3 ms·

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

• orthoxerox 6mo ago
  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.
  • cbility 6mo ago
    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/#we-need-to-talk-about-iceberg https://ducklake.select/2026/04/02/data-inlining-in-ducklake... [2] https://ducklake.select/manifesto/#features https://ducklake.select/manifesto/#features
    • orthoxerox 6mo ago
      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.