ]} } The nested searchers along the MustNot pathway would end up looking roughly like... booleanSearcher MustNot => disjunctionSearcher => disjunctionSearcher => termSearcher On the first Next() call by the collector, the two disjunction searchers would run through their respective Next() method processing, which includes their initSearcher() processing on the first time. This has the effect of driving the leaf termSearcher through two Next() invocations. That is, if there were 3 docs (doc-1, doc-2, doc-3), the leaf termSearcher would at this point have moved to point to doc-3, while the topmost MustNot would have received doc-1. Next, the booleanSearcher's Must searcher would produce doc-2, so the booleanSearcher would try to Advance() the MustNot searcher to doc-2. But, in scorch, the leafmost termSearcher had already gotten past doc-2 and would return its doc-3. In upsidedown, in contrast, the leaf termSearcher would then drive the KVStore iterator with a Seek(doc-2), and the KVStore iterator would perform a backwards seek to reach doc-2. In scorch, however, backwards iteration seeking isn't supported. So, this fix checks the state of the disjunction searcher to see if we already have the necessary state so that we don't have to perform actual Advance()'es on the underlying searchers. This not only fixes the behavior w.r.t. scorch, but also can have an effect of potentially making upsidedown slightly faster as we're avoiding some backwards KVStore iterator seeks.
2017-12-22 02:49:55 +01:00
"about": "must-not-same-as-must -- see: MB-27291",
"query": {
"must_not": {"disjuncts": [
{"field": "body", "match": "{{word}}"}
]},
"must": {"conjuncts": [
{"field": "body", "match": "{{word}}"}
]}
}
}`,
`{
scorch fix disjunction searcher Advance() Found with "versus" test (TestScorchVersusUpsideDownBoltSmallMNSAM), which had a boolean query with a MustNot that was the same as the Must parameters. This replicates a situation found by Aruna/Mihir/testrunner/RQG (MB-27291). Example: "query": { "must_not": {"disjuncts": [ {"field": "body", "match": "hello"} ]}, "must": {"conjuncts": [ {"field": "body", "match": "hello"} ]} } The nested searchers along the MustNot pathway would end up looking roughly like... booleanSearcher MustNot => disjunctionSearcher => disjunctionSearcher => termSearcher On the first Next() call by the collector, the two disjunction searchers would run through their respective Next() method processing, which includes their initSearcher() processing on the first time. This has the effect of driving the leaf termSearcher through two Next() invocations. That is, if there were 3 docs (doc-1, doc-2, doc-3), the leaf termSearcher would at this point have moved to point to doc-3, while the topmost MustNot would have received doc-1. Next, the booleanSearcher's Must searcher would produce doc-2, so the booleanSearcher would try to Advance() the MustNot searcher to doc-2. But, in scorch, the leafmost termSearcher had already gotten past doc-2 and would return its doc-3. In upsidedown, in contrast, the leaf termSearcher would then drive the KVStore iterator with a Seek(doc-2), and the KVStore iterator would perform a backwards seek to reach doc-2. In scorch, however, backwards iteration seeking isn't supported. So, this fix checks the state of the disjunction searcher to see if we already have the necessary state so that we don't have to perform actual Advance()'es on the underlying searchers. This not only fixes the behavior w.r.t. scorch, but also can have an effect of potentially making upsidedown slightly faster as we're avoiding some backwards KVStore iterator seeks.
2017-12-22 02:49:55 +01:00
"about": "must-not-same-as-should",
"query": {
"must_not": {"disjuncts": [
{"field": "body", "match": "{{word}}"}
]},
"should": {"disjuncts": [
{"field": "body", "match": "{{word}}"}
]}
}
}`,
`{
scorch fix disjunction searcher Advance() Found with "versus" test (TestScorchVersusUpsideDownBoltSmallMNSAM), which had a boolean query with a MustNot that was the same as the Must parameters. This replicates a situation found by Aruna/Mihir/testrunner/RQG (MB-27291). Example: "query": { "must_not": {"disjuncts": [ {"field": "body", "match": "hello"} ]}, "must": {"conjuncts": [ {"field": "body", "match": "hello"} ]} } The nested searchers along the MustNot pathway would end up looking roughly like... booleanSearcher MustNot => disjunctionSearcher => disjunctionSearcher => termSearcher On the first Next() call by the collector, the two disjunction searchers would run through their respective Next() method processing, which includes their initSearcher() processing on the first time. This has the effect of driving the leaf termSearcher through two Next() invocations. That is, if there were 3 docs (doc-1, doc-2, doc-3), the leaf termSearcher would at this point have moved to point to doc-3, while the topmost MustNot would have received doc-1. Next, the booleanSearcher's Must searcher would produce doc-2, so the booleanSearcher would try to Advance() the MustNot searcher to doc-2. But, in scorch, the leafmost termSearcher had already gotten past doc-2 and would return its doc-3. In upsidedown, in contrast, the leaf termSearcher would then drive the KVStore iterator with a Seek(doc-2), and the KVStore iterator would perform a backwards seek to reach doc-2. In scorch, however, backwards iteration seeking isn't supported. So, this fix checks the state of the disjunction searcher to see if we already have the necessary state so that we don't have to perform actual Advance()'es on the underlying searchers. This not only fixes the behavior w.r.t. scorch, but also can have an effect of potentially making upsidedown slightly faster as we're avoiding some backwards KVStore iterator seeks.
2017-12-22 02:49:55 +01:00
"about": "inspired by testrunner RQG issue -- see: MB-27291",
"query": {
"must_not": {"disjuncts": [
{"field": "title", "match": "Trista Allen"},
{"field": "body", "match": "{{word}}"}
]},
"should": {"disjuncts": [
{"field": "title", "match": "Kallie Safiya Amara"},
{"field": "body", "match": "{{word}}"}
]}
}
}`,
`{
"about": "conjuncts-match-phrase-1-1 inspired by testrunner RQG issue -- see: MB-27291",
"query": {
"conjuncts": [
{"field": "body", "match": "{{bodyWord 0}}"},
{"field": "body", "match_phrase": "{{bodyWord 1}} {{bodyWord 1}}"}
]
}
}`,
`{
"about": "conjuncts-match-phrase-1-2 inspired by testrunner RQG issue -- see: MB-27291 -- FAILS!!",
"query": {
"conjuncts": [
{"field": "body", "match": "{{bodyWord 0}}"},
{"field": "body", "match_phrase": "{{bodyWord 1}} {{bodyWord 2}}"}
]
}
}`,
}
// -------------------------------------------------------
type VersusTest struct {
t *testing.T
// Use environment variable VERBOSE=<integer> that's > 0 for more
// verbose output.
Verbose int
// Allow user to focus on particular search templates, where
// where the search template must contain the Focus string.
Focus string
NumDocs int // Number of docs to insert.
MaxWordsPerDoc int // Max number words in each doc's Body field.
NumWords int // Total number of words in the dictionary.
BatchSize int // Batch size when inserting docs.
NumAttemptsPerSearch int // For each search template, number of searches to try.
// The Bodies is an array with length NumDocs, where each entry
// is the words in a doc's Body field.
Bodies [][]string
CurAttempt int
TotAttempts int
}
// -------------------------------------------------------
func testVersusSearches(vt *VersusTest, searchTemplates []string, idxA, idxB bleve.Index) {
t := vt.t
funcMap := template.FuncMap{
// Returns a word. The word may or may not be in any
// document's body.
"word": func() string {
return vt.genWord(vt.CurAttempt % vt.NumWords)
},
// Picks a document and returns the i'th word in that
// document's body. You can use this in searches to
// definitely find at least one document.
"bodyWord": func(i int) string {
body := vt.Bodies[vt.CurAttempt%len(vt.Bodies)]
if len(body) <= 0 {
return ""
}
return body[i%len(body)]
},
}
// Optionally allow call to focus on a particular search templates,
// where the search template must contain the vt.Focus string.
if vt.Focus == "" {
vt.Focus = os.Getenv("FOCUS")
}
for i, searchTemplate := range searchTemplates {
if vt.Focus != "" && !strings.Contains(searchTemplate, vt.Focus) {
continue
}
tmpl, err := template.New("search").Funcs(funcMap).Parse(searchTemplate)
if err != nil {
t.Fatalf("could not parse search template: %s, err: %v", searchTemplate, err)
}
for j := 0; j < vt.NumAttemptsPerSearch; j++ {
vt.CurAttempt = j
var buf bytes.Buffer
err = tmpl.Execute(&buf, vt)
if err != nil {
t.Fatalf("could not execute search template: %s, err: %v", searchTemplate, err)
}
bufBytes := buf.Bytes()
if vt.Verbose > 0 {
fmt.Printf(" %s\n", bufBytes)
}
var search bleve.SearchRequest
err = json.Unmarshal(bufBytes, &search)
if err != nil {
t.Fatalf("could not unmarshal search: %s, err: %v", bufBytes, err)
}
search.Size = vt.NumDocs * 10 // Crank up limit to get all results.
searchA := search
searchB := search
resA, errA := idxA.Search(&searchA)
resB, errB := idxB.Search(&searchB)
if errA != errB {
t.Errorf("search: (%d) %s,\n err mismatch, errA: %v, errB: %v",
i, bufBytes, errA, errB)
}
// Scores might have float64 vs float32 wobbles, so truncate precision.
resA.MaxScore = math.Trunc(resA.MaxScore*1000.0) / 1000.0
resB.MaxScore = math.Trunc(resB.MaxScore*1000.0) / 1000.0
// Timings may be different between A & B, so force equality.
resA.Took = resB.Took
// Hits might have different ordering since some indexers
// (like upsidedown) have a natural secondary sort on id
// while others (like scorch) don't. So, we compare by
// putting the hits from A & B into maps.
hitsA := hitsById(resA)
hitsB := hitsById(resB)
if !reflect.DeepEqual(hitsA, hitsB) {
t.Errorf("=========\nsearch: (%d) %s,\n res hits mismatch,\n len(hitsA): %d,\n len(hitsB): %d",
i, bufBytes, len(hitsA), len(hitsB))
t.Errorf("\n hitsA: %#v,\n hitsB: %#v",
hitsA, hitsB)
for id, hitA := range hitsA {
hitB := hitsB[id]
if !reflect.DeepEqual(hitA, hitB) {
t.Errorf("\n driving from hitsA\n hitA: %#v,\n hitB: %#v", hitA, hitB)
idx, _ := strconv.Atoi(id)
t.Errorf("\n doc: %d, body: %s", idx, strings.Join(vt.Bodies[idx], " "))
}
}
for id, hitB := range hitsB {
hitA := hitsA[id]
if !reflect.DeepEqual(hitA, hitB) {
t.Errorf("\n driving from hitsB\n hitA: %#v,\n hitB: %#v", hitA, hitB)
idx, _ := strconv.Atoi(id)
t.Errorf("\n doc: %d, body: %s", idx, strings.Join(vt.Bodies[idx], " "))
}
}
}
resA.Hits = nil
resB.Hits = nil
if !reflect.DeepEqual(resA, resB) {
resAj, _ := json.Marshal(resA)
resBj, _ := json.Marshal(resB)
t.Errorf("search: (%d) %s,\n res mismatch,\n resA: %s,\n resB: %s",
i, bufBytes, resAj, resBj)
}
if vt.Verbose > 0 {
fmt.Printf(" Total: (%t) %d\n", resA.Total == resB.Total, resA.Total)
}
vt.TotAttempts++
}
}
}
// Organizes the hits into a map keyed by id.
func hitsById(res *bleve.SearchResult) map[string]*search.DocumentMatch {
rv := make(map[string]*search.DocumentMatch, len(res.Hits))
for _, hit := range res.Hits {
// Clear out or truncate precision of hit fields that might be
// different across different indexer implementations.
hit.Index = ""
hit.Score = math.Trunc(hit.Score*1000.0) / 1000.0
hit.IndexInternalID = nil
hit.HitNumber = 0
rv[hit.ID] = hit
}
return rv
}
// -------------------------------------------------------
func (vt *VersusTest) run(indexTypeA, kvStoreA, indexTypeB, kvStoreB string,
cb func(versusTest *VersusTest, searchTemplates []string, idxA, idxB bleve.Index),
searchTemplates []string) {
if cb == nil {
cb = testVersusSearches
}
if searchTemplates == nil {
searchTemplates = testVersusSearchTemplates
}
if vt.Verbose <= 0 {
vt.Verbose, _ = strconv.Atoi(os.Getenv("VERBOSE"))
}
dirA := "/tmp/bleve-versus-test-a"
dirB := "/tmp/bleve-versus-test-b"
2017-12-22 01:49:12 +01:00
defer func() {
_ = os.RemoveAll(dirA)
_ = os.RemoveAll(dirB)
}()
2017-12-22 01:49:12 +01:00
_ = os.RemoveAll(dirA)
_ = os.RemoveAll(dirB)
imA := vt.makeIndexMapping()
imB := vt.makeIndexMapping()
kvConfigA := map[string]interface{}{}
kvConfigB := map[string]interface{}{}
idxA, err := bleve.NewUsing(dirA, imA, indexTypeA, kvStoreA, kvConfigA)
if err != nil || idxA == nil {
vt.t.Fatalf("new using err: %v", err)
}
2017-12-22 01:49:12 +01:00
defer func() { _ = idxA.Close() }()
idxB, err := bleve.NewUsing(dirB, imB, indexTypeB, kvStoreB, kvConfigB)
if err != nil || idxB == nil {
vt.t.Fatalf("new using err: %v", err)
}
2017-12-22 01:49:12 +01:00
defer func() { _ = idxB.Close() }()
rand.Seed(0)
if vt.Bodies == nil {
vt.Bodies = vt.genBodies()
}
vt.insertBodies(idxA)
vt.insertBodies(idxB)
cb(vt, searchTemplates, idxA, idxB)
}
// -------------------------------------------------------
func (vt *VersusTest) makeIndexMapping() mapping.IndexMapping {
standardFM := bleve.NewTextFieldMapping()
standardFM.Store = false
standardFM.IncludeInAll = false
standardFM.IncludeTermVectors = true
standardFM.Analyzer = "standard"
dm := bleve.NewDocumentMapping()
dm.AddFieldMappingsAt("title", standardFM)
dm.AddFieldMappingsAt("body", standardFM)
im := bleve.NewIndexMapping()
im.DefaultMapping = dm
im.DefaultAnalyzer = "standard"
return im
}
func (vt *VersusTest) insertBodies(idx bleve.Index) {
batch := idx.NewBatch()
for i, bodyWords := range vt.Bodies {
title := fmt.Sprintf("%d", i)
body := strings.Join(bodyWords, " ")
2017-12-22 01:49:12 +01:00
err := batch.Index(title, map[string]interface{}{"title": title, "body": body})
if err != nil {
vt.t.Fatalf("batch.Index err: %v", err)
}
if i%vt.BatchSize == 0 {
2017-12-22 01:49:12 +01:00
err = idx.Batch(batch)
if err != nil {
vt.t.Fatalf("batch err: %v", err)
}
batch.Reset()
}
}
err := idx.Batch(batch)
if err != nil {
vt.t.Fatalf("last batch err: %v", err)
}
}
func (vt *VersusTest) genBodies() (rv [][]string) {
for i := 0; i < vt.NumDocs; i++ {
rv = append(rv, vt.genBody())
}
return rv
}
func (vt *VersusTest) genBody() (rv []string) {
m := rand.Intn(vt.MaxWordsPerDoc)
for j := 0; j < m; j++ {
rv = append(rv, vt.genWord(rand.Intn(vt.NumWords)))
}
return rv
}
func (vt *VersusTest) genWord(i int) string {
return fmt.Sprintf("%x", i)
}