Laktory

An open-source DataOps and dataframe-centric ETL framework for building lakehouses.

Laktory is your all-in-one solution for defining both data transformations and Databricks resources. Imagine if Terraform, Databricks Asset Bundles, and dbt combined forces—that’s essentially Laktory.

This open-source framework simplifies the creation, deployment, and execution of data pipelines while adhering to essential DevOps practices like version control, code reviews, and CI/CD integration. With Apache Spark and Polars driving its data transformation, Laktory ensures reliable and scalable data processing. Its modular, flexible approach allows you to seamlessly combine SQL statements with DataFrame operations.

Since Laktory pipelines are built on top of Spark and Polars, they can run in any environment that supports python—from your local machine to a Kubernetes cluster. They can also be deployed and orchestrated as Databricks Jobs or Delta Live Tables, offering a simple, fully managed, and low-maintenance solution.

But Laktory goes beyond data pipelines. It empowers you to define and deploy your entire Databricks data platform—from Unity Catalog and access grants to compute and quality monitoring—providing a complete, modern solution for data platform management. This empowers your data team to take full ownership of the solution, eliminating the need to juggle multiple technologies. Say goodbye to relying on external Terraform experts to handle compute, workspace configuration, and Unity Catalog, while your data engineers and analysts try to combine Databricks Asset Bundles and dbt to build data pipelines. Laktory consolidates these functions, simplifying the entire process and reducing the overall cost.

Help

See documentation for more details.

Installation

Install using

pip install laktory

For more installation options, see the Install section in the documentation.

A Basic Example

from laktory import models


node_brz = models.PipelineNode(
    name="brz_stock_prices",
    source={
        "format": "PARQUET",
        "path": "./data/brz_stock_prices/"
    },
    transformer={
        "nodes": [
        ]
    }
)

node_slv = models.PipelineNode(
    name="slv_stock_prices",
    source={
        "node_name": "brz_stock_prices"
    },
    sink={
        "path": "./data/slv_stock_prices",
        "mode": "OVERWRITE",
        "format": "PARQUET",
    },
    transformer={
        "nodes": [
            
            # SQL Transformation
            {
                "sql_expr": """
                    SELECT
                      data.created_at AS created_at,
                      data.symbol AS symbol,
                      data.open AS open,
                      data.close AS close,
                      data.high AS high,
                      data.low AS low,
                      data.volume AS volume
                    FROM
                      {df}
                """   
            },
            
            # Spark Transformation
            {
                "func_name": "drop_duplicates",
                "func_kwargs": {
                    "subset": ["created_at", "symbol"]
                }
            },
        ]
    }
)

pipeline = models.Pipeline(
    name="stock_prices",
    nodes=[node_brz, node_slv],
)

print(pipeline)
#> resource_name_=None options=ResourceOptions(variables={}, depends_on=[], provider=None, aliases=None, delete_before_replace=True, ignore_changes=None, import_=None, parent=None, replace_on_changes=None) variables={} databricks_job=None dlt=None name='stock_prices' nodes=[PipelineNode(variables={}, add_layer_columns=True, dlt_template='DEFAULT', description=None, drop_duplicates=None, drop_source_columns=False, transformer=SparkChain(variables={}, nodes=[SparkChainNode(variables={}, allow_missing_column_args=False, column=None, spark_func_args=[SparkFuncArg(variables={}, value='symbol'), SparkFuncArg(variables={}, value='timestamp'), SparkFuncArg(variables={}, value='open'), SparkFuncArg(variables={}, value='close')], spark_func_kwargs={}, spark_func_name='select', sql_expression=None)]), expectations=[], layer='BRONZE', name='brz_stock_prices', primary_key=None, sink=None, source=FileDataSource(variables={}, as_stream=False, broadcast=False, cdc=None, dataframe_type='SPARK', drops=None, filter=None, mock_df=None, renames=None, selects=None, watermark=None, format='PARQUET', header=True, multiline=False, path='./data/brz_stock_prices/', read_options={}, schema_location=None), timestamp_key=None), PipelineNode(variables={}, add_layer_columns=True, dlt_template='DEFAULT', description=None, drop_duplicates=None, drop_source_columns=True, transformer=SparkChain(variables={}, nodes=[SparkChainNode(variables={}, allow_missing_column_args=False, column=None, spark_func_args=[], spark_func_kwargs={'subset': SparkFuncArg(variables={}, value=['timestamp', 'symbol'])}, spark_func_name='drop_duplicates', sql_expression=None)]), expectations=[], layer='SILVER', name='slv_stock_prices', primary_key=None, sink=FileDataSink(variables={}, mode='OVERWRITE', checkpoint_location=None, format='PARQUET', path='./data/slv_stock_prices', write_options={}), source=PipelineNodeDataSource(variables={}, as_stream=False, broadcast=False, cdc=None, dataframe_type='SPARK', drops=None, filter=None, mock_df=None, renames=None, selects=None, watermark=None, node_name='brz_stock_prices', node=PipelineNode(variables={}, add_layer_columns=True, dlt_template='DEFAULT', description=None, drop_duplicates=None, drop_source_columns=False, transformer=SparkChain(variables={}, nodes=[SparkChainNode(variables={}, allow_missing_column_args=False, column=None, spark_func_args=[SparkFuncArg(variables={}, value='symbol'), SparkFuncArg(variables={}, value='timestamp'), SparkFuncArg(variables={}, value='open'), SparkFuncArg(variables={}, value='close')], spark_func_kwargs={}, spark_func_name='select', sql_expression=None)]), expectations=[], layer='BRONZE', name='brz_stock_prices', primary_key=None, sink=None, source=FileDataSource(variables={}, as_stream=False, broadcast=False, cdc=None, dataframe_type='SPARK', drops=None, filter=None, mock_df=None, renames=None, selects=None, watermark=None, format='PARQUET', header=True, multiline=False, path='./data/brz_stock_prices/', read_options={}, schema_location=None), timestamp_key=None)), timestamp_key=None)] orchestrator=None udfs=[]

pipeline.execute(spark=spark)

To get started with a more useful example, jump into the Quickstart.

A Lakehouse DataOps Template

A comprehensive template on how to deploy a lakehouse as code using Laktory is maintained here: https://github.com/okube-ai/lakehouse-as-code.

In this template, 4 pulumi projects are used to:

• {cloud_provider}_infra: Deploy the required resources on your cloud provider
• unity-catalog: Setup users, groups, catalogs, schemas and manage grants
• workspace: Setup secrets, clusters and warehouses and common files/notebooks
• workflows: The data workflows to build your lakehouse

Okube Company

Okube is dedicated to building open source frameworks, known as the kubes, empowering businesses to build, deploy and operate highly scalable data platforms and AI models.