Released v0.9.0 — 2026-10-07
Focus: Performance and correctness hardening.

• Materialized statistics CTE, hoisted casts, and partition pre-filtering cut file listing ~400×: 2,703 → 6.8 ms.
• UPDATE pushes predicates into Parquet; keyed DELETE reads only predicate columns (~10 based counts with delete files; one read per cold footer.
• Added DuckLake comments/tags, more object-store URL schemes, batch file registration, and snapshot metadata.

github.com/datafusion-con…

Here is the playlist of San Francisco DataFusion meetup.

Thanks to all the presenters from:
@LangChain @LakeSail @columnar_tech @datadoghq @hotdata_dev

youtube.com/playlist?list=…

cc: @andrewlamb1111 @ApacheDataFusio

New features in the CLI: fork your database like a Git branch.

We’ve also added a dialect option to the query command, so you can run queries using DuckDB, Postgres, or Snowflake syntax.

We've released major updates to datafusion-ducklake.

• Time travel by snapshot or timestamp
• Partition-aware writes and updates
• Atomic multi-file commits
• Improved CDC support for renamed and nested columns
• SORTED BY for writes, updates, and compaction

Check it out GhLA7