Track · 2:32 · Liner note
The Bloom Filter Says Maybe
A Bloom filter is a compact probabilistic structure that says a value is either definitely absent or possibly present, and ClickHouse uses it in skipping indexes to avoid reading blocks that cannot contain the value searched for.
A Bloom filter answers one question: is this value definitely absent, or possibly present? It never says no to something that is there, but it will sometimes say maybe to something that is not. In a ClickHouse skipping index, that is enough to skip the blocks of rows that cannot contain the value. The Bloom Filter Says Maybe is worth a pause when you search a high-cardinality column.
It helps most for equality and membership lookups on columns outside the sort key, and less for range filters or values scattered across every block. Tune the false-positive rate and confirm in the plan that blocks are really being skipped.