Method multiLibraryVectorSearch

api/service/rag/search.go:3333–3534

multiLibraryVectorSearch 多知识库向量搜索

(eid int64, req *SearchRequest, configService *ChunkConfigService)

// multiLibraryVectorSearch 多知识库向量搜索
func (s *SearchService) multiLibraryVectorSearch(eid int64, req *SearchRequest, configService *ChunkConfigService) ([]SearchResultItem, error) {
	multiSearchStart := time.Now()
	logger.SysLogf("🔍 开始多知识库向量搜索 (eid=%d, libraries=%v)", eid, req.LibraryIDs)
	logger.SysDebugf("【向量检索】多库并发参数: eid=%d, query=%q, top_k=%d, libraries=%v, files=%v, chunk_types=%v",
		eid, truncateForDebug(req.Query, 256), req.TopK,
		previewInt64IDsForDebug(req.LibraryIDs, 50), previewInt64IDsForDebug(req.FileIDs, 50), req.ChunkTypes)

	// 检查 singleVectorSearchFn 是否被 mock（测试场景）
	// 如果被 mock，则走原来的流程（不预获取 embedding）
	if !isDefaultSingleVectorSearchFn() {
		// 测试场景：singleVectorSearchFn 被 mock，走原来的流程
		return s.multiLibraryVectorSearchLegacy(eid, req, configService)
	}

	if configService == nil {
		configService = NewChunkConfigService(s.db)
	}

	// 预先获取企业全局配置和 embedding（只调用一次，避免多次调用 embedding API）
	configStart := time.Now()
	config, configErr := configService.GetConfig(eid, nil, model.ChunkTypeDefault)
	if configErr != nil {
		logger.SysLogf("❌ 获取企业全局配置失败: eid=%d, err=%v", eid, configErr)
		return nil, fmt.Errorf("获取企业全局配置失败: %v", configErr)
	}
	logger.SysDebugf("【向量检索】获取配置耗时: eid=%d, elapsed_ms=%d", eid, time.Since(configStart).Milliseconds())

	if config.EmbeddingChannelID == nil {
		return nil, fmt.Errorf("未配置向量化渠道")
	}

	// 预先获取 embedding（多库并发搜索时，所有库共用同一个 embedding）
	embeddingStart := time.Now()
	queryVector64, embeddingErr := s.embedding.GetQueryEmbedding(eid, req.Query, *config.EmbeddingChannelID, config)
	if embeddingErr != nil {
		return nil, fmt.Errorf("生成查询向量失败: %v", embeddingErr)
	}
	logger.SysDebugf("【向量检索】获取Embedding耗时: eid=%d, channel_id=%d, elapsed_ms=%d, vector_dim=%d",
		eid, *config.EmbeddingChannelID, time.Since(embeddingStart).Milliseconds(), len(queryVector64))

	// 转换向量格式 (float64 -> float32)
	queryVector := make([]float32, len(queryVector64))
	for i, v := range queryVector64 {
		queryVector[i] = float32(v)
	}

	libraryMap, err := s.batchGetLibrariesByIDs(eid, req.LibraryIDs)
	if err != nil {
		logger.SysLogf("批量获取库信息失败: %v", err)
	}

	// 并发搜索各个知识库，先保留原始向量结果，随后统一富化，避免每个库重复查一轮 DB。
	type libraryResult struct {
		libraryID     int64
		collection    string
		vectorResults []vectorstore.SearchResult
		err           error
		elapsedMs     int64