etlantic 0.17.0

Typed, contract-driven data pipeline modeling for Python.

ETLantic

Design once. Validate everywhere.
Typed, contract-driven data pipelines for Python.

Documentation · Quickstart · Capabilities · Roadmap

---

ETLantic catches incompatible wiring before data is processed. Define datasets, transformations, and pipelines as typed Python classes, then validate, plan, run, or compile the same logical pipeline for different execution engines.

Typed contracts ──▶ Validation ──▶ Deterministic plan ──▶ Run or compile

Why ETLantic?

• Fail earlier. Detect broken references, incompatible contracts, missing implementations, unsupported capabilities, and untrusted plugins before a write occurs.
• Keep logic portable. Separate logical pipeline structure from local, Polars, Pandas, SQL, PySpark, and orchestration implementations.
• Make plans reviewable. Generate deterministic, immutable, secret-free execution plans with stable fingerprints.
• Preserve evidence. Produce structured diagnostics, lineage, schema observations, and run reports instead of opaque task logs.
• Adopt incrementally. The core has no dataframe, SQL, Spark, or Airflow dependency. Install only the integrations you need.

Project status: 0.17.0 is production/stable for documented single-tenant reference deployments. The local runtime and reference plugins are available today. Structured Streaming remains experimental. Portable transformation authoring and Polars/PySpark/Pandas/SQL relational compilers plus the public conformance SDK are available; advanced portable profiles shipped under 0.17 where documented. Multi-tenant isolation, deployment topology, compliance/SBOM/signing, and advanced supply-chain controls remain adopter-owned. See the capabilities guide before choosing a production architecture.

Quickstart

ETLantic requires Python 3.11 or newer.

pip install etlantic
etlantic --version

Create pipeline.py:

from etlantic import (
    Data,
    Extract,
    Input,
    Load,
    Output,
    Pipeline,
    PipelineRuntime,
    Transformation,
)


class RawCustomer(Data):
    customer_id: int
    first_name: str
    last_name: str


class Customer(Data):
    customer_id: int
    full_name: str


class NormalizeCustomers(Transformation):
    customers: Input[RawCustomer]
    result: Output[Customer]


@NormalizeCustomers.implementation("local")
def normalize_customers(customers: list[RawCustomer]) -> list[Customer]:
    return [
        Customer(
            customer_id=row.customer_id,
            full_name=f"{row.first_name} {row.last_name}",
        )
        for row in customers
    ]


class CustomerPipeline(Pipeline):
    raw: Extract[RawCustomer] = Extract(asset="customer_source")
    normalized = NormalizeCustomers.step(customers=raw)
    curated: Load[Customer] = Load(
        input=normalized.result,
        asset="customer_sink",
    )


def main() -> None:
    # Validation and planning do not execute transformation code.
    CustomerPipeline.validate(profile="development").raise_for_errors()
    plan = CustomerPipeline.plan(profile="development")
    print(plan.fingerprint)

    runtime = PipelineRuntime()
    runtime.memory.seed(
        "customer_source",
        [RawCustomer(customer_id=1, first_name="Ada", last_name="Lovelace")],
    )
    report = CustomerPipeline.run(profile="development", runtime=runtime)

    print(report.status)  # succeeded
    print(runtime.memory.get("customer_sink")[0].model_dump())
    # {"customer_id": 1, "full_name": "Ada Lovelace"}


if __name__ == "__main__":
    main()

Keep contracts, the transformation registration, and CustomerPipeline at module scope so the CLI can import them. Put validate/seed/run under if __name__ == "__main__" so etlantic validate / plan do not execute the pipeline during import.

Change the sink contract to an incompatible type and validate() returns a structured diagnostic before any transformation or write is attempted.

The complete tested example is examples/quickstart.py.

CLI workflow

The CLI follows the same validate-first lifecycle. Validation and planning do not require seeded data; in-memory execution does, so run the seeded example with Python:

# Inspect and validate a pipeline (import-safe when side effects are guarded)
etlantic inspect pipeline.py:CustomerPipeline --format json
etlantic validate pipeline.py:CustomerPipeline --profile development --format json

# Build and explain a deterministic execution plan
etlantic plan pipeline.py:CustomerPipeline --profile development --format json
etlantic plan explain pipeline.py:CustomerPipeline --profile development --format json

# Execute the seeded in-memory example
python pipeline.py

# Emit CI diagnostics
etlantic validate pipeline.py:CustomerPipeline --profile development --format sarif

Airflow compilation requires the optional etlantic-airflow package:

pip install "etlantic[airflow]"
etlantic compile pipeline.py:CustomerPipeline --target airflow -o dags/

Other public command groups cover contract generation and diffs, plugins, schema drift, reliability, visualization, and reports. Run etlantic --help for the complete command surface.

Choose an engine

Start with the core package, then add engines as needed:

pip install "etlantic[polars]"
pip install "etlantic[pandas]"
pip install "etlantic[sql]"
pip install "etlantic[pyspark]"
pip install "etlantic[airflow]"
pip install "etlantic[prefect]"

Integration	Package	Purpose
Polars	etlantic-polars	Eager/lazy dataframe execution and portable kernel compilation
Pandas	etlantic-pandas	Eager dataframe execution
SQL	etlantic-sql	Parameterized relational execution and SQL-to-SQL plans
PySpark	etlantic-pyspark	Spark execution and local session provider
Airflow	etlantic-airflow	Compile plans into Airflow DAG artifacts
Prefect	etlantic-prefect	Optional direct-execution scheduler (local MVP)
Keyring	etlantic-keyring	Resolve runtime secrets from the OS keyring
SQLModel	etlantic-sqlmodel	Bridge ContractModel schemas and SQLModel
SparkForge	etlantic-sparkforge	Migrate SparkForge pipeline definitions

Plugins are discovered through Python entry points and scoped to a runtime registry. Production profiles require an explicit plugin allowlist and reject untrusted plugins by default.

How it works

ETLantic keeps logical intent separate from physical execution:

1. Author typed Data, Transformation, and Pipeline classes.
2. Inspect an immutable logical graph without running user code.
3. Validate structure, references, contracts, policies, capabilities, and plugin trust in ordered phases.
4. Plan engine selections, execution regions, bindings, artifacts, and materialization boundaries.
5. Execute or compile the plan through small backend protocols.
6. Report step outcomes, diagnostics, lineage, artifacts, and schema observations.

Plans and reports contain secret references, never resolved secret values. Secrets are resolved only at runtime. Capability and trust failures occur before mutation.

Capability boundary

Capability	0.17
Typed modeling, validation, contracts, and deterministic planning	Available
Local Python execution and structured run reports	Available
Memory, callable, JSON, CSV, and no-write storage	Available
Polars and Pandas dataframe plugins	Available
SQL and PySpark plugins	Available
Airflow plan compiler	Available
ODCS, DTCS, and DPCS interchange	Available
Schema drift, reliability, visualization, and SARIF tooling	Available
Production plugin allowlists and runtime secret providers	Available
Portable transformation authoring	Available
Polars + PySpark + Pandas + SQL portable compilers (kernel + relational /1)	Available
Public portable transform conformance SDK	Available
Structured Streaming	Experimental
Advanced portable profile graduation	Available on Polars + PySpark (0.17); Pandas/SQL baseline only

See Capabilities and Limitations and the roadmap for the precise support boundary.

Documentation

• Hosted documentation
• Getting Started
• Current 0.17 User Guide
• Quickstart
• Core Concepts
• Architecture
• Portable Transformations
• Contributing
• Roadmap

Development

The repository uses uv for its workspace and development environment:

git clone https://github.com/eddiethedean/etlantic.git
cd etlantic
uv sync
uv run python examples/quickstart.py
uv run pytest
uv run ruff check .
uv run ruff format --check .
uv run mkdocs serve

See CONTRIBUTING.md for package-specific test groups and development conventions.

License

MIT