mirror of
https://github.com/minio/minio.git
synced 2024-12-27 15:45:55 -05:00
a2a8d54bb6
This change adds `access` format support for notifications to a Elasticsearch server, and it refactors `namespace` format support. In the case of `access` format, for each event in Minio, a JSON document is inserted into Elasticsearch with its timestamp set to the event's timestamp, and with the ID generated automatically by elasticsearch. No events are modified or deleted in this mode. In the case of `namespace` format, for each event in Minio, a JSON document is keyed together by the bucket and object name is updated in Elasticsearch. In the case of an object being created or over-written in Minio, a new document or an existing document is inserted into the Elasticsearch index. If an object is deleted in Minio, the corresponding document is deleted from the Elasticsearch index. Additionally, this change upgrades Elasticsearch support to the 5.x series. This is a breaking change, and users of previous elasticsearch versions should upgrade. Also updates documentation on Elasticsearch notification target usage and has a link to an elasticsearch upgrade guide. This is the last patch that finally resolves #3928.
547 lines
14 KiB
Go
547 lines
14 KiB
Go
// Copyright 2012-present Oliver Eilhard. All rights reserved.
|
|
// Use of this source code is governed by a MIT-license.
|
|
// See http://olivere.mit-license.org/license.txt for details.
|
|
|
|
package elastic
|
|
|
|
// PhraseSuggester provides an API to access word alternatives
|
|
// on a per token basis within a certain string distance.
|
|
// For more details, see
|
|
// https://www.elastic.co/guide/en/elasticsearch/reference/5.2/search-suggesters-phrase.html.
|
|
type PhraseSuggester struct {
|
|
Suggester
|
|
name string
|
|
text string
|
|
field string
|
|
analyzer string
|
|
size *int
|
|
shardSize *int
|
|
contextQueries []SuggesterContextQuery
|
|
|
|
// fields specific to a phrase suggester
|
|
maxErrors *float64
|
|
separator *string
|
|
realWordErrorLikelihood *float64
|
|
confidence *float64
|
|
generators map[string][]CandidateGenerator
|
|
gramSize *int
|
|
smoothingModel SmoothingModel
|
|
forceUnigrams *bool
|
|
tokenLimit *int
|
|
preTag, postTag *string
|
|
collateQuery *string
|
|
collatePreference *string
|
|
collateParams map[string]interface{}
|
|
collatePrune *bool
|
|
}
|
|
|
|
// NewPhraseSuggester creates a new PhraseSuggester.
|
|
func NewPhraseSuggester(name string) *PhraseSuggester {
|
|
return &PhraseSuggester{
|
|
name: name,
|
|
collateParams: make(map[string]interface{}),
|
|
}
|
|
}
|
|
|
|
func (q *PhraseSuggester) Name() string {
|
|
return q.name
|
|
}
|
|
|
|
func (q *PhraseSuggester) Text(text string) *PhraseSuggester {
|
|
q.text = text
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) Field(field string) *PhraseSuggester {
|
|
q.field = field
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) Analyzer(analyzer string) *PhraseSuggester {
|
|
q.analyzer = analyzer
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) Size(size int) *PhraseSuggester {
|
|
q.size = &size
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) ShardSize(shardSize int) *PhraseSuggester {
|
|
q.shardSize = &shardSize
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) ContextQuery(query SuggesterContextQuery) *PhraseSuggester {
|
|
q.contextQueries = append(q.contextQueries, query)
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) ContextQueries(queries ...SuggesterContextQuery) *PhraseSuggester {
|
|
q.contextQueries = append(q.contextQueries, queries...)
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) GramSize(gramSize int) *PhraseSuggester {
|
|
if gramSize >= 1 {
|
|
q.gramSize = &gramSize
|
|
}
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) MaxErrors(maxErrors float64) *PhraseSuggester {
|
|
q.maxErrors = &maxErrors
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) Separator(separator string) *PhraseSuggester {
|
|
q.separator = &separator
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) RealWordErrorLikelihood(realWordErrorLikelihood float64) *PhraseSuggester {
|
|
q.realWordErrorLikelihood = &realWordErrorLikelihood
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) Confidence(confidence float64) *PhraseSuggester {
|
|
q.confidence = &confidence
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) CandidateGenerator(generator CandidateGenerator) *PhraseSuggester {
|
|
if q.generators == nil {
|
|
q.generators = make(map[string][]CandidateGenerator)
|
|
}
|
|
typ := generator.Type()
|
|
if _, found := q.generators[typ]; !found {
|
|
q.generators[typ] = make([]CandidateGenerator, 0)
|
|
}
|
|
q.generators[typ] = append(q.generators[typ], generator)
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) CandidateGenerators(generators ...CandidateGenerator) *PhraseSuggester {
|
|
for _, g := range generators {
|
|
q = q.CandidateGenerator(g)
|
|
}
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) ClearCandidateGenerator() *PhraseSuggester {
|
|
q.generators = nil
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) ForceUnigrams(forceUnigrams bool) *PhraseSuggester {
|
|
q.forceUnigrams = &forceUnigrams
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) SmoothingModel(smoothingModel SmoothingModel) *PhraseSuggester {
|
|
q.smoothingModel = smoothingModel
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) TokenLimit(tokenLimit int) *PhraseSuggester {
|
|
q.tokenLimit = &tokenLimit
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) Highlight(preTag, postTag string) *PhraseSuggester {
|
|
q.preTag = &preTag
|
|
q.postTag = &postTag
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) CollateQuery(collateQuery string) *PhraseSuggester {
|
|
q.collateQuery = &collateQuery
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) CollatePreference(collatePreference string) *PhraseSuggester {
|
|
q.collatePreference = &collatePreference
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) CollateParams(collateParams map[string]interface{}) *PhraseSuggester {
|
|
q.collateParams = collateParams
|
|
return q
|
|
}
|
|
|
|
func (q *PhraseSuggester) CollatePrune(collatePrune bool) *PhraseSuggester {
|
|
q.collatePrune = &collatePrune
|
|
return q
|
|
}
|
|
|
|
// phraseSuggesterRequest is necessary because the order in which
|
|
// the JSON elements are routed to Elasticsearch is relevant.
|
|
// We got into trouble when using plain maps because the text element
|
|
// needs to go before the simple_phrase element.
|
|
type phraseSuggesterRequest struct {
|
|
Text string `json:"text"`
|
|
Phrase interface{} `json:"phrase"`
|
|
}
|
|
|
|
// Source generates the source for the phrase suggester.
|
|
func (q *PhraseSuggester) Source(includeName bool) (interface{}, error) {
|
|
ps := &phraseSuggesterRequest{}
|
|
|
|
if q.text != "" {
|
|
ps.Text = q.text
|
|
}
|
|
|
|
suggester := make(map[string]interface{})
|
|
ps.Phrase = suggester
|
|
|
|
if q.analyzer != "" {
|
|
suggester["analyzer"] = q.analyzer
|
|
}
|
|
if q.field != "" {
|
|
suggester["field"] = q.field
|
|
}
|
|
if q.size != nil {
|
|
suggester["size"] = *q.size
|
|
}
|
|
if q.shardSize != nil {
|
|
suggester["shard_size"] = *q.shardSize
|
|
}
|
|
switch len(q.contextQueries) {
|
|
case 0:
|
|
case 1:
|
|
src, err := q.contextQueries[0].Source()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
suggester["context"] = src
|
|
default:
|
|
var ctxq []interface{}
|
|
for _, query := range q.contextQueries {
|
|
src, err := query.Source()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
ctxq = append(ctxq, src)
|
|
}
|
|
suggester["context"] = ctxq
|
|
}
|
|
|
|
// Phase-specified parameters
|
|
if q.realWordErrorLikelihood != nil {
|
|
suggester["real_word_error_likelihood"] = *q.realWordErrorLikelihood
|
|
}
|
|
if q.confidence != nil {
|
|
suggester["confidence"] = *q.confidence
|
|
}
|
|
if q.separator != nil {
|
|
suggester["separator"] = *q.separator
|
|
}
|
|
if q.maxErrors != nil {
|
|
suggester["max_errors"] = *q.maxErrors
|
|
}
|
|
if q.gramSize != nil {
|
|
suggester["gram_size"] = *q.gramSize
|
|
}
|
|
if q.forceUnigrams != nil {
|
|
suggester["force_unigrams"] = *q.forceUnigrams
|
|
}
|
|
if q.tokenLimit != nil {
|
|
suggester["token_limit"] = *q.tokenLimit
|
|
}
|
|
if q.generators != nil && len(q.generators) > 0 {
|
|
for typ, generators := range q.generators {
|
|
var arr []interface{}
|
|
for _, g := range generators {
|
|
src, err := g.Source()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
arr = append(arr, src)
|
|
}
|
|
suggester[typ] = arr
|
|
}
|
|
}
|
|
if q.smoothingModel != nil {
|
|
src, err := q.smoothingModel.Source()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
x := make(map[string]interface{})
|
|
x[q.smoothingModel.Type()] = src
|
|
suggester["smoothing"] = x
|
|
}
|
|
if q.preTag != nil {
|
|
hl := make(map[string]string)
|
|
hl["pre_tag"] = *q.preTag
|
|
if q.postTag != nil {
|
|
hl["post_tag"] = *q.postTag
|
|
}
|
|
suggester["highlight"] = hl
|
|
}
|
|
if q.collateQuery != nil {
|
|
collate := make(map[string]interface{})
|
|
suggester["collate"] = collate
|
|
if q.collateQuery != nil {
|
|
collate["query"] = *q.collateQuery
|
|
}
|
|
if q.collatePreference != nil {
|
|
collate["preference"] = *q.collatePreference
|
|
}
|
|
if len(q.collateParams) > 0 {
|
|
collate["params"] = q.collateParams
|
|
}
|
|
if q.collatePrune != nil {
|
|
collate["prune"] = *q.collatePrune
|
|
}
|
|
}
|
|
|
|
if !includeName {
|
|
return ps, nil
|
|
}
|
|
|
|
source := make(map[string]interface{})
|
|
source[q.name] = ps
|
|
return source, nil
|
|
}
|
|
|
|
// -- Smoothing models --
|
|
|
|
type SmoothingModel interface {
|
|
Type() string
|
|
Source() (interface{}, error)
|
|
}
|
|
|
|
// StupidBackoffSmoothingModel implements a stupid backoff smoothing model.
|
|
// See https://www.elastic.co/guide/en/elasticsearch/reference/5.2/search-suggesters-phrase.html#_smoothing_models
|
|
// for details about smoothing models.
|
|
type StupidBackoffSmoothingModel struct {
|
|
discount float64
|
|
}
|
|
|
|
func NewStupidBackoffSmoothingModel(discount float64) *StupidBackoffSmoothingModel {
|
|
return &StupidBackoffSmoothingModel{
|
|
discount: discount,
|
|
}
|
|
}
|
|
|
|
func (sm *StupidBackoffSmoothingModel) Type() string {
|
|
return "stupid_backoff"
|
|
}
|
|
|
|
func (sm *StupidBackoffSmoothingModel) Source() (interface{}, error) {
|
|
source := make(map[string]interface{})
|
|
source["discount"] = sm.discount
|
|
return source, nil
|
|
}
|
|
|
|
// --
|
|
|
|
// LaplaceSmoothingModel implements a laplace smoothing model.
|
|
// See https://www.elastic.co/guide/en/elasticsearch/reference/5.2/search-suggesters-phrase.html#_smoothing_models
|
|
// for details about smoothing models.
|
|
type LaplaceSmoothingModel struct {
|
|
alpha float64
|
|
}
|
|
|
|
func NewLaplaceSmoothingModel(alpha float64) *LaplaceSmoothingModel {
|
|
return &LaplaceSmoothingModel{
|
|
alpha: alpha,
|
|
}
|
|
}
|
|
|
|
func (sm *LaplaceSmoothingModel) Type() string {
|
|
return "laplace"
|
|
}
|
|
|
|
func (sm *LaplaceSmoothingModel) Source() (interface{}, error) {
|
|
source := make(map[string]interface{})
|
|
source["alpha"] = sm.alpha
|
|
return source, nil
|
|
}
|
|
|
|
// --
|
|
|
|
// LinearInterpolationSmoothingModel implements a linear interpolation
|
|
// smoothing model.
|
|
// See https://www.elastic.co/guide/en/elasticsearch/reference/5.2/search-suggesters-phrase.html#_smoothing_models
|
|
// for details about smoothing models.
|
|
type LinearInterpolationSmoothingModel struct {
|
|
trigramLamda float64
|
|
bigramLambda float64
|
|
unigramLambda float64
|
|
}
|
|
|
|
func NewLinearInterpolationSmoothingModel(trigramLamda, bigramLambda, unigramLambda float64) *LinearInterpolationSmoothingModel {
|
|
return &LinearInterpolationSmoothingModel{
|
|
trigramLamda: trigramLamda,
|
|
bigramLambda: bigramLambda,
|
|
unigramLambda: unigramLambda,
|
|
}
|
|
}
|
|
|
|
func (sm *LinearInterpolationSmoothingModel) Type() string {
|
|
return "linear_interpolation"
|
|
}
|
|
|
|
func (sm *LinearInterpolationSmoothingModel) Source() (interface{}, error) {
|
|
source := make(map[string]interface{})
|
|
source["trigram_lambda"] = sm.trigramLamda
|
|
source["bigram_lambda"] = sm.bigramLambda
|
|
source["unigram_lambda"] = sm.unigramLambda
|
|
return source, nil
|
|
}
|
|
|
|
// -- CandidateGenerator --
|
|
|
|
type CandidateGenerator interface {
|
|
Type() string
|
|
Source() (interface{}, error)
|
|
}
|
|
|
|
// DirectCandidateGenerator implements a direct candidate generator.
|
|
// See https://www.elastic.co/guide/en/elasticsearch/reference/5.2/search-suggesters-phrase.html#_smoothing_models
|
|
// for details about smoothing models.
|
|
type DirectCandidateGenerator struct {
|
|
field string
|
|
preFilter *string
|
|
postFilter *string
|
|
suggestMode *string
|
|
accuracy *float64
|
|
size *int
|
|
sort *string
|
|
stringDistance *string
|
|
maxEdits *int
|
|
maxInspections *int
|
|
maxTermFreq *float64
|
|
prefixLength *int
|
|
minWordLength *int
|
|
minDocFreq *float64
|
|
}
|
|
|
|
func NewDirectCandidateGenerator(field string) *DirectCandidateGenerator {
|
|
return &DirectCandidateGenerator{
|
|
field: field,
|
|
}
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) Type() string {
|
|
return "direct_generator"
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) Field(field string) *DirectCandidateGenerator {
|
|
g.field = field
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) PreFilter(preFilter string) *DirectCandidateGenerator {
|
|
g.preFilter = &preFilter
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) PostFilter(postFilter string) *DirectCandidateGenerator {
|
|
g.postFilter = &postFilter
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) SuggestMode(suggestMode string) *DirectCandidateGenerator {
|
|
g.suggestMode = &suggestMode
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) Accuracy(accuracy float64) *DirectCandidateGenerator {
|
|
g.accuracy = &accuracy
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) Size(size int) *DirectCandidateGenerator {
|
|
g.size = &size
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) Sort(sort string) *DirectCandidateGenerator {
|
|
g.sort = &sort
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) StringDistance(stringDistance string) *DirectCandidateGenerator {
|
|
g.stringDistance = &stringDistance
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) MaxEdits(maxEdits int) *DirectCandidateGenerator {
|
|
g.maxEdits = &maxEdits
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) MaxInspections(maxInspections int) *DirectCandidateGenerator {
|
|
g.maxInspections = &maxInspections
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) MaxTermFreq(maxTermFreq float64) *DirectCandidateGenerator {
|
|
g.maxTermFreq = &maxTermFreq
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) PrefixLength(prefixLength int) *DirectCandidateGenerator {
|
|
g.prefixLength = &prefixLength
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) MinWordLength(minWordLength int) *DirectCandidateGenerator {
|
|
g.minWordLength = &minWordLength
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) MinDocFreq(minDocFreq float64) *DirectCandidateGenerator {
|
|
g.minDocFreq = &minDocFreq
|
|
return g
|
|
}
|
|
|
|
func (g *DirectCandidateGenerator) Source() (interface{}, error) {
|
|
source := make(map[string]interface{})
|
|
if g.field != "" {
|
|
source["field"] = g.field
|
|
}
|
|
if g.suggestMode != nil {
|
|
source["suggest_mode"] = *g.suggestMode
|
|
}
|
|
if g.accuracy != nil {
|
|
source["accuracy"] = *g.accuracy
|
|
}
|
|
if g.size != nil {
|
|
source["size"] = *g.size
|
|
}
|
|
if g.sort != nil {
|
|
source["sort"] = *g.sort
|
|
}
|
|
if g.stringDistance != nil {
|
|
source["string_distance"] = *g.stringDistance
|
|
}
|
|
if g.maxEdits != nil {
|
|
source["max_edits"] = *g.maxEdits
|
|
}
|
|
if g.maxInspections != nil {
|
|
source["max_inspections"] = *g.maxInspections
|
|
}
|
|
if g.maxTermFreq != nil {
|
|
source["max_term_freq"] = *g.maxTermFreq
|
|
}
|
|
if g.prefixLength != nil {
|
|
source["prefix_length"] = *g.prefixLength
|
|
}
|
|
if g.minWordLength != nil {
|
|
source["min_word_length"] = *g.minWordLength
|
|
}
|
|
if g.minDocFreq != nil {
|
|
source["min_doc_freq"] = *g.minDocFreq
|
|
}
|
|
if g.preFilter != nil {
|
|
source["pre_filter"] = *g.preFilter
|
|
}
|
|
if g.postFilter != nil {
|
|
source["post_filter"] = *g.postFilter
|
|
}
|
|
return source, nil
|
|
}
|