Crosshire
← Back to Databricks
Crosshire / Learn · Tech · Databricks

DWH Handbook

How a team designs, builds, tests, governs and operates a data warehouse on Databricks. This is not a platform tutorial — the generic material is in the Databricks docs and is linked, not restated. What is here is the 90% a warehouse team hits every week, plus the 10% that separates someone who has shipped one from someone who has read about it.

23 sections124 topics17 tables in one worked modelplatform facts verified 29.07.2026

How to read it

Every subject is one block, six parts

What it is, precisely enough to act on, with runnable code. Why here — the concrete failure it prevents in this warehouse, not generic benefits. Then two lanes: Doing it, what you do with your hands, and Embedding it, how a lead gets the team to own it instead of watching. Then the Defaults, and the Gotchas — each one naming what actually breaks and how you notice.

One worked model, everywhere

Every code sample on every page uses the same warehouse: a B2B distributor selling into DE, AT, CH and NL, with SAP and a webshop as sources. No page invents its own table names. That model supplies the four problems a real warehouse has and a toy example never does — a full extract with no change timestamp, CDC, semi-structured files, and multi-currency conversion.

Dated facts, not confident guesses

Databricks renames and re-stages things faster than a handbook can be rewritten. Every volatile claim is anchored to the currency register with the date it was checked and a link to the official source. 6 entries are currently preview, beta or unverified and are marked as such — do not build a client commitment on one without re-checking.

Foundation

The decisions that are expensive to reverse: layers, model, ingestion, layout, semantics.

Build

How code is structured, shipped and proven — standards and the four kinds of test.

Run

Knowing the warehouse is right, not merely that the jobs went green.

Govern

Unity Catalog as the control plane: topology, access, classification, evidence.

Engagement

How the work gets specified, written down, and handed over to a team that owns it.

Reference

Verified facts, system tables, glossary, sources.

The audit is one part of this

An engagement usually opens with an audit, because it converts opinion about the platform into evidence before anyone proposes changes. This handbook covers what you sweep and why; the actual SQL is a separate library of 100 copy-paste queries built entirely on system.* tables.

Databricks Audit — 100 system-tables queries →