DuckLake v1.0

ducklake.select

Very exciting! The numbers seem to crush iceberg. Has anyone tried it out for "real" workloads?

Where did you find these numbers? The way I see it, it's basically Iceberg, but with all metadata and small data changes stored in an RDBMS.

It depends what you're measuring. For streaming workloads they're reporting 923x speedup for aggregations, and 105x for insertions, over iceberg [1].

There are also some improvements other areas, e.g. cross table transactions are not possible in iceberg/delta, but they are in ducklake. [2]

[1] https://ducklake.select/2026/04/02/data-inlining-in-ducklake...

[2] https://ducklake.select/manifesto/#features

But that's a single use case. I also can put a Postgres DB in front of Iceberg, add some views in Trino to consume the data and get faster streaming ingestion. I know this sounds like that Dropbox is just rsync+cron post, but as soon as you replace pure insertions with upserts the gap should vanish.

Reply on news.ycombinator.com