Looking at the source code, it looks like using RowIter is efficient in the sense that it will load one row group at a time.
My problem is this: The data we're dealing with has no row group metadata, meaning, I think, every row is treated as a group.
And these files are very large - one is 24 Gb (or, rather, it is as a CSV - should be smaller in parquet).
I'm worried this is going to cause poor performance. There's probably nothing I can do about, but, if someone very familiar with the crate can comment, great.