vgi-easter 0.1.4

A minimal VGI (Vector Gateway Interface) worker exposing a SQL easter_date(year) scalar function to DuckDB

vgi-easter

A tiny VGI (Vector Gateway Interface) worker that gives DuckDB one SQL function — easter_date(year) — returning the date of Western (Gregorian) Easter Sunday. It has no external data and almost no code, which makes it a clean, copyable example of a VGI scalar-function worker.

      

Quick start

In DuckDB:

-- VGI isn't bundled with DuckDB yet, so load it from the community channel
-- (INSTALL is a one-time download; LOAD is once per session).
INSTALL vgi FROM community;
LOAD vgi;

ATTACH 'easter' (TYPE 'vgi', LOCATION 'uvx vgi-easter');

SELECT easter.easter_date(2025);   -- 2025-04-20

SELECT year, easter.easter_date(year) AS easter
FROM range(2020, 2025) t(year);

DuckDB launches the worker for you, and easter_date then behaves like a native function (a null year yields a null date). The uvx vgi-easter location fetches the worker on demand; to keep it around, pip install vgi-easter.

How it works

A VGI worker publishes catalogs, schemas, and functions that DuckDB can ATTACH and query as if they were built in. Values cross the boundary as Apache Arrow, so they stay columnar end to end.

This worker publishes a single function:

easter                                  (catalog)
└── main                                (schema)
    └── easter_date(year BIGINT) → DATE

The whole implementation is ~160 lines in easter_worker.py: the date calculation (_easter_sunday, the Anonymous Gregorian Computus — pure standard library), a ScalarFunction that maps an Arrow array of years to dates, and a few lines wiring it into a catalog.

Running it

Once installed, the package gives you one command per VGI transport:

Command	Transport	Use it when
vgi-easter	stdio	DuckDB spawns the worker as a subprocess (the quickstart)
vgi-easter-http	HTTP	you want a long-running server to attach to

To run over HTTP, start the server and attach to its URL:

VGI_SIGNING_KEY=dev vgi-easter-http --host 0.0.0.0 --port 8000

LOAD vgi;   -- after a one-time INSTALL vgi FROM community
ATTACH 'easter' (TYPE 'vgi', LOCATION 'http://localhost:8000');

Working from a checkout instead? Both modules carry PEP 723 metadata, so uv run easter_worker.py (stdio) and uv run serve.py (HTTP) run without installing anything.

Configuration

Variable	Purpose
VGI_SIGNING_KEY	Signing key for HTTP state tokens (required by the HTTP server).
VGI_HTTP_HOST / VGI_HTTP_PORT	HTTP bind address (default: all interfaces / 8000).
VGI_EASTER_GIT_COMMIT	Reported as the catalog's implementation_version.
VGI_WORKER_DEBUG	Set to 1 for debug logging.

Development

Requires Python 3.13+ and uv; the only dependency is vgi-python.

uv run --frozen pytest tests/ -q

The tests/ suite covers the Easter calculation (including the March 22 / April 25 extremes) and the Arrow compute path. A separate sqllogictest suite in test/sql/ drives the worker through the real DuckDB vgi extension. CI runs both on Linux, macOS, and Windows — see ci/README.md.

Releasing

vgi-easter is built with hatchling and published to PyPI by CI when a GitHub Release is created (it runs the test suites, then uv build && uv publish). To cut a release, bump version in pyproject.toml and publish a GitHub Release.

Built with

• DuckDB attaches and queries the worker — install the VGI extension with INSTALL vgi FROM community; LOAD vgi;.
• vgi-python is the VGI worker SDK this is built on.
• Haybarn provides the DuckDB-compatible unittest runner that drives the cross-platform CI.

License

MIT — see LICENSE. Copyright 2026 Query Farm LLC — https://query.farm