Querying JSON in open lakehouse tables means parsing everything at runtime — no predicate pushdown, no skipping.
Iceberg v3 Variant flips that by:
✦ Shredding nested data into typed columns at write time
✦ Enabling predicate pushdown at the column level
Result: In an 11-query benchmark on identical Iceberg tables, Snowflake was 2.3x faster at equivalent compute.
顯示更多