Documentation ¶
Index ¶
- func ConfigFromModel(iicm *models.InvertedIndexConfig) schema.InvertedIndexConfig
- func CopyIntoMap(a, b map[string]interface{}) map[string]interface{}
- func HasFrequency(dt schema.DataType) bool
- func LexicographicallySortableFloat64(in float64) ([]byte, error)
- func LexicographicallySortableInt64(in int64) ([]byte, error)
- func LexicographicallySortableUint64(in uint64) ([]byte, error)
- func ParseLexicographicallySortableFloat64(in []byte) (float64, error)
- func ParseLexicographicallySortableInt64(in []byte) (int64, error)
- func ParseLexicographicallySortableUint64(in []byte) (uint64, error)
- func ValidateConfig(conf *models.InvertedIndexConfig) error
- func ValidateUserConfigUpdate(initial, updated *models.InvertedIndexConfig) error
- type Analyzer
- func (a *Analyzer) Bool(in bool) ([]Countable, error)
- func (a *Analyzer) BoolArray(in []bool) ([]Countable, error)
- func (a *Analyzer) Float(in float64) ([]Countable, error)
- func (a *Analyzer) FloatArray(in []float64) ([]Countable, error)
- func (a *Analyzer) Int(in int64) ([]Countable, error)
- func (a *Analyzer) IntArray(in []int64) ([]Countable, error)
- func (a *Analyzer) Object(input map[string]interface{}, props []*models.Property, uuid strfmt.UUID) ([]Property, error)
- func (a *Analyzer) Ref(in models.MultipleRef) ([]Countable, error)
- func (a *Analyzer) RefCount(in models.MultipleRef) ([]Countable, error)
- func (a *Analyzer) String(tokenization, in string) []Countable
- func (a *Analyzer) StringArray(tokenization string, in []string) []Countable
- func (a *Analyzer) Text(tokenization, in string) []Countable
- func (a *Analyzer) TextArray(tokenization string, in []string) []Countable
- type BM25Searcher
- type CacheEntry
- type CacheEntryType
- type ClassSearcher
- type Cleaner
- type Countable
- type DeletedDocIDChecker
- type DeltaMergeResult
- type DeltaMerger
- type DeltaResults
- type MergeDocIDWithFrequency
- type MergeItem
- type MergeProperty
- type Property
- type PropertyLengthTracker
- func (t *PropertyLengthTracker) Close() error
- func (t *PropertyLengthTracker) Drop() error
- func (t *PropertyLengthTracker) FileName() string
- func (t *PropertyLengthTracker) Flush() error
- func (t *PropertyLengthTracker) PropertyMean(propName string) (float32, error)
- func (t *PropertyLengthTracker) TrackProperty(propName string, value float32)
- type ReadFn
- type ReadFnFrequency
- type RowCacher
- type RowReader
- type RowReaderFrequency
- type Searcher
- func (s *Searcher) DocIDs(ctx context.Context, filter *filters.LocalFilter, ...) (helpers.AllowList, error)
- func (s *Searcher) DocIDsPreventCaching(ctx context.Context, filter *filters.LocalFilter, ...) (helpers.AllowList, error)
- func (s *Searcher) Objects(ctx context.Context, limit int, filter *filters.LocalFilter, ...) ([]*storobj.Object, error)
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func ConfigFromModel ¶ added in v1.12.0
func ConfigFromModel(iicm *models.InvertedIndexConfig) schema.InvertedIndexConfig
func CopyIntoMap ¶ added in v1.17.0
func HasFrequency ¶ added in v1.5.0
func LexicographicallySortableFloat64 ¶
LexicographicallySortableFloat64 transforms a conversion to a lexicographically sortable byte slice. In general, for lexicographical sorting big endian notatino is required. Additionally the sign needs to be flipped in any case, but additionally each remaining byte also needs to be flipped if the number is negative
func LexicographicallySortableInt64 ¶
LexicographicallySortableInt64 performs a conversion to a lexicographically sortable byte slice. For this, big endian notation is required and the sign must be flipped
func LexicographicallySortableUint64 ¶ added in v1.0.0
LexicographicallySortableUint64 performs a conversion to a lexicographically sortable byte slice. For this, big endian notation is required.
func ParseLexicographicallySortableFloat64 ¶ added in v0.22.19
ParseLexicographicallySortableFloat64 reverses the changes in LexicographicallySortableFloat64
func ParseLexicographicallySortableInt64 ¶ added in v0.22.19
ParseLexicographicallySortableInt64 reverses the changes in LexicographicallySortableInt64
func ParseLexicographicallySortableUint64 ¶ added in v1.0.0
ParseLexicographicallySortableUint64 reverses the changes in LexicographicallySortableUint64
func ValidateConfig ¶ added in v1.12.0
func ValidateConfig(conf *models.InvertedIndexConfig) error
func ValidateUserConfigUpdate ¶ added in v1.12.0
func ValidateUserConfigUpdate(initial, updated *models.InvertedIndexConfig) error
Types ¶
type Analyzer ¶
type Analyzer struct {
// contains filtered or unexported fields
}
func NewAnalyzer ¶
func NewAnalyzer(stopwords stopwords.StopwordDetector) *Analyzer
func (*Analyzer) Bool ¶
Bool requires no analysis, so it's actually just a simple conversion to a little-endian ordered byte slice
func (*Analyzer) BoolArray ¶ added in v1.7.2
BoolArray requires no analysis, so it's actually just a simple conversion to a little-endian ordered byte slice
func (*Analyzer) Float ¶
Float requires no analysis, so it's actually just a simple conversion to a lexicographically sortable byte slice.
func (*Analyzer) FloatArray ¶ added in v1.7.0
Float array requires no analysis, so it's actually just a simple conversion to a lexicographically sortable byte slice.
func (*Analyzer) Int ¶
Int requires no analysis, so it's actually just a simple conversion to a string-formatted byte slice of the int
func (*Analyzer) IntArray ¶ added in v1.7.0
Int array requires no analysis, so it's actually just a simple conversion to a string-formatted byte slice of the int
func (*Analyzer) Ref ¶ added in v0.22.20
func (a *Analyzer) Ref(in models.MultipleRef) ([]Countable, error)
Ref indexes references as beacon-strings
func (*Analyzer) RefCount ¶
func (a *Analyzer) RefCount(in models.MultipleRef) ([]Countable, error)
RefCount does not index the content of the refs, but only the count with 0 being an explicitly allowed value as well.
func (*Analyzer) String ¶
String splits only on spaces and does not lowercase, then aggregates duplicates
func (*Analyzer) StringArray ¶ added in v1.12.0
StringArray splits only on spaces and does not lowercase, then aggregates duplicates
type BM25Searcher ¶ added in v1.11.0
type BM25Searcher struct {
// contains filtered or unexported fields
}
func NewBM25Searcher ¶ added in v1.11.0
func NewBM25Searcher(config schema.BM25Config, store *lsmkv.Store, schema schema.Schema, rowCache cacher, propIndices propertyspecific.Indices, classSearcher ClassSearcher, deletedDocIDs DeletedDocIDChecker, propLengths propLengthRetriever, logger logrus.FieldLogger, shardVersion uint16, ) *BM25Searcher
func (*BM25Searcher) BM25F ¶ added in v1.17.0
func (b *BM25Searcher) BM25F(ctx context.Context, className schema.ClassName, limit int, keywordRanking *searchparams.KeywordRanking, filter *filters.LocalFilter, sort []filters.Sort, additional additional.Properties, objectByIndexID func(index uint64) *storobj.Object, ) ([]*storobj.Object, []float32, error)
func (*BM25Searcher) Objects ¶ added in v1.17.1
func (b *BM25Searcher) Objects(ctx context.Context, limit int, keywordRanking *searchparams.KeywordRanking, filter *filters.LocalFilter, sort []filters.Sort, additional additional.Properties, className schema.ClassName, ) ([]*storobj.Object, []float32, error)
Objects returns a list of full objects
type CacheEntry ¶ added in v1.8.0
type CacheEntry struct { Type CacheEntryType Hash []byte Partial *docPointers AllowList helpers.AllowList }
func (*CacheEntry) Size ¶ added in v1.8.0
func (ce *CacheEntry) Size() uint64
Size cannot be determined accurately since a golang map does not have fixed size per elements. However, through experimentation we have found that a map[uint64]struct{} rarely exceeds 25 bytes per entry, so we are using this as an estimate. In addition, we know that the partial content uses an array where we can assume full efficiency, i.e. 8 bytes per entry.
type CacheEntryType ¶ added in v1.8.0
type CacheEntryType uint8
const ( CacheTypePartial CacheEntryType = iota CacheTypeAllowList )
func (CacheEntryType) String ¶ added in v1.8.0
func (t CacheEntryType) String() string
type ClassSearcher ¶ added in v0.22.20
type ClassSearcher interface { ClassSearch(ctx context.Context, params traverser.GetParams) ([]search.Result, error) GetQueryMaximumResults() int }
ClassSearcher is anything that allows a root-level ClassSearch
type Cleaner ¶ added in v1.0.0
type Cleaner struct {
// contains filtered or unexported fields
}
func NewCleaner ¶ added in v1.0.0
type DeletedDocIDChecker ¶ added in v0.22.20
type DeltaMergeResult ¶ added in v1.1.0
type DeltaMergeResult struct { Additions []MergeProperty Deletions []MergeProperty }
type DeltaMerger ¶ added in v1.1.0
type DeltaMerger struct {
// contains filtered or unexported fields
}
DeltaMerger can be used to condense the number of single writes into one big one. Additionally it removes overlaps between additions and deletions. It is meant to be used in batch situation, where 5 ref objects in a row might each increase the doc count by one. Instead of writing 5 additions and 4 deletions, this can be condensed to write just one addition
func NewDeltaMerger ¶ added in v1.1.0
func NewDeltaMerger() *DeltaMerger
func (*DeltaMerger) AddAdditions ¶ added in v1.1.0
func (dm *DeltaMerger) AddAdditions(props []Property, docID uint64)
func (*DeltaMerger) AddDeletions ¶ added in v1.1.0
func (dm *DeltaMerger) AddDeletions(props []Property, docID uint64)
func (*DeltaMerger) Merge ¶ added in v1.1.0
func (dm *DeltaMerger) Merge() DeltaMergeResult
type DeltaResults ¶
func Delta ¶
func Delta(previous, next []Property) DeltaResults
type MergeDocIDWithFrequency ¶ added in v1.1.0
type MergeItem ¶ added in v1.1.0
type MergeItem struct { Data []byte DocIDs []MergeDocIDWithFrequency }
type MergeProperty ¶ added in v1.1.0
type PropertyLengthTracker ¶ added in v1.11.0
Page Design | Bytes | Description | | --------- | ------------------------------------------------ | | start | page is now 0 | 0-1 | uint16 pointer to last index byte | 2-3 | uint16 pointer for property name length | 4-n | property name | ... | repeat length+pointer pattern | 3584-3840 | second property buckets (64 buckets of float32) | 3840-4096 | first property buckets | repeat | page is now 1, repeat all of above
Fixed Assumptions:
- First two bytes always used to indicate end of index, minimal value is 02, as the first possible value with index length=0 is after the two bytes themselves.
- 64 buckets of float32 per property (=256B per prop), excluding the index
- One index row is always 4+len(propName), consisting of a uint16 prop name length pointer, the name itself and an offset pointer pointing to the start (first byte) of the buckets
func NewPropertyLengthTracker ¶ added in v1.11.0
func NewPropertyLengthTracker(path string) (*PropertyLengthTracker, error)
func (*PropertyLengthTracker) Close ¶ added in v1.11.0
func (t *PropertyLengthTracker) Close() error
func (*PropertyLengthTracker) Drop ¶ added in v1.11.0
func (t *PropertyLengthTracker) Drop() error
func (*PropertyLengthTracker) FileName ¶ added in v1.15.0
func (t *PropertyLengthTracker) FileName() string
func (*PropertyLengthTracker) Flush ¶ added in v1.11.0
func (t *PropertyLengthTracker) Flush() error
func (*PropertyLengthTracker) PropertyMean ¶ added in v1.11.0
func (t *PropertyLengthTracker) PropertyMean(propName string) (float32, error)
func (*PropertyLengthTracker) TrackProperty ¶ added in v1.11.0
func (t *PropertyLengthTracker) TrackProperty(propName string, value float32, )
type ReadFn ¶ added in v0.22.19
ReadFn will be called 1..n times per match. This means it will also be called on a non-match, in this case v == nil. It is up to the caller to decide if that is an error case or not.
Note that because what we are parsing is an inverted index row, it can sometimes become confusing what a key and value actually resembles. The variables k and v are the literal row key and value. So this means, the data-value as in "less than 17" where 17 would be the "value" is in the key variable "k". The value will contain the docCount, hash and list of pointers (with optional frequency) to the docIDs
The boolean return argument is a way to stop iteration (e.g. when a limit is reached) without producing an error. In normal operation always return true, if false is returned once, the loop is broken.
type ReadFnFrequency ¶ added in v1.5.0
ReadFnFrequency will be called 1..n times per match. This means it will also be called on a non-match, in this case v == nil. It is up to the caller to decide if that is an error case or not.
Note that because what we are parsing is an inverted index row, it can sometimes become confusing what a key and value actually resembles. The variables k and v are the literal row key and value. So this means, the data-value as in "less than 17" where 17 would be the "value" is in the key variable "k". The value will contain the docCount, hash and list of pointers (with optional frequency) to the docIDs
The boolean return argument is a way to stop iteration (e.g. when a limit is reached) without producing an error. In normal operation always return true, if false is returned once, the loop is broken.
type RowCacher ¶
type RowCacher struct {
// contains filtered or unexported fields
}
func NewRowCacher ¶
func (*RowCacher) Store ¶
func (rc *RowCacher) Store(id []byte, row *CacheEntry)
type RowReader ¶ added in v0.22.19
type RowReader struct {
// contains filtered or unexported fields
}
RowReader reads one or many row(s) depending on the specified operator
type RowReaderFrequency ¶ added in v1.5.0
type RowReaderFrequency struct {
// contains filtered or unexported fields
}
RowReaderFrequency reads one or many row(s) depending on the specified operator
func NewRowReaderFrequency ¶ added in v1.5.0
func (*RowReaderFrequency) Read ¶ added in v1.5.0
func (rr *RowReaderFrequency) Read(ctx context.Context, readFn ReadFnFrequency) error
type Searcher ¶
type Searcher struct {
// contains filtered or unexported fields
}
func NewSearcher ¶
func NewSearcher(store *lsmkv.Store, schema schema.Schema, rowCache cacher, propIndices propertyspecific.Indices, classSearcher ClassSearcher, deletedDocIDs DeletedDocIDChecker, stopwords stopwords.StopwordDetector, shardVersion uint16, ) *Searcher
func (*Searcher) DocIDs ¶
func (s *Searcher) DocIDs(ctx context.Context, filter *filters.LocalFilter, additional additional.Properties, className schema.ClassName, ) (helpers.AllowList, error)
DocIDs is similar to Objects, but does not actually resolve the docIDs to full objects. Instead it returns the pure object id pointers. They can then be used in a secondary index (e.g. vector index)
DocID queries does not contain a limit by design, as we won't know if the limit wouldn't remove the item that is most important for the follow up query. Imagine the user sets the limit to 1 and the follow-up is a vector search. If we already limited the allowList to 1, the vector search would be pointless, as only the first element would be allowed, regardless of which had the shortest distance
func (*Searcher) DocIDsPreventCaching ¶ added in v1.14.0
func (s *Searcher) DocIDsPreventCaching(ctx context.Context, filter *filters.LocalFilter, additional additional.Properties, className schema.ClassName, ) (helpers.AllowList, error)
DocIDsPreventCaching is the same as DocIDs, but makes sure that no filter cache entries are written. This can be used when we can guarantee that the filter is part of an operation that will lead to a state change, such as batch delete. The state change would make the cached filter unusable anyway, so we don't need to unnecessarily populate the cache with an entry.
Source Files ¶
- analyzer.go
- bm25_searcher.go
- cleaner.go
- config.go
- config_update.go
- delta_analyzer.go
- delta_merger.go
- like_regexp.go
- merge.go
- objects.go
- prop_length_tracker.go
- prop_value_pairs.go
- prop_value_pairs_hashes.go
- row_cacher.go
- row_reader.go
- row_reader_frequency.go
- score_merger.go
- searcher.go
- searcher_doc_pointers.go
- searcher_ref_filter.go
- searcher_value_extractors.go
- serialization.go