I have a notebook with five Parquet files for a total of ~230MB. Much of the size can be attributed to columns that will rarely ever be queried (but are still essential and cumbersome to split out).
Currently even a simple count(*) across all files causes each file to get fetched in full. I do recall that range requests had been tricky in the past, and I wanted to check if you’d expect them to work (and the issue might be on my end) or if I would need to look for other ways to split the load. ![]()