Portable Data Stack

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

Stack

• Dagster
• Docker (Docker Compose)
• DuckDB
• dbt core
• 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 using the same dbt-core model, check the below:

• portable-data-stack-mage
• portable-data-stack-airflow
• portable-data-stack-sqlmesh

Implementations with other tools with a different model:

• portable-data-stack-bruin
• postcard-company-dataform

For the legacy version involving OLTP, CSV and JSON sources, check out the legacy-oltp branch.

System requirements

• Docker

Setup

1. Rename .env.example file to .env and set your desired 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

   Note that this may take several minutes to completed. Check out the console to see when the Dagster interface is ready.

4. Once the Docker suite has finished loading, open up Dagster (dagit) , go to Assets, select all and click Materialize selected

1. When the assets have been materialized, you can open the Superset interface

Demo Credentials

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

Ports exposed locally

• Dagster (dagit): 3000
• Superset: 8088

Generated parquet files are saved in the shared folder.

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

General flow

1. Generate test data as parquet files using Python
2. Import data to the staging area in the Data Warehouse (DuckDB), orchestrated by Dagster
3. Model data, build fact and dimension tables, load the Data Warehouse using dbt
   • installs dbt dependencies
   • seeds the database with static data (e.g. geography)
   • runs the model
   • tests the model
4. Analyze and visually explore the data using Superset or directly query the Data Warehouse database instance

For superset, the default credentials are: 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
• dbt: copies the data model from the postcard-company-datamart project into dbt/postcard_company, where Dagster runs it. Edits you make there survive restarts; they are replaced only when DATAMART_REF moves to a different version of the model.
• dagster: this is the orchestrator tool that will trigger the ETL tasks; its GUI is locally available on port 3000;
• superset: this contains the web-based Business Intelligence application we will use to explore the data; exposed on port 8088; built from the postcard-company-datamart project, dashboard included.

The generator, the dbt project 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

Once the Docker building process has completed, we may open the Dagster GUI (locally: localhost:3000) to view and materialize our assets.

After the assets have been materialized 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)

Credits

Inspired by:

• Build a poor man’s data lake from scratch with DuckDB
• Using dbt with Dagster software-defined assets