common

package
v1.36.21 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jul 8, 2026 License: BSD-3-Clause Imports: 16 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	IndexTypeHNSW    = "hnsw"
	IndexTypeFlat    = "flat"
	IndexTypeNoop    = "noop"
	IndexTypeDynamic = "dynamic"
	IndexTypeHFresh  = "hfresh"
)
View Source 
const (
	// DefaultSearchByDistInitialLimit :
	// the initial limit of 100 here is an
	// arbitrary decision, and can be tuned
	// as needed
	DefaultSearchByDistInitialLimit = 100

	// DefaultSearchByDistLimitMultiplier :
	// the decision to increase the limit in
	// multiples of 10 here is an arbitrary
	// decision, and can be tuned as needed
	DefaultSearchByDistLimitMultiplier = 10
)
View Source 
const (
	DefaultShardedLocksCount = 512
	DefaultPageSize          = 1
)

Variables

View Source 
var ErrWrongDimensions = errors.New("vector dimensions do not match the index dimensions")

Functions

func AddVectorsToIndex added in v1.29.0 

func AddVectorsToIndex(ctx context.Context, vectors []VectorRecord, vectorIndex VectorIndex) error

func CalculateOptimalSegments added in v1.25.23 

func CalculateOptimalSegments(dims int) int

func IsDynamic added in v1.32.1 

func IsDynamic(indexType IndexType) bool

func MultiVectorsEqual added in v1.29.0 

func MultiVectorsEqual(vecA, vecB [][]float32) bool

func VectorsEqual added in v1.24.0 

func VectorsEqual(vecA, vecB []float32) bool

Types

type BoltStore added in v1.35.0 

type BoltStore struct {
	// contains filtered or unexported fields
}

BoltStore is a SequenceStore implementation that uses BoltDB as the backend.

func NewBoltStore added in v1.35.0 

func NewBoltStore(db *bolt.DB, bucket, key []byte) *BoltStore

func (*BoltStore) Load added in v1.35.0 

func (s *BoltStore) Load() (uint64, error)

func (*BoltStore) Store added in v1.35.0 

func (s *BoltStore) Store(upperBound uint64) error

type BucketView added in v1.34.11 

type BucketView interface {
	ReleaseView()
}

BucketView represents a consistent view of an LSM bucket that can be used for multiple reads without acquiring locks for each read. The caller must call ReleaseView() when done to avoid blocking compactions.

type FS added in v1.34.0 

type FS interface {
	Open(name string) (File, error)
	OpenFile(name string, flag int, perm os.FileMode) (File, error)
	Create(name string) (File, error)
	MkdirAll(path string, perm os.FileMode) error
	ReadDir(name string) ([]os.DirEntry, error)
	Stat(name string) (os.FileInfo, error)
	Remove(name string) error
	RemoveAll(path string) error
	Rename(oldpath, newpath string) error
	Truncate(name string, size int64) error
}

func NewOSFS added in v1.34.0 

func NewOSFS() FS

type File added in v1.34.0 

type File interface {
	io.Reader
	io.Writer
	io.Closer
	io.ReaderAt
	io.Seeker
	Sync() error
	Stat() (os.FileInfo, error)
}

type GetViewThunk added in v1.34.11 

type GetViewThunk func() BucketView

GetViewThunk returns a consistent view of the underlying bucket.

type GroupedPagedArray added in v1.36.0 

type GroupedPagedArray[T any] struct {
	// contains filtered or unexported fields
}

GroupedPagedArray is a two-level paged array optimized for concurrent access

Level 1: a slice of atomic pointers to groups (allocated upfront, small). Level 2: each group holds 128 atomic pointers to pages (allocated on demand). Pages: fixed-size slices of T (allocated on demand).

Read path is lock-free (two atomic loads). Write path uses a mutex only for allocating new groups/pages via double-checked locking.

Caller is responsible for ensuring that reads and writes to elements within returned pages are performed safely using atomic operations or mutexes.

func NewGroupedPagedArray added in v1.36.0 

func NewGroupedPagedArray[T any](maxPages, pageSize uint64) *GroupedPagedArray[T]

NewGroupedPagedArray creates a new GroupedPagedArray.

maxPages is the maximum number of pages that can be addressed. pageSize is the number of elements per page (rounded up to a power of 2, minimum 64).

Total addressable elements = maxPages × pageSize.

The upfront cost is len(groups) × 8 bytes, where len(groups) = ceil(maxPages / 128). For example, 16K max pages costs only 1KB upfront.

func (*GroupedPagedArray[T]) EnsurePageFor added in v1.36.0 

func (p *GroupedPagedArray[T]) EnsurePageFor(id uint64) ([]T, int)

EnsurePageFor ensures that a page exists for the given ID, allocating the group and/or page if necessary. Returns the page and the index within the page.

Panics if the ID exceeds the maximum addressable capacity.

func (*GroupedPagedArray[T]) GetPageFor added in v1.36.0 

func (p *GroupedPagedArray[T]) GetPageFor(id uint64) ([]T, int)

GetPageFor returns the page containing the given ID and the index within that page. Returns (nil, -1) if the page has not been allocated.

Lock-free: performs two atomic loads.

type IndexStats added in v1.24.23 

type IndexStats interface {
	IndexType() IndexType
}

type IndexType added in v1.24.23 

type IndexType string

func (IndexType) String added in v1.32.1 

func (i IndexType) String() string

type Iterator added in v1.27.0 

type Iterator interface {
	Next() (int, error)
	IsDone() bool
}

Iterator interface defines the methods for sampling elements.

type MonotonicCounter added in v1.34.0 

type MonotonicCounter struct {
	// contains filtered or unexported fields
}

MonotonicCounter is a thread-safe counter that increments monotonically. It is a simple wrapper to avoid accidental resets, decrements, overflow, or other operations that could break the monotonicity. The zero value is a valid counter that starts at 0.

func NewMonotonicCounter added in v1.34.0 

func NewMonotonicCounter(start uint64) *MonotonicCounter

NewMonotonicCounter creates a new MonotonicCounter with the given starting value. Next call to Next() will return the starting value + 1. For a counter that starts at 0, you can use the zero value directly.

func (*MonotonicCounter) Next added in v1.34.0 

func (c *MonotonicCounter) Next() uint64

Next returns the next value of the counter. If the counter has hit its maximum value, it panics.

func (*MonotonicCounter) NextN added in v1.34.0 

func (c *MonotonicCounter) NextN(n uint64) (start uint64, end uint64)

NextN returns the next n values as a range: [start, end]. If n is 0, it returns (0, 0). Panics if the range would overflow.

func (*MonotonicCounter) TryNext added in v1.34.0 

func (c *MonotonicCounter) TryNext() (uint64, bool)

TryNext returns (next, ok). When ok == false the counter has hit its maximum and next == zero(T).

func (*MonotonicCounter) TryNextN added in v1.34.0 

func (c *MonotonicCounter) TryNextN(n uint64) (start, end uint64, ok bool)

TryNextN returns the next n values as a range [start, end]. If n is 0, it returns (0, 0, true). If the counter has hit its maximum value or the range would overflow, it returns (0, 0, false).

type MultiVectorForID 

type MultiVectorForID func(ctx context.Context, ids []uint64) ([][]float32, []error)

type MultipleVectorForID added in v1.29.0 

type MultipleVectorForID[T float32 | uint64 | byte] func(ctx context.Context, id uint64, relativeID uint64) ([]T, error)

type PagedArray added in v1.34.0 

type PagedArray[T any] struct {
	// contains filtered or unexported fields
}

PagedArray is an array that stores elements in pages of a fixed size. It is optimized for concurrent access patterns where multiple goroutines may read and write to different pages simultaneously. The API is thread-safe and returns direct references to the pages, allowing for efficient concurrent access. Caller is responsible for ensuring that reads and writes to the returned pages are performed safely, using either atomic operations or mutexes.

func NewPagedArray added in v1.34.0 

func NewPagedArray[T any](pages, pageSize uint64) *PagedArray[T]

NewPagedArray creates a new PagedArray with the given page size. It will round up to the next power of 2 and enforce a minimum size of 64.

func (*PagedArray[T]) EnsurePageFor added in v1.34.0 

func (p *PagedArray[T]) EnsurePageFor(id uint64) ([]T, int)

EnsurePageFor makes sure that a page exists for the given ID. If the page already exists, it does nothing. It returns the page and its index within the page for the given ID.

func (*PagedArray[T]) GetPageFor added in v1.34.0 

func (p *PagedArray[T]) GetPageFor(id uint64) ([]T, int)

GetPageFor takes an ID and returns the associated page and its index. If the page does not exist, it returns nil. It doesn't return a copy of the page, so modifications to the returned slice will affect the original data. Caller is responsible for ensuring that reads and writes to the returned page are performed safely, using either atomic operations or mutexes.

type PqMaxPool 

type PqMaxPool struct {
	// contains filtered or unexported fields
}

func NewPqMaxPool 

func NewPqMaxPool(defaultCap int) *PqMaxPool

func (*PqMaxPool) GetMax 

func (pqh *PqMaxPool) GetMax(capacity int) *priorityqueue.Queue[any]

func (*PqMaxPool) Put 

func (pqh *PqMaxPool) Put(pq *priorityqueue.Queue[any])

type QueryVectorDistancer added in v1.26.0 

type QueryVectorDistancer struct {
	DistanceFunc func(uint64) (float32, error)
	CloseFunc    func()
}

func (*QueryVectorDistancer) Close added in v1.26.0 

func (q *QueryVectorDistancer) Close()

func (*QueryVectorDistancer) DistanceToNode added in v1.26.0 

func (q *QueryVectorDistancer) DistanceToNode(nodeID uint64) (float32, error)

type SearchByDistParams 

type SearchByDistParams struct {
	// contains filtered or unexported fields
}

func NewSearchByDistParams 

func NewSearchByDistParams(
	offset int,
	limit int,
	totalLimit int,
	maximumSearchLimit int64,
) *SearchByDistParams

func (*SearchByDistParams) Iterate 

func (params *SearchByDistParams) Iterate()

func (*SearchByDistParams) MaxLimitReached 

func (params *SearchByDistParams) MaxLimitReached() bool

func (*SearchByDistParams) MaximumSearchLimit 

func (params *SearchByDistParams) MaximumSearchLimit() int64

func (*SearchByDistParams) OffsetCapacity 

func (params *SearchByDistParams) OffsetCapacity(ids []uint64) int

func (*SearchByDistParams) TotalLimit 

func (params *SearchByDistParams) TotalLimit() int

func (*SearchByDistParams) TotalLimitCapacity 

func (params *SearchByDistParams) TotalLimitCapacity(ids []uint64) int

type Sequence added in v1.35.0 

type Sequence struct {
	// contains filtered or unexported fields
}

Sequence represents a monotonic uint64 generator that can be used to generate unique incrementing ids for stored postings and other entities. It is inspired by Postgres Sequences, although much simpler and it's designed to ensure the following properties: 1. Monotonicity: Each generated id is guaranteed to be greater than the previously generated id. 2. Persistence: The state of the generator is persisted to disk. 3. Low disk I/O: The generator is designed to minimize the number of disk writes by allocating ranges of ids in memory. This allows it to serve most requests from memory without touching the disk. 4. Concurrency: The generator is safe for concurrent access. However because of its design, it does not guarantee gapless ids. A Sequence first allocates a range of ids and reserves them in the persistent storage. It then serves those ids from memory without touching the disk. If the sequence is closed gracefully, if persists the last used id, however in case of a crash, the sequence will start from the upper bound of the reserved range potentially leaving gaps. The range is configurable and can be tuned based on the expected throughput requirements.

func NewSequence added in v1.35.0 

func NewSequence(store SequenceStore, rangeSize uint64) (*Sequence, error)

NewSequence loads the upper bound from the store and returns a ready to use Sequence.

func (*Sequence) Flush added in v1.35.0 

func (s *Sequence) Flush() error

Flush persists the last used id to the store, to be used as the next starting point when the sequence is re-opened. This must be called to gracefully close the sequence and avoid gaps after a restart. Callers need to ensure no concurrent calls to Next() are happening while Flush() is in progress.

func (*Sequence) Next added in v1.35.0 

func (s *Sequence) Next() (uint64, error)

Next returns the next value in the sequence. Most of the time it will be served from memory without touching the disk.

type SequenceStore added in v1.35.0 

type SequenceStore interface {
	Store(upperBound uint64) error
	Load() (uint64, error)
}

SequenceStore defines the interface for persisting the state of a Sequence. Implementations don't need to be thread-safe.

type ShardedLocks 

type ShardedLocks struct {
	PageSize uint64
	// contains filtered or unexported fields
}

func NewDefaultShardedLocks 

func NewDefaultShardedLocks() *ShardedLocks

func NewShardedLocks 

func NewShardedLocks(count uint64) *ShardedLocks

func NewShardedLocksWithPageSize added in v1.28.0 

func NewShardedLocksWithPageSize(pageSize uint64) *ShardedLocks

func (*ShardedLocks) Hash added in v1.34.0 

func (sl *ShardedLocks) Hash(id uint64) uint64

func (*ShardedLocks) Lock 

func (sl *ShardedLocks) Lock(id uint64)

func (*ShardedLocks) LockAll 

func (sl *ShardedLocks) LockAll()

func (*ShardedLocks) Locked 

func (sl *ShardedLocks) Locked(id uint64, callback func())

func (*ShardedLocks) LockedAll 

func (sl *ShardedLocks) LockedAll(callback func())

func (*ShardedLocks) Unlock 

func (sl *ShardedLocks) Unlock(id uint64)

func (*ShardedLocks) UnlockAll 

func (sl *ShardedLocks) UnlockAll()

type ShardedRWLocks added in v1.24.0 

type ShardedRWLocks struct {
	PageSize uint64
	// contains filtered or unexported fields
}

func NewDefaultShardedRWLocks added in v1.24.0 

func NewDefaultShardedRWLocks() *ShardedRWLocks

func NewShardedRWLocks added in v1.24.0 

func NewShardedRWLocks(count uint64) *ShardedRWLocks

func NewShardedRWLocksWith added in v1.34.0 

func NewShardedRWLocksWith(pages, pageSize uint64) *ShardedRWLocks

func NewShardedRWLocksWithPageSize added in v1.28.0 

func NewShardedRWLocksWithPageSize(pageSize uint64) *ShardedRWLocks

func (*ShardedRWLocks) Hash added in v1.34.0 

func (sl *ShardedRWLocks) Hash(id uint64) uint64

func (*ShardedRWLocks) Lock added in v1.24.0 

func (sl *ShardedRWLocks) Lock(id uint64)

func (*ShardedRWLocks) LockAll added in v1.24.0 

func (sl *ShardedRWLocks) LockAll()

func (*ShardedRWLocks) Locked added in v1.24.0 

func (sl *ShardedRWLocks) Locked(id uint64, callback func())

func (*ShardedRWLocks) LockedAll added in v1.24.0 

func (sl *ShardedRWLocks) LockedAll(callback func())

func (*ShardedRWLocks) RLock added in v1.24.0 

func (sl *ShardedRWLocks) RLock(id uint64)

func (*ShardedRWLocks) RLockAll added in v1.24.0 

func (sl *ShardedRWLocks) RLockAll()

func (*ShardedRWLocks) RLocked added in v1.24.0 

func (sl *ShardedRWLocks) RLocked(id uint64, callback func())

func (*ShardedRWLocks) RLockedAll added in v1.24.0 

func (sl *ShardedRWLocks) RLockedAll(callback func())

func (*ShardedRWLocks) RUnlock added in v1.24.0 

func (sl *ShardedRWLocks) RUnlock(id uint64)

func (*ShardedRWLocks) RUnlockAll added in v1.24.0 

func (sl *ShardedRWLocks) RUnlockAll()

func (*ShardedRWLocks) TryLock added in v1.34.0 

func (sl *ShardedRWLocks) TryLock(id uint64) bool

func (*ShardedRWLocks) Unlock added in v1.24.0 

func (sl *ShardedRWLocks) Unlock(id uint64)

func (*ShardedRWLocks) UnlockAll added in v1.24.0 

func (sl *ShardedRWLocks) UnlockAll()

type SharedGauge added in v1.24.22 

type SharedGauge struct {
	// contains filtered or unexported fields
}

SharedGauge is a thread-safe gauge that can be shared between multiple goroutines. It is used to track the number of running tasks, and allows to wait until all tasks are done.

func NewSharedGauge added in v1.24.22 

func NewSharedGauge() *SharedGauge

func (*SharedGauge) Count added in v1.24.22 

func (sc *SharedGauge) Count() int64

func (*SharedGauge) Decr added in v1.24.22 

func (sc *SharedGauge) Decr() int64

Decr decrements the gauge and returns the new count.

func (*SharedGauge) Incr added in v1.24.22 

func (sc *SharedGauge) Incr() int64

Incr increments the gauge and returns the new count.

func (*SharedGauge) Wait added in v1.24.22 

func (sc *SharedGauge) Wait(ctx context.Context) error

Wait blocks until the count reaches zero or the context is cancelled. Returns ctx.Err() if the context was cancelled before the count reached zero.

type SparseFisherYatesIterator added in v1.27.0 

type SparseFisherYatesIterator struct {
	// contains filtered or unexported fields
}

SparseFisherYatesIterator implements the Iterator interface using the Sparse Fisher-Yates algorithm.

func NewSparseFisherYatesIterator added in v1.27.0 

func NewSparseFisherYatesIterator(size int) *SparseFisherYatesIterator

NewSparseFisherYatesIterator creates a new SparseFisherYatesIterator with the given size.

func (*SparseFisherYatesIterator) IsDone added in v1.27.0 

func (s *SparseFisherYatesIterator) IsDone() bool

IsDone checks if all elements have been sampled.

func (*SparseFisherYatesIterator) Next added in v1.27.0 

func (s *SparseFisherYatesIterator) Next() *int

Next returns the next sampled index using the Sparse Fisher-Yates algorithm.

type TargetTempVectorForID added in v1.24.11 

type TargetTempVectorForID[T []float32 | float32] struct {
	TargetVector         string
	TempVectorForIDThunk func(ctx context.Context, id uint64, container *VectorSlice, targetVector string) ([]T, error)
}

func (TargetTempVectorForID[T]) TempVectorForID added in v1.24.11 

func (t TargetTempVectorForID[T]) TempVectorForID(ctx context.Context, id uint64, container *VectorSlice) ([]T, error)

type TargetTempVectorForIDWithView added in v1.34.11 

type TargetTempVectorForIDWithView[T []float32 | float32] struct {
	TargetVector                 string
	TempVectorForIDWithViewThunk func(ctx context.Context, id uint64, container *VectorSlice, targetVector string, view BucketView) ([]T, error)
}

TargetTempVectorForIDWithView wraps a view-aware vector thunk with a target vector name.

func (TargetTempVectorForIDWithView[T]) TempVectorForIDWithView added in v1.34.11 

func (t TargetTempVectorForIDWithView[T]) TempVectorForIDWithView(ctx context.Context, id uint64, container *VectorSlice, view BucketView) ([]T, error)

TempVectorForIDWithView returns the view-aware vector lookup function.

type TargetVectorForID added in v1.24.11 

type TargetVectorForID[T []float32 | float32 | byte | uint64] struct {
	TargetVector     string
	VectorForIDThunk func(ctx context.Context, id uint64, targetVector string) ([]T, error)
}

func (TargetVectorForID[T]) VectorForID added in v1.24.11 

func (t TargetVectorForID[T]) VectorForID(ctx context.Context, id uint64) ([]T, error)

type TempVectorForID 

type TempVectorForID[T []float32 | float32] func(ctx context.Context, id uint64, container *VectorSlice) ([]T, error)

type TempVectorForIDWithView added in v1.34.11 

type TempVectorForIDWithView[T []float32 | float32] func(ctx context.Context, id uint64, container *VectorSlice, view BucketView) ([]T, error)

TempVectorForIDWithView is like TempVectorForID but uses an existing bucket view.

type TempVectorUint64Pool added in v1.27.0 

type TempVectorUint64Pool struct {
	// contains filtered or unexported fields
}

func NewTempUint64VectorsPool added in v1.27.0 

func NewTempUint64VectorsPool() *TempVectorUint64Pool

func (*TempVectorUint64Pool) Get added in v1.27.0 

func (pool *TempVectorUint64Pool) Get(capacity int) *VectorUint64Slice

func (*TempVectorUint64Pool) Put added in v1.27.0 

func (pool *TempVectorUint64Pool) Put(container *VectorUint64Slice)

type TempVectorsPool 

type TempVectorsPool struct {
	// contains filtered or unexported fields
}

func NewTempVectorsPool 

func NewTempVectorsPool() *TempVectorsPool

func (*TempVectorsPool) Get 

func (pool *TempVectorsPool) Get(capacity int) *VectorSlice

func (*TempVectorsPool) Put 

func (pool *TempVectorsPool) Put(container *VectorSlice)

type TestFS added in v1.34.0 

type TestFS struct {
	FS
	OnOpenFile func(f File) File
	OnOpen     func(f File) File
	OnCreate   func(f File) File
	OnRename   func(oldpath, newpath string) error
	OnRemove   func(name string) error
}

func NewTestFS added in v1.34.0 

func NewTestFS() *TestFS

func (*TestFS) Create added in v1.34.0 

func (fs *TestFS) Create(name string) (File, error)

func (*TestFS) Open added in v1.34.0 

func (fs *TestFS) Open(name string) (File, error)

func (*TestFS) OpenFile added in v1.34.0 

func (fs *TestFS) OpenFile(name string, flag int, perm os.FileMode) (File, error)

func (*TestFS) ReadDir added in v1.34.0 

func (fs *TestFS) ReadDir(name string) ([]os.DirEntry, error)

func (*TestFS) Remove added in v1.34.0 

func (fs *TestFS) Remove(name string) error

func (*TestFS) RemoveAll added in v1.34.0 

func (fs *TestFS) RemoveAll(path string) error

func (*TestFS) Rename added in v1.34.0 

func (fs *TestFS) Rename(oldpath, newpath string) error

func (*TestFS) Stat added in v1.34.0 

func (fs *TestFS) Stat(name string) (os.FileInfo, error)

func (*TestFS) Truncate added in v1.34.0 

func (fs *TestFS) Truncate(name string, size int64) error

type TestFile added in v1.34.0 

type TestFile struct {
	File
	OnWrite func(b []byte) (n int, err error)
	OnRead  func(b []byte) (n int, err error)
	OnSync  func() error
}

func (*TestFile) Read added in v1.34.0 

func (f *TestFile) Read(b []byte) (n int, err error)

func (*TestFile) Sync added in v1.34.0 

func (f *TestFile) Sync() error

func (*TestFile) Write added in v1.34.0 

func (f *TestFile) Write(b []byte) (n int, err error)

type UnboundedChannel added in v1.34.0 

type UnboundedChannel[T any] struct {
	// contains filtered or unexported fields
}

func MakeUnboundedChannel added in v1.34.0 

func MakeUnboundedChannel[T any]() *UnboundedChannel[T]

MakeUnboundedChannel creates a channel with unbounded capacity. Any push is guaranteed to succeed without blocking, regardless of the number of pending reads on `out`. The implementation uses a background goroutine that continuously drains an internal queue to the output channel. If the channel is empty, the goroutine sleeps until a new item is added. This channel doesn't implement backpressure and should be used with caution to avoid excessive memory usage.

func (*UnboundedChannel[T]) Close added in v1.34.0 

func (u *UnboundedChannel[T]) Close(ctx context.Context)

Close the channel and initiates the shutdown of the background goroutine. The provided context can be used to cancel the draining of remaining items to the output channel.

func (*UnboundedChannel[T]) Len added in v1.34.0 

func (u *UnboundedChannel[T]) Len() int

Len returns the number of items currently buffered in the channel.

func (*UnboundedChannel[T]) Out added in v1.34.0 

func (u *UnboundedChannel[T]) Out() <-chan T

Out returns a read-only channel from which values can be received.

func (*UnboundedChannel[T]) Push added in v1.34.0 

func (u *UnboundedChannel[T]) Push(v T) bool

Push adds a value to the channel. This operation does not wait for a corresponding read and buffers the value internally until it can be sent. If the channel has been closed, Push returns false.

type Vector added in v1.29.0 

type Vector[T dto.Embedding] struct {
	ID     uint64
	Vector T
}

func (*Vector[T]) Len added in v1.29.0 

func (v *Vector[T]) Len() int

func (*Vector[T]) Validate added in v1.29.0 

func (v *Vector[T]) Validate(vectorIndex VectorIndex) error

type VectorForID 

type VectorForID[T []float32 | []uint64 | float32 | byte | uint64] func(ctx context.Context, id uint64) ([]T, error)

type VectorIndex added in v1.29.0 

type VectorIndex interface {
	AddBatch(ctx context.Context, ids []uint64, vector [][]float32) error
	ValidateBeforeInsert(vector []float32) error
}

type VectorIndexMulti added in v1.33.0 

type VectorIndexMulti interface {
	AddMultiBatch(ctx context.Context, docIds []uint64, vectors [][][]float32) error
	ValidateMultiBeforeInsert(vector [][]float32) error
}

type VectorRecord added in v1.28.0 

type VectorRecord interface {
	Len() int
	Validate(vectorIndex VectorIndex) error
}

type VectorSlice 

type VectorSlice struct {
	Slice []float32
	Mem   []float32
	Buff8 []byte
	Buff  []byte
}

type VectorUint64Slice added in v1.27.0 

type VectorUint64Slice struct {
	Slice []uint64
}

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.