Skip to main content

Getting Started

Prerequisites​

  • DuckDB v1.5.5 — the release line this extension is built on (Versions).
  • The mssql extension v0.2.5 or newer. Older versions answer DuckDB's main when DuckLake asks the catalog for its default schema, and the attach fails with Schema 'main' not found. INSTALL mssql FROM community gives a current version; SELECT mssql_version() says which one is loaded.
  • SQL Server 2019 or later, or Azure SQL Database — the catalog's strings need a UTF-8 collation (Requirements).
  • A place for the data files: a local directory, S3, Azure Blob Storage — anything DuckDB can write to. DuckLake puts only the catalog in SQL Server.

Step 1: install and load​

INSTALL mssql FROM community;
INSTALL mssql_ducklake FROM community;
LOAD mssql_ducklake;          -- loads mssql as well, and refuses if stock ducklake is already loaded

Two rules that come from embedding DuckLake:

  • Never together with the stock ducklake extension. Both provide the same functions and the same ducklake: attach prefix; LOAD mssql_ducklake refuses with a clear message when stock ducklake is loaded.
  • Load before the first ATTACH 'ducklake:…'. DuckDB autoloads an extension by attach prefix, and for ducklake: that is the stock one; once mssql_ducklake is loaded the prefix is taken.

Step 2: attach​

ATTACH 'ducklake:mssql:Server=localhost,1433;Database=lake_meta;User Id=sa;Password=…' AS lake
    (DATA_PATH '/data/lake/');

ducklake:mssql: is followed by an ADO-style connection string, exactly what the mssql extension takes. The URI form works too, with the type spelled out — DuckDB deliberately does not treat mssql:// as a prefix:

ATTACH 'ducklake:mssql://sa:…@localhost:1433/lake_meta' AS lake
    (DATA_PATH 's3://my-bucket/lake/', META_TYPE 'mssql');

The first attach creates the catalog: DuckLake's tables in the database's dbo schema, then the keys, collations and indexes this extension adds (Shaping). Every later attach finds it and opens it. METADATA_SCHEMA puts the catalog in another schema; the other attach options are DuckLake's own.

Step 3: use it​

CREATE SCHEMA lake.sales;
CREATE TABLE lake.sales.orders(id BIGINT, customer VARCHAR, amount DECIMAL(18, 2), day DATE);

INSERT INTO lake.sales.orders VALUES (1, 'acme', 10.5, DATE '2026-01-01');       -- inlined into the catalog
INSERT INTO lake.sales.orders
    SELECT i, 'c' || i, i * 1.5, DATE '2026-01-01' + i FROM range(10000) t(i);   -- a parquet file under DATA_PATH

UPDATE lake.sales.orders SET amount = amount * 2 WHERE id = 1;
DELETE FROM lake.sales.orders WHERE id > 9000;

SELECT * FROM ducklake_snapshots('lake');
SELECT count(*) FROM lake.sales.orders AT (VERSION => 2);
SELECT * FROM ducklake_table_info('lake');

Small inserts (10 rows by default) are inlined into the catalog — they land in a ducklake_inlined_data_* table in SQL Server — and larger ones go to Parquet files under DATA_PATH; ducklake_flush_inlined_data('lake') moves inlined rows to files. See Writing for what is specific to SQL Server on the write side.

Step 4: look at what landed​

Attach the catalog database with the mssql extension directly and the catalog is ordinary tables:

ATTACH 'Server=localhost,1433;Database=lake_meta;User Id=sa;Password=…' AS meta (TYPE mssql);
SELECT table_name FROM meta.dbo.ducklake_table;
FROM mssql_scan('meta', 'SELECT name FROM sys.tables WHERE name LIKE ''ducklake%'' ORDER BY name');

Next: the catalog in SQL Server, performance, and the settings.