How does data partitioning and data skipping reduce the number of partitions scanned in Snowflake?

Snowflake stores data in micro‑partitions, each containing a few megabytes of data. Data partitioning refers to how the table’s rows are distributed across these micro‑partitions, often based on a clustering key. When a query includes a predicate on a clustering key, Snowflake’s optimizer consults the metadata for each micro‑partition to see if the key’s value range overlaps the predicate. If a partition’s range is entirely outside the predicate, Snowflake skips reading that partition—a process called data skipping. This reduces I/O and improves performance. The effectiveness of data skipping depends on how well the data is clustered; a well‑clustered table will have tighter value ranges per partition, allowing the optimizer to skip more partitions. Snowflake automatically tracks the min/max values for each clustering key in each micro‑partition, enabling efficient pruning without requiring explicit partition tables.