Forty Eight Mebibytes Of Three Hundred Sixty Five Gibibytes

A year of events sits in S3 as 365 Parquet files, one per day, each 1 GiB with 80 columns in a single row group. The paths carry no date, so the engine cannot prune by filename. Its plan is: read every file's footer (about 64 KiB each) for column chunk offsets and row group statistics, discard files whose event_date min/max cannot match, then read only the needed column chunks from what is left.

SELECT user_id, amount FROM events WHERE event_date = DATE '2026-03-01';

user_id and amount are each about 1/80 of a file's bytes, and exactly one file's statistics overlap that date. Roughly how many bytes get read?