Enha: migrations with different emb tables

rag/rag.go

@@ -43,10 +43,7 @@ func New(l *slog.Logger, s storage.FullRepo, cfg *config.Config) *RAG {
 		storage:  NewVectorStorage(l, s),
 	}
 
-	// Create the necessary tables
+	// Note: Vector tables are created via database migrations, not at runtime
-	if err := rag.storage.CreateTables(); err != nil {
-		l.Error("failed to create vector tables", "error", err)
-	}
 
 	return rag
 }

rag/storage.go

@@ -28,45 +28,6 @@ func NewVectorStorage(logger *slog.Logger, store storage.FullRepo) *VectorStorag
 	}
 }
 
-// CreateTables creates the necessary tables for vector storage
-func (vs *VectorStorage) CreateTables() error {
-	// Create tables for common embedding dimensions
-	embeddingSizes := []int{384, 768, 1024, 1536, 2048, 3072, 4096, 5120}
-	// Pre-allocate queries slice: each embedding size needs 1 table + 3 indexes = 4 queries per size
-	queries := make([]string, 0, len(embeddingSizes)*4)
-
-	// Generate table creation queries for each embedding size
-	for _, size := range embeddingSizes {
-		tableName := fmt.Sprintf("embeddings_%d", size)
-		queries = append(queries,
-			fmt.Sprintf(`CREATE TABLE IF NOT EXISTS %s (
-				id INTEGER PRIMARY KEY AUTOINCREMENT,
-				embeddings BLOB NOT NULL,
-				slug TEXT NOT NULL,
-				raw_text TEXT NOT NULL,
-				filename TEXT NOT NULL,
-				created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
-			)`, tableName),
-		)
-	}
-
-	// Add indexes for all supported sizes
-	for _, size := range embeddingSizes {
-		tableName := fmt.Sprintf("embeddings_%d", size)
-		queries = append(queries,
-			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS idx_%s_filename ON %s(filename)`, tableName, tableName),
-			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS idx_%s_slug ON %s(slug)`, tableName, tableName),
-			fmt.Sprintf(`CREATE INDEX IF NOT EXISTS idx_%s_created_at ON %s(created_at)`, tableName, tableName),
-		)
-	}
-
-	for _, query := range queries {
-		if _, err := vs.sqlxDB.Exec(query); err != nil {
-			return fmt.Errorf("failed to create table: %w", err)
-		}
-	}
-	return nil
-}
 
 // SerializeVector converts []float32 to binary blob
 func SerializeVector(vec []float32) []byte {

storage/migrations/002_add_vector.down.sql

@@ -1,10 +1,34 @@
 -- Drop vector storage tables
 DROP INDEX IF EXISTS idx_embeddings_384_filename;
+DROP INDEX IF EXISTS idx_embeddings_768_filename;
+DROP INDEX IF EXISTS idx_embeddings_1024_filename;
+DROP INDEX IF EXISTS idx_embeddings_1536_filename;
+DROP INDEX IF EXISTS idx_embeddings_2048_filename;
+DROP INDEX IF EXISTS idx_embeddings_3072_filename;
+DROP INDEX IF EXISTS idx_embeddings_4096_filename;
 DROP INDEX IF EXISTS idx_embeddings_5120_filename;
 DROP INDEX IF EXISTS idx_embeddings_384_slug;
+DROP INDEX IF EXISTS idx_embeddings_768_slug;
+DROP INDEX IF EXISTS idx_embeddings_1024_slug;
+DROP INDEX IF EXISTS idx_embeddings_1536_slug;
+DROP INDEX IF EXISTS idx_embeddings_2048_slug;
+DROP INDEX IF EXISTS idx_embeddings_3072_slug;
+DROP INDEX IF EXISTS idx_embeddings_4096_slug;
 DROP INDEX IF EXISTS idx_embeddings_5120_slug;
 DROP INDEX IF EXISTS idx_embeddings_384_created_at;
+DROP INDEX IF EXISTS idx_embeddings_768_created_at;
+DROP INDEX IF EXISTS idx_embeddings_1024_created_at;
+DROP INDEX IF EXISTS idx_embeddings_1536_created_at;
+DROP INDEX IF EXISTS idx_embeddings_2048_created_at;
+DROP INDEX IF EXISTS idx_embeddings_3072_created_at;
+DROP INDEX IF EXISTS idx_embeddings_4096_created_at;
 DROP INDEX IF EXISTS idx_embeddings_5120_created_at;
 
 DROP TABLE IF EXISTS embeddings_384;
+DROP TABLE IF EXISTS embeddings_768;
+DROP TABLE IF EXISTS embeddings_1024;
+DROP TABLE IF EXISTS embeddings_1536;
+DROP TABLE IF EXISTS embeddings_2048;
+DROP TABLE IF EXISTS embeddings_3072;
+DROP TABLE IF EXISTS embeddings_4096;
 DROP TABLE IF EXISTS embeddings_5120;

storage/migrations/002_add_vector.up.sql

@@ -8,6 +8,60 @@ CREATE TABLE IF NOT EXISTS embeddings_384 (
     created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
 );
 
+CREATE TABLE IF NOT EXISTS embeddings_768 (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    embeddings BLOB NOT NULL,
+    slug TEXT NOT NULL,
+    raw_text TEXT NOT NULL,
+    filename TEXT NOT NULL DEFAULT '',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS embeddings_1024 (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    embeddings BLOB NOT NULL,
+    slug TEXT NOT NULL,
+    raw_text TEXT NOT NULL,
+    filename TEXT NOT NULL DEFAULT '',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS embeddings_1536 (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    embeddings BLOB NOT NULL,
+    slug TEXT NOT NULL,
+    raw_text TEXT NOT NULL,
+    filename TEXT NOT NULL DEFAULT '',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS embeddings_2048 (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    embeddings BLOB NOT NULL,
+    slug TEXT NOT NULL,
+    raw_text TEXT NOT NULL,
+    filename TEXT NOT NULL DEFAULT '',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS embeddings_3072 (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    embeddings BLOB NOT NULL,
+    slug TEXT NOT NULL,
+    raw_text TEXT NOT NULL,
+    filename TEXT NOT NULL DEFAULT '',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS embeddings_4096 (
+    id INTEGER PRIMARY KEY AUTOINCREMENT,
+    embeddings BLOB NOT NULL,
+    slug TEXT NOT NULL,
+    raw_text TEXT NOT NULL,
+    filename TEXT NOT NULL DEFAULT '',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
 CREATE TABLE IF NOT EXISTS embeddings_5120 (
     id INTEGER PRIMARY KEY AUTOINCREMENT,
     embeddings BLOB NOT NULL,
@@ -19,8 +73,26 @@ CREATE TABLE IF NOT EXISTS embeddings_5120 (
 
 -- Indexes for better performance
 CREATE INDEX IF NOT EXISTS idx_embeddings_384_filename ON embeddings_384(filename);
+CREATE INDEX IF NOT EXISTS idx_embeddings_768_filename ON embeddings_768(filename);
+CREATE INDEX IF NOT EXISTS idx_embeddings_1024_filename ON embeddings_1024(filename);
+CREATE INDEX IF NOT EXISTS idx_embeddings_1536_filename ON embeddings_1536(filename);
+CREATE INDEX IF NOT EXISTS idx_embeddings_2048_filename ON embeddings_2048(filename);
+CREATE INDEX IF NOT EXISTS idx_embeddings_3072_filename ON embeddings_3072(filename);
+CREATE INDEX IF NOT EXISTS idx_embeddings_4096_filename ON embeddings_4096(filename);
 CREATE INDEX IF NOT EXISTS idx_embeddings_5120_filename ON embeddings_5120(filename);
 CREATE INDEX IF NOT EXISTS idx_embeddings_384_slug ON embeddings_384(slug);
+CREATE INDEX IF NOT EXISTS idx_embeddings_768_slug ON embeddings_768(slug);
+CREATE INDEX IF NOT EXISTS idx_embeddings_1024_slug ON embeddings_1024(slug);
+CREATE INDEX IF NOT EXISTS idx_embeddings_1536_slug ON embeddings_1536(slug);
+CREATE INDEX IF NOT EXISTS idx_embeddings_2048_slug ON embeddings_2048(slug);
+CREATE INDEX IF NOT EXISTS idx_embeddings_3072_slug ON embeddings_3072(slug);
+CREATE INDEX IF NOT EXISTS idx_embeddings_4096_slug ON embeddings_4096(slug);
 CREATE INDEX IF NOT EXISTS idx_embeddings_5120_slug ON embeddings_5120(slug);
 CREATE INDEX IF NOT EXISTS idx_embeddings_384_created_at ON embeddings_384(created_at);
+CREATE INDEX IF NOT EXISTS idx_embeddings_768_created_at ON embeddings_768(created_at);
+CREATE INDEX IF NOT EXISTS idx_embeddings_1024_created_at ON embeddings_1024(created_at);
+CREATE INDEX IF NOT EXISTS idx_embeddings_1536_created_at ON embeddings_1536(created_at);
+CREATE INDEX IF NOT EXISTS idx_embeddings_2048_created_at ON embeddings_2048(created_at);
+CREATE INDEX IF NOT EXISTS idx_embeddings_3072_created_at ON embeddings_3072(created_at);
+CREATE INDEX IF NOT EXISTS idx_embeddings_4096_created_at ON embeddings_4096(created_at);
 CREATE INDEX IF NOT EXISTS idx_embeddings_5120_created_at ON embeddings_5120(created_at);

storage/vector.go

@@ -45,17 +45,24 @@ func mathBitsToFloat32(b uint32) float32 {
 	return *(*float32)(unsafe.Pointer(&b))
 }
 
-var (
-	vecTableName5120 = "embeddings_5120"
-	vecTableName384  = "embeddings_384"
-)
-
 func fetchTableName(emb []float32) (string, error) {
 	switch len(emb) {
-	case 5120:
-		return vecTableName5120, nil
 	case 384:
-		return vecTableName384, nil
+		return "embeddings_384", nil
+	case 768:
+		return "embeddings_768", nil
+	case 1024:
+		return "embeddings_1024", nil
+	case 1536:
+		return "embeddings_1536", nil
+	case 2048:
+		return "embeddings_2048", nil
+	case 3072:
+		return "embeddings_3072", nil
+	case 4096:
+		return "embeddings_4096", nil
+	case 5120:
+		return "embeddings_5120", nil
 	default:
 		return "", fmt.Errorf("no table for the size of %d", len(emb))
 	}
@@ -185,8 +192,12 @@ func sqrt(f float32) float32 {
 func (p ProviderSQL) ListFiles() ([]string, error) {
 	fileLists := make([][]string, 0)
 
-	// Query both tables and combine results
+	// Query all supported tables and combine results
-	for _, table := range []string{vecTableName384, vecTableName5120} {
+	tableNames := []string{
+		"embeddings_384", "embeddings_768", "embeddings_1024", "embeddings_1536",
+		"embeddings_2048", "embeddings_3072", "embeddings_4096", "embeddings_5120",
+	}
+	for _, table := range tableNames {
 		query := "SELECT DISTINCT filename FROM " + table
 		rows, err := p.db.Query(query)
 		if err != nil {
@@ -225,7 +236,11 @@ func (p ProviderSQL) ListFiles() ([]string, error) {
 func (p ProviderSQL) RemoveEmbByFileName(filename string) error {
 	var errors []string
 
-	for _, table := range []string{vecTableName384, vecTableName5120} {
+	tableNames := []string{
+		"embeddings_384", "embeddings_768", "embeddings_1024", "embeddings_1536",
+		"embeddings_2048", "embeddings_3072", "embeddings_4096", "embeddings_5120",
+	}
+	for _, table := range tableNames {
 		query := fmt.Sprintf("DELETE FROM %s WHERE filename = ?", table)
 		if _, err := p.db.Exec(query, filename); err != nil {
 			errors = append(errors, err.Error())