A filter that misses the first cluster column saves far less

What it is

A clustered table keeps its data sorted by the cluster columns, in the order they were declared. BigQuery skips blocks only when a filter compares those columns directly, starting with the first.

Why it costs money

• Measured by dry run on one day of wikipedia.pageviews_2024, clustered on wiki then title: no filter 5.68 GiB, wiki 1.00 GiB, title alone 2.07 GiB, both 0.11 GiB.
• Wrapping the first column in a function loses the saving: UPPER(wiki) = 'EN' read 6.47 GiB, no better than no filter.
• A dry run gives an upper bound on a clustered table. The completed job shows the real bytes.

Worked example

SELECT title, views
FROM `bigquery-public-data.wikipedia.pageviews_2024`
WHERE DATE(datehour) = '2024-06-01'
  AND title = 'BigQuery'

SELECT title, views
FROM `bigquery-public-data.wikipedia.pageviews_2024`
WHERE DATE(datehour) = '2024-06-01'
  AND wiki = 'en'
  AND title = 'BigQuery'

How to fix it

1. 1Filter on the first cluster column, as well as any later ones.
2. 2Compare the bare column. Move any function to the other side of the comparison.
3. 3When creating a table, put the column queries filter on most first.