LinkinStars commented on code in PR #1510: URL: https://github.com/apache/answer/pull/1510#discussion_r2889699958
########## internal/repo/embedding/embedding_repo.go: ########## @@ -0,0 +1,197 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you 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 embedding + +import ( + "context" + "encoding/json" + "math" + "sort" + + "github.com/apache/answer/internal/base/data" + "github.com/apache/answer/internal/entity" + "github.com/segmentfault/pacman/log" + "xorm.io/builder" +) + +// EmbeddingRepo defines the interface for embedding data access. +type EmbeddingRepo interface { + Upsert(ctx context.Context, emb *entity.Embedding) error + GetByObjectID(ctx context.Context, objectID, objectType string) (*entity.Embedding, bool, error) + GetAll(ctx context.Context) ([]*entity.Embedding, error) + SearchSimilar(ctx context.Context, queryVector []float32, topK int) ([]SimilarResult, error) + DeleteByObjectID(ctx context.Context, objectID, objectType string) error + Count(ctx context.Context) (int64, error) +} + +// SimilarResult holds a similarity search result. +type SimilarResult struct { + ObjectID string `json:"object_id"` + ObjectType string `json:"object_type"` + Metadata string `json:"metadata"` + Score float64 `json:"score"` +} + +type embeddingRepo struct { + data *data.Data +} + +// NewEmbeddingRepo creates a new EmbeddingRepo. +func NewEmbeddingRepo(data *data.Data) EmbeddingRepo { + return &embeddingRepo{data: data} +} + +// Upsert inserts or updates an embedding by (object_id, object_type). +func (r *embeddingRepo) Upsert(ctx context.Context, emb *entity.Embedding) error { + existing := &entity.Embedding{} + exist, err := r.data.DB.Context(ctx). + Where(builder.Eq{"object_id": emb.ObjectID, "object_type": emb.ObjectType}). + Get(existing) + if err != nil { + log.Errorf("check embedding existence failed: %v", err) + return err + } + + if exist { + emb.ID = existing.ID + _, err = r.data.DB.Context(ctx).ID(existing.ID). + Cols("content_hash", "metadata", "embedding", "dimensions", "updated_at"). + Update(emb) + if err != nil { + log.Errorf("update embedding failed: %v", err) + return err + } + return nil + } + + _, err = r.data.DB.Context(ctx).Insert(emb) + if err != nil { + log.Errorf("insert embedding failed: %v", err) + return err + } + return nil +} + +// GetByObjectID returns an embedding by object ID and type. +func (r *embeddingRepo) GetByObjectID(ctx context.Context, objectID, objectType string) (*entity.Embedding, bool, error) { + emb := &entity.Embedding{} + exist, err := r.data.DB.Context(ctx). + Where(builder.Eq{"object_id": objectID, "object_type": objectType}). + Get(emb) + if err != nil { + log.Errorf("get embedding failed: %v", err) + return nil, false, err + } + return emb, exist, nil +} + +// GetAll returns all embeddings. +func (r *embeddingRepo) GetAll(ctx context.Context) ([]*entity.Embedding, error) { + var list []*entity.Embedding + err := r.data.DB.Context(ctx).Find(&list) + if err != nil { + log.Errorf("get all embeddings failed: %v", err) + return nil, err + } + return list, nil +} + +// SearchSimilar performs brute-force cosine similarity search in Go. +func (r *embeddingRepo) SearchSimilar(ctx context.Context, queryVector []float32, topK int) ([]SimilarResult, error) { Review Comment: The biggest issue with this approach lies here: **all data queries are performed in-memory.** While this works for small datasets, it is undoubtedly unacceptable for large-scale data. Therefore, my suggestion is that if users are using PostgreSQL, they could directly utilize PostgreSQL + pgvector to store vector data and perform searches within the database. Furthermore, a more 'ideal' approach would be to expose this component as a plugin, allowing user Q&A data to be synchronized to external systems. This design is similar to how search plugins operate. For instance, an Elasticsearch (ES) plugin synchronizes Q&A data to ES for retrieval. The benefit of a plugin-based implementation is extensibility: users aren't restricted to the built-in database and can use their own custom vector databases. The required interfaces would likely mirror those of a search plugin, such as a data synchronization interface and a search interface. What do you think? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
