Predicate pushdown means applying filters as early as possible, ideally close to the data source.
Instead of reading:
1 billion rows ↓ filter ↓ 10,000 rows
the system tries to read only relevant data:
filter pushed to source ↓ 10,000 rows
Example:
SELECT * FROM sales WHERE year = 2026;
With a partitioned data source, the engine may read only the relevant year=2026 partitions.
This reduces:
- Data scanned
- I/O
- Network transfer
- Processing