finally figured out why our storage costs are such a mess lately. it's not even about the compute or dbus, it's just all those stupid nightly exports and duplicate parquet files we're running to keep snowflake in sync with databricks. if unifrom actually works, we wouldn't need that extra layer of metadata just to read delta as iceberg.
>one copy of Parquet with multiple metadata layers describing itsounds great on paper but i'm skeptical about how much latency we'd actually see on the snowflake side when writing. it's basically trying to kill the vendor lock-in by making the catalog irrelevant.
good luck with the maintenance overhead though is anyone here actually running unifrom in production yet?
full read:
https://hackernoon.com/delta-lake-uniform-is-the-real-fix-for-the-snowflake-vs-databricks-standoff?source=rss