How BEGIN TRANSACTION; COMMIT; creates parquet files in DuckLake?

[intersynct]

If I wrap a number of insert statements between BEGIN TRANSACTION; COMMIT; statements does DuckLake create one parquet file, rather than several files if the inserts were not grouped in a transaction? Or should one not worry and use merge_adjacent_files with ducklake_cleanup_old_files and let DuckLake manage files?

[Tishj]

No, parquet files are written optimistically on INSERT.
They're automatically cleaned up on abort of the transaction.

If you are worried about small files you should indeed rely on merge_adjacent_files or make use of data-inlining