Yes! Thanks for sharing. I’m really excited to check it out. Have you played around? Any initial thoughts?

Oh really cool. Thanks!

@benesch.bsky.social, sounds like you might have more to write about soon…

Ooo, I haven’t heard of any of the assumptions — what have you heard?

…it could be interesting if e.g. they added built-in caching of Iceberg metadata and Parquet files to offer Express One Zone-like performance on frequently accessed objects. Maybe that would move the performance needle…

I wish their claim of “3x query performance and 10x transactions per second compared to self-managed tables in general purpose buckets” was based on new tech, not worst-case assumptions about how general purpose tables are used…

This also feels like something Cloudflare should do on top of R2 only with an Iceberg REST catalog implementation. To me, it actually feels more like the type of product that Cloudflare has been shipping, and would be a nice compliment to their serverless D1 database (~SQLite for up to 10GB)

So at first glance, it looks to be like a tighter integration with GlueCatalog, but not necessarily a new underlying catalog implementation (as far as I can tell).

In the AWS console, when you follow the "create table bucket" flow and look under the integration details dropdown, you see: "A catalog named s3tablescatalog is created in AWS Glue Data Catalog, as well as subcatalogs within it for each table bucket in this Region."

You lost me at “ground up rewrite of DD in C” (as fun as that sounds) 😅

Really cool!… but aren’t the egress fees on 10GB of free data is going to add up quick? A single query across the full dataset could cost them $0.50 (at a very conservative $0.05/GB)

Woah congrats! This is super cool. Is it possible/planned to enable queries from a non Postgres query engine? (even if read-only)

100% — some of our use cases are uniquely good fits for duckdb! E.g in-process Parquet de/serialization.

Great meeting you @chris.blue !