1
0
Files
easy_rag/internal/pkg/database/milvus/operations.go

271 lines
9.6 KiB
Go

package milvus
import (
"context"
"fmt"
"sort"
"easy_rag/internal/models"
"github.com/milvus-io/milvus-sdk-go/v2/client"
"github.com/milvus-io/milvus-sdk-go/v2/entity"
)
// InsertDocuments inserts documents into the "documents" collection.
func (m *Client) InsertDocuments(ctx context.Context, docs []models.Document) error {
idColumn := entity.NewColumnVarChar("ID", extractIDs(docs))
contentColumn := entity.NewColumnVarChar("Content", extractContents(docs))
linkColumn := entity.NewColumnVarChar("Link", extractLinks(docs))
filenameColumn := entity.NewColumnVarChar("Filename", extractFilenames(docs))
categoryColumn := entity.NewColumnVarChar("Category", extractCategories(docs))
embeddingModelColumn := entity.NewColumnVarChar("EmbeddingModel", extractEmbeddingModels(docs))
summaryColumn := entity.NewColumnVarChar("Summary", extractSummaries(docs))
metadataColumn := entity.NewColumnVarChar("Metadata", extractMetadata(docs))
vectorColumn := entity.NewColumnFloatVector("Vector", 1024, extractVectorsDocs(docs))
// Insert the data
_, err := m.Instance.Insert(ctx, "documents", "_default", idColumn, contentColumn, linkColumn, filenameColumn,
categoryColumn, embeddingModelColumn, summaryColumn, metadataColumn, vectorColumn)
if err != nil {
return fmt.Errorf("failed to insert documents: %w", err)
}
// Flush the collection
err = m.Instance.Flush(ctx, "documents", false)
if err != nil {
return fmt.Errorf("failed to flush documents collection: %w", err)
}
return nil
}
// InsertEmbeddings inserts embeddings into the "chunks" collection.
func (m *Client) InsertEmbeddings(ctx context.Context, embeddings []models.Embedding) error {
idColumn := entity.NewColumnVarChar("ID", extractEmbeddingIDs(embeddings))
documentIDColumn := entity.NewColumnVarChar("DocumentID", extractDocumentIDs(embeddings))
vectorColumn := entity.NewColumnFloatVector("Vector", 1024, extractVectors(embeddings))
textChunkColumn := entity.NewColumnVarChar("TextChunk", extractTextChunks(embeddings))
dimensionColumn := entity.NewColumnInt32("Dimension", extractDimensions(embeddings))
orderColumn := entity.NewColumnInt32("Order", extractOrders(embeddings))
_, err := m.Instance.Insert(ctx, "chunks", "_default", idColumn, documentIDColumn, vectorColumn,
textChunkColumn, dimensionColumn, orderColumn)
if err != nil {
return fmt.Errorf("failed to insert embeddings: %w", err)
}
err = m.Instance.Flush(ctx, "chunks", false)
if err != nil {
return fmt.Errorf("failed to flush chunks collection: %w", err)
}
return nil
}
// GetDocumentByID retrieves a document from the "documents" collection by ID.
func (m *Client) GetDocumentByID(ctx context.Context, id string) (map[string]interface{}, error) {
collectionName := "documents"
expr := fmt.Sprintf("ID == '%s'", id)
projections := []string{"ID", "Content", "Link", "Filename", "Category", "EmbeddingModel", "Summary", "Metadata"} // Fetch all fields
results, err := m.Instance.Query(ctx, collectionName, nil, expr, projections)
if err != nil {
return nil, fmt.Errorf("failed to query document by ID: %w", err)
}
if len(results) == 0 {
return nil, fmt.Errorf("document with ID '%s' not found", id)
}
mp, err := transformResultSet(results, "ID", "Content", "Link", "Filename", "Category", "EmbeddingModel", "Summary", "Metadata")
if err != nil {
return nil, fmt.Errorf("failed to unmarshal document: %w", err)
}
// convert metadata to map
mp[0]["Metadata"] = convertToMetadata(mp[0]["Metadata"].(string))
return mp[0], err
}
// GetAllDocuments retrieves all documents from the "documents" collection.
func (m *Client) GetAllDocuments(ctx context.Context) ([]models.Document, error) {
collectionName := "documents"
projections := []string{"ID", "Content", "Link", "Filename", "Category", "EmbeddingModel", "Summary", "Metadata"} // Fetch all fields
expr := ""
rs, err := m.Instance.Query(ctx, collectionName, nil, expr, projections, client.WithLimit(1000))
if err != nil {
return nil, fmt.Errorf("failed to query all documents: %w", err)
}
if len(rs) == 0 {
return nil, fmt.Errorf("no documents found in the collection")
}
results, err := transformResultSet(rs, "ID", "Content", "Link", "Filename", "Category", "EmbeddingModel", "Summary", "Metadata")
if err != nil {
return nil, fmt.Errorf("failed to unmarshal all documents: %w", err)
}
var docs []models.Document = make([]models.Document, len(results))
for i, result := range results {
docs[i] = models.Document{
ID: result["ID"].(string),
Content: result["Content"].(string),
Link: result["Link"].(string),
Filename: result["Filename"].(string),
Category: result["Category"].(string),
EmbeddingModel: result["EmbeddingModel"].(string),
Summary: result["Summary"].(string),
Metadata: convertToMetadata(results[0]["Metadata"].(string)),
}
}
return docs, nil
}
// GetAllEmbeddingByDocID retrieves all embeddings linked to a specific DocumentID from the "chunks" collection.
func (m *Client) GetAllEmbeddingByDocID(ctx context.Context, documentID string) ([]models.Embedding, error) {
collectionName := "chunks"
projections := []string{"ID", "DocumentID", "TextChunk", "Order"} // Fetch all fields
expr := fmt.Sprintf("DocumentID == '%s'", documentID)
rs, err := m.Instance.Query(ctx, collectionName, nil, expr, projections, client.WithLimit(1000))
if err != nil {
return nil, fmt.Errorf("failed to query embeddings by DocumentID: %w", err)
}
if rs.Len() == 0 {
return nil, fmt.Errorf("no embeddings found for DocumentID '%s'", documentID)
}
results, err := transformResultSet(rs, "ID", "DocumentID", "TextChunk", "Order")
if err != nil {
return nil, fmt.Errorf("failed to unmarshal all documents: %w", err)
}
var embeddings []models.Embedding = make([]models.Embedding, rs.Len())
for i, result := range results {
embeddings[i] = models.Embedding{
ID: result["ID"].(string),
DocumentID: result["DocumentID"].(string),
TextChunk: result["TextChunk"].(string),
Order: result["Order"].(int64),
}
}
return embeddings, nil
}
func (m *Client) Search(ctx context.Context, vectors [][]float32, topK int) ([]models.Embedding, error) {
const (
collectionName = "chunks"
vectorDim = 1024 // Replace with your actual vector dimension
)
projections := []string{"ID", "DocumentID", "TextChunk", "Order"}
metricType := entity.L2 // Default metric type
// Validate and convert input vectors
searchVectors, err := validateAndConvertVectors(vectors, vectorDim)
if err != nil {
return nil, err
}
// Set search parameters
searchParams, err := entity.NewIndexIvfFlatSearchParam(16) // 16 is the number of clusters for IVF_FLAT index
if err != nil {
return nil, fmt.Errorf("failed to create search params: %w", err)
}
// Perform the search
searchResults, err := m.Instance.Search(ctx, collectionName, nil, "", projections, searchVectors, "Vector", metricType, topK, searchParams, client.WithLimit(10))
if err != nil {
return nil, fmt.Errorf("failed to search collection: %w", err)
}
// Process search results
embeddings, err := processSearchResults(searchResults)
if err != nil {
return nil, fmt.Errorf("failed to process search results: %w", err)
}
return embeddings, nil
}
// validateAndConvertVectors validates vector dimensions and converts them to Milvus-compatible format.
func validateAndConvertVectors(vectors [][]float32, expectedDim int) ([]entity.Vector, error) {
searchVectors := make([]entity.Vector, len(vectors))
for i, vector := range vectors {
if len(vector) != expectedDim {
return nil, fmt.Errorf("vector dimension mismatch: expected %d, got %d", expectedDim, len(vector))
}
searchVectors[i] = entity.FloatVector(vector)
}
return searchVectors, nil
}
// processSearchResults transforms and aggregates the search results into embeddings and sorts by score.
func processSearchResults(results []client.SearchResult) ([]models.Embedding, error) {
var embeddings []models.Embedding
for _, result := range results {
for i := 0; i < result.ResultCount; i++ {
embeddingMap, err := transformSearchResultSet(result, "ID", "DocumentID", "TextChunk", "Order")
if err != nil {
return nil, fmt.Errorf("failed to transform search result set: %w", err)
}
for _, embedding := range embeddingMap {
embeddings = append(embeddings, models.Embedding{
ID: embedding["ID"].(string),
DocumentID: embedding["DocumentID"].(string),
TextChunk: embedding["TextChunk"].(string),
Order: embedding["Order"].(int64), // Assuming 'Order' is a float64 type
Score: embedding["Score"].(float32),
})
}
}
}
// Sort embeddings by score in descending order (higher is better)
sort.Slice(embeddings, func(i, j int) bool {
return embeddings[i].Score > embeddings[j].Score
})
return embeddings, nil
}
// DeleteDocument deletes a document from the "documents" collection by ID.
func (m *Client) DeleteDocument(ctx context.Context, id string) error {
collectionName := "documents"
partitionName := "_default"
expr := fmt.Sprintf("ID == '%s'", id)
err := m.Instance.Delete(ctx, collectionName, partitionName, expr)
if err != nil {
return fmt.Errorf("failed to delete document by ID: %w", err)
}
return nil
}
// DeleteEmbedding deletes an embedding from the "chunks" collection by ID.
func (m *Client) DeleteEmbedding(ctx context.Context, id string) error {
collectionName := "chunks"
partitionName := "_default"
expr := fmt.Sprintf("DocumentID == '%s'", id)
err := m.Instance.Delete(ctx, collectionName, partitionName, expr)
if err != nil {
return fmt.Errorf("failed to delete embedding by DocumentID: %w", err)
}
return nil
}