Introduction

dagster-rocky is a thin adapter. It calls the rocky command-line binary through rocky-sdk’s RockyClient, then turns each result into Dagster assets and asset checks.

Rocky and Dagster do different jobs. Rocky checks and runs your SQL. It type-checks each model, traces column lineage, and detects schema drift. It also enforces compile-time contracts, which are schema agreements it checks before any row is written.

Dagster schedules the work, retries it, alerts on it, and draws the asset graph. dagster-rocky reports each Rocky result as a native Dagster event, so the asset graph shows what Rocky checked.

RockyResource builds a RockyClient from your config and delegates every command to it. To drive Rocky from a notebook, a script, or a non-Dagster orchestrator, use the SDK directly.

Quick start

There are two ways to wire Rocky into Dagster. Start with the component. It discovers your tables for you.

Option A: component (defs.yaml):

type: dagster_rocky.RockyComponent
attributes:
  binary_path: rocky
  config_path: config/rocky.toml
  models_dir: models

Option B: resource + asset:

import dagster as dg
from dagster_rocky import RockyResource

rocky = RockyResource(binary_path="rocky", config_path="config/rocky.toml")

@dg.asset
def acme_orders(rocky: RockyResource) -> dg.MaterializeResult:
    result = rocky.run(filter="tenant=acme")
    return dg.MaterializeResult(
        metadata={"tables_copied": result.tables_copied, "duration_ms": result.duration_ms},
    )

defs = dg.Definitions(assets=[acme_orders], resources={"rocky": rocky})

What it provides

Symbol	Purpose
RockyResource	ConfigurableResource wrapping the CLI; 25+ methods; three run modes (buffered, streaming, Pipes)
RockyComponent	State-backed component that caches discovery; dag_mode=True builds connected asset graphs
RockyDagsterTranslator	Customize asset keys, groups, tags, and metadata per Rocky table
load_rocky_assets()	Returns one AssetSpec per enabled Rocky table
emit_check_results() / emit_materializations()	Convert Rocky results into Dagster events

Architecture

Every Rocky call travels down the same chain.

  ┌────────────────────────────────────────────────────────┐
  │ Dagster asset or check                                 │
  │   your code calls rocky.run(...) on the resource       │
  └───────────────────────────┬────────────────────────────┘
                              │ Python method call
                              ▼
  ┌────────────────────────────────────────────────────────┐
  │ RockyResource          (dagster-rocky)                 │
  │   adds the Dagster parts: logging, Pipes, dg.Failure   │
  └───────────────────────────┬────────────────────────────┘
                              │ Python method call
                              ▼
  ┌────────────────────────────────────────────────────────┐
  │ RockyClient            (rocky-sdk)                     │
  │   builds the argument list, parses stdout into types   │
  └───────────────────────────┬────────────────────────────┘
                              │ subprocess:
                              │ rocky run --output json
                              ▼
  ┌────────────────────────────────────────────────────────┐
  │ rocky CLI              (Rust binary)                   │
  │   checks the project, then executes the command        │
  └───────────────────────────┬────────────────────────────┘
                              │ warehouse operations
                              ▼
  ┌────────────────────────────────────────────────────────┐
  │ your warehouse                                         │
  │   DuckDB, Databricks, Snowflake, BigQuery, and others  │
  └────────────────────────────────────────────────────────┘

Results travel back up the same chain. The CLI prints typed JSON on stdout. RockyClient parses that JSON into Pydantic models. RockyResource turns the models into asset materializations, asset check results, and metadata.

Requirements

• dagster >= 1.13.8
• rocky-sdk >= 0.6.0
• pydantic >= 2.0
• pygments >= 2.20.0
• The rocky binary must be available on PATH (or configured via binary_path). For deployment, you can vendor the binary under a vendor/ directory and point binary_path to it.

RockyResource exposes one Python method per Rocky CLI command. See the RockyResource page for the full method list and signatures.