Background

This is a small demo project that sets up an Iceberg datalake and allows you to run dbt-spark models on it. It was based on the work of Muhammed Irshad, with a few minor tweaks. The orignial hive-metastore/maria-db configuration was failing, so I used PostgreSQL and a different (relatively) lightweight hive-metastore image (naushadh/hive-metastore). Docker and docker-compose were replaced with podman and podman-compose.
With a few tweaks you can quickly:

• Reconfigure to a Delta Lake instead of Iceberg stack.
• Replace Minio with an AWS back-end if you want to use cloud instead of local storage.

My intention is to make this as light as possible so I can have my own little data-lake-in-a-box.

Instructions

Only podman and podman-compose need to be installed locally.
Everything else runs in containers (even python and dbt).
Podman can run rootless containers and you may need to check that you use. I needed to change the network backend from CNI to netavark to get it to work rootless.
This should get you started:

1. Change directory to the docker folder.
2. Run command docker-compose up -d.
3. Change directory to the dbt_iceberg_sample folder.
4. Run commmands:
   ./dbt-pod seed
./dbt-pod run
   or
   ./dbt-pod build
   to populate data lake.
5. Run the following command to start a beeline console, from where you can query the datalake in SQL:
   podman exec -it spark-thrift /usr/spark/bin/beeline -u “jdbc:hive2://spark-thrift:10000/test_schema;auth=noSasl” -n root
   Enter show tables; to see the tables.
   You can continue to execute SQL commands to interrogate the data.

Technology stack

Last tested with the following technologies:

• dbt - 1.8.2
• dbt-spark - 1.8.0
• Apache Spark and Hadoop - Spark 3.5.1 & Hadoop 3.3
• Apache Iceberg - 1.5.2
• Hive Standalone Metastore - 3.0.0
• PostgreSQL - 16.3
• Minio - RELEASE.2024-06-29T01-20-47Z
• Hadoop Aws - 3.3.3
• AWS Java SDK - 1.12.754
• podman - 4.9.3
• podman-compose - 1.0.6

Podman will pull the latest images when you build, so breaking changes may start to creep as things change.
You can update SparkThriftDockerfile and docker-compose.yml to lock the image versions to revert to a working configuration like the one above.

Acknowledgements

Thanks to Muhammed Irshad for leading the way:

• Medium article: Integrating dbt-spark with Apache Iceberg
• Github repo: irshadgit/dbt-spark-iceberg