Portable Data Stack with SQLMesh

This application is a containerized Analytics suite for an imaginary company selling postcards. The company sells both directly but also through resellers in the majority of European countries.

Stack

• Docker (docker compose)
• DuckDB
• SQLMesh using dbt adapter
• Superset

Interested in the data model?

Generation of example data and the underlying dbt-core model is available in the postcard-company-datamart project.

For other stacks, check the below:

• portable-data-stack-dagster
• portable-data-stack-airflow
• portable-data-stack-mage
• portable-data-stack-bruin
• postcard-company-dataform

System requirements

• Docker

Setup

1. Rename .env.example file to .env and set your desired Superset password. Remember to never commit files containing passwords or any other sensitive information.

   The SUPERSET_SECRET_KEY in .env.example is a placeholder that every fork of this repository shares, and Superset signs session cookies with it. Generate your own before the stack is reachable by anyone but you:

   openssl rand -base64 42

2. Rename shared/db/datamart.duckdb.example to shared/db/datamart.duckdb or init an empty database file there with that name.

3. With Docker engine installed, change directory to the root folder of the project (also the one that contains docker-compose.yml) and run

   docker compose up --build

4. Once the Docker suite has finished starting, you will see the output of the SQLMesh plan & apply commands, which run the dbt core models against DuckDB.

5. To explore the data and build dashboards you can open the Superset interface

Demo Credentials

Demo credentials are set in the .env file mentioned above.

Ports exposed locally

• Superset: 8088

Generated parquet files are saved in the shared/parquet folder.

The data is fictional and automatically generated. Any similarities with existing persons, entities, products or businesses are purely coincidental.

General flow

1. Test data is generated as parquet files using Python (generator)
2. Data is imported from parquet files to the staging area in the Data Warehouse (DuckDB)
3. The data is modelled, building fact and dimension tables, loading the Data Warehouse using SQLMesh (with the dbt adapter for model compatibility)
4. Analyze and visually explore the data using Superset, or query the DuckDB warehouse directly at shared/db/datamart.duckdb

For Superset, the default credentials are set in the .env file: user = admin, password = admin

Overview of architecture

The Docker process will begin building the application suite. The suite is made up of the following components, each within its own Docker container:

• generator: a Python script, from the postcard-company-datamart project, that generates the example data and exports it to parquet files
• sqlmesh-dbt: the data model, sourced from the postcard-company-datamart project
• superset: the web-based Business Intelligence application used to explore the data; exposed on port 8088; built from the postcard-company-datamart project, dashboard included.

The generator, the dbt project that sqlmesh-dbt loads and Superset with its dashboard are not kept in this repository. Docker builds them straight from postcard-company-datamart, by git URL, at the tag set by DATAMART_REF in docker-compose.yml, so every portable data stack runs the same model and the same dashboard. To build from a newer tag or a branch instead:

DATAMART_REF=main docker compose build

After the models have been applied you can either analyze the data using the querying and visualization tools provided by Superset (available locally on port 8088), or query the Data Warehouse (available as a DuckDB database).