Columns you never needed
A query often reads columns it does not need. Finding them is the quickest way to cut a bill.
How it works
A query pays for the columns its result really depends on. An outermost SELECT * depends on all of them, and a nested column such as defects is one of the largest. To lower the estimate, depend on fewer columns and read fewer months.
- A plain, outermost
SELECT *reads every column, includingdefects. Name the columns you need. - A column that appears only in a
WHERE,JOINorORDER BYis read as well. - The date filter still decides how many months are read, and fewer months mean fewer bytes for every column.
Common mistake
Taking a column out of the final SELECT does not lower the estimate while the same column still appears in a WHERE, JOIN or ORDER BY.
Worked example
SELECT vehicle_id, test_result
FROM `learn.mot.fact_mot_test`
WHERE completed_date = '2024-06-12'
AND make = 'AUDI'The result depends on four columns: vehicle_id, test_result, completed_date and make.
All four are read for the June partition, even though completed_date and make are not returned.