Snowflake Speeds Up Workloads With Hierarchical Selectivity Estimates

A recent Snowflake Automatic Performance enhancement in the core execution engine has led to improved query plan quality for specific query patterns - this has resulted in an 8x speedup, on average, for relevant workloads. This feature is on by default and customers benefit from it automatically, without needing to manually adjust any configurations.

Specifically, this new capability - Hierarchical Selectivity Estimates - skips computing the selectivity of predicates based on the table-level aggregated statistics, and instead drills down and applies estimation logic on the micropartition-level statistics. Compared to table-level statistics, the new approach provides high quality table statistics while allowing for efficient updates to the data.