Skip to content

lakehouse

lakehouse is the analytical read path. Trino queries Iceberg tables, and dbt declares the transformations on top of them.

The catalog homelab.sim_customers points at the Iceberg warehouse written by sim_customers_daily. Trino reads those tables. It does not publish a new copy back to that warehouse. Spark in the cluster remains the writer.

Silver and gold models, when built, land in the lakehouse's own local warehouse. That local warehouse is a separate hop from the cluster snapshot. A missing table in the local warehouse does not mean the cluster failed to write the snapshot, and a missing snapshot does not mean the lakehouse models are wrong. Check the writer first.

Checked-in schema files describe the bronze tables. They are contracts for those tables. Applying them is not what loads the cluster snapshot.