It’s great to write carefree SQL, with the DB platform totally abstracted, until there’s a massive spike in query latency. Then hash joins vs nested loops and full table scans is useful knowledge which is a) mostly vendor specific b) requires a good prod-sized dataset to test on.