Project Scaffold

There are two ways to bootstrap a new project. Take the first for a bare start, the second for a skeleton that already runs.

1. dg scaffold defs dagster_rocky.RockyComponent <name>: uses Dagster’s built-in scaffolder via the registered entry point. Writes a bare defs.yaml.
2. init_rocky_project(target_dir): Python helper that writes a complete skeleton with defs.yaml, rocky.toml, .gitignore, models/, and a README.md.

dg scaffold (canonical)

Install dagster-rocky and the dg CLI finds RockyComponent on its own. The discovery comes from the [project.entry-points."dagster_dg_cli.registry_modules"] entry in dagster-rocky’s pyproject.toml.

uv add dagster-rocky
uv run dg list components | grep -i rocky
# dagster_rocky.RockyComponent

uv run dg scaffold defs dagster_rocky.RockyComponent my_pipeline

This writes a single my_pipeline/defs.yaml with type: dagster_rocky.RockyComponent.

init_rocky_project (richer)

For a complete project skeleton, use the Python helper:

from pathlib import Path
from dagster_rocky import init_rocky_project

init_rocky_project(Path("my_pipeline"))

After running, my_pipeline/ contains:

my_pipeline/
├── defs.yaml          # type: dagster_rocky.RockyComponent + attributes
├── rocky.toml         # DuckDB-backed full-refresh starter
├── .gitignore         # the DuckDB files and state the engine creates
├── models/            # empty, with .gitkeep
│   └── .gitkeep
└── README.md          # quickstart instructions

The default rocky.toml uses the DuckDB local-execution adapter. The scaffold therefore runs end to end with no warehouse credentials. It copies every table in each src__<source> schema of warehouse.duckdb into raw__<source>, with row_count and column_match checks.

The file also carries a commented-out incremental strategy and a [pipeline.main.checks.freshness] block. Turn them on once every source table has a load timestamp column. The freshness block then gives your Dagster assets a freshness policy.

Overwrite protection

init_rocky_project refuses to overwrite existing files by default; pass overwrite=True to force:

init_rocky_project(Path("my_pipeline"), overwrite=True)

This stops an accidental re-run from clobbering your own edits to rocky.toml, defs.yaml, or README.md.

Quickstart for end users

After scaffolding, the README guides users through:

# Install the rocky binary (once)
curl -fsSL https://raw.githubusercontent.com/rocky-data/rocky/main/engine/install.sh | bash

# Check the pipeline config (DuckDB, no credentials required)
rocky --config rocky.toml validate

# Run the pipeline locally
rocky --config rocky.toml run

# Launch the Dagster UI
dg dev