[chimerasaurus]

I’d take issue with the “Iceberg is slow” theme that Databricks in particular has tried to push.

If that were true, Snowflake would not be as fast on Iceberg/Parquet as its native format. The engine makes something fast or slow, not the table format.

Disclaimer - am at Snowflake.

[orthoxerox]

Back when were choosing between the three formats about 1.5 years ago, Iceberg was definitely the slowest. If the situation has changed since then, I would love to see an updated comparison.

We tested all three of them using Spark batches that converted a stream of changes into SCD2.