What is pruning in Snowflake, and how does it differ from partitioning and clustering? Is clustering mandatory for pruning?

Pruning in Snowflake refers to the query optimizer’s ability to skip over large portions of a table that cannot contain rows matching the query predicates. It uses metadata about the data’s distribution—specifically the clustering keys—to determine which micro‑partitions can be excluded. Partitioning is a physical division of data on disk, while clustering is a logical ordering of data within those partitions. Clustering is not mandatory for pruning; Snowflake can still prune based on the table’s internal metadata, but clustering keys dramatically improve pruning efficiency by creating tighter bounds on the data ranges. When a table is clustered, the optimizer can quickly identify and skip entire micro‑partitions that fall outside the predicate range, reducing I/O and speeding up query execution. Without clustering, pruning still occurs but is less selective, leading to more data being scanned. Therefore, clustering is a best practice for large tables that benefit from frequent filtering on key columns.