diff --git a/executor.go b/executor.go
index 885200c4f..d8422bd05 100644
--- a/executor.go
+++ b/executor.go
@@ -8504,7 +8504,7 @@ func DeleteRowsWithFlow(ctx context.Context, src *Row, idx *Index, shard uint64,
 	}
 	var change bool
 	var err error
-	limit := 65536
+	limit := idx.holder.cfg.RBFConfig.MaxDelete
 	for i := 0; i < len(bits); i += limit {
 		batch := roaring.NewBitmap(bits[i:min(i+limit, len(bits))]...)
 		if idx.Keys() {
diff --git a/rbf/cfg/cfg.go b/rbf/cfg/cfg.go
index 184775e33..2b65a2b2b 100644
--- a/rbf/cfg/cfg.go
+++ b/rbf/cfg/cfg.go
@@ -41,6 +41,9 @@ type Config struct {
 	// background checkpoints. It cannot be set from toml. The default is
 	// to use stderr.
 	Logger logger.Logger `toml:"-"`
+
+	// The maximum number of bits to be deleted in a single transaction default(65536)
+	MaxDelete int `toml:"max-delete"`
 }
 
 func NewDefaultConfig() *Config {
@@ -50,6 +53,7 @@ func NewDefaultConfig() *Config {
 		MinWALCheckpointSize: DefaultMinWALCheckpointSize,
 		MaxWALCheckpointSize: DefaultMaxWALCheckpointSize,
 		FsyncEnabled:         true,
+		MaxDelete:            DefaultMaxDelete,
 
 		// CI passed with 20. 50 was too big for CI, even on X-large instances.
 		// For now we default to 0, which means use sync.Pool.
diff --git a/rbf/cfg/os.go b/rbf/cfg/os.go
index 5ca1f60d9..057b4320f 100644
--- a/rbf/cfg/os.go
+++ b/rbf/cfg/os.go
@@ -13,3 +13,6 @@ const DefaultMaxSize = 4 * (1 << 30)
 // size of the WAL. The size can be increased by updating the DB.MaxWALSize
 // and reopening the database. This setting mainly affects virtual space usage.
 const DefaultMaxWALSize = 4 * (1 << 30)
+
+// DefaultMaxDelete is the maximum number of bits that will be deleted in a single batch
+const DefaultMaxDelete = 65536