motherduck-ducklake

Use DuckLake when you need open-table-format semantics, object storage as the source of truth, bring-your-own-bucket ownership, or file-aware maintenance. Native MotherDuck storage is often 2x-10x faster for reads, so avoid DuckLake just because a workload is large.

Run CREATE DATABASE my_ducklake (TYPE DUCKLAKE) for a fully managed instance. For BYOB, add DATA_PATH 's3://my-bucket/my-prefix/', and optionally set DATA_INLINING_ROW_LIMIT for small frequent writes.

The S3 bucket must be in the same AWS region as your MotherDuck organization: us-east-1 for US orgs and eu-central-1 for EU orgs. Other clouds are not currently supported for BYOB DuckLake storage.

No, MotherDuck documentation states DuckLake maintenance is not automatic. You must define who runs compaction and cleanup, from which compute surface, and on what schedule, using operations like CHECKPOINT and ducklake_flush_inlined_data.

Write permissions are effectively single-account at the database level today. Concurrent append-only writes can work, but concurrent updates, deletes, or DDL are much more constrained, and sharing is limited to auto-update read-only shares.

Attach the metadata database with ATTACH 'ducklake:md:__ducklake_metadata_<database_name>' AS my_ducklake. Only the database owner can attach it, and you must verify the DuckDB and DuckLake version matrix before direct catalog access.