Home › Data engineering › Metric views
Data engineering · BI and analytics · free guide
Metric views, taught from zero
Ask three teams for "revenue last quarter" and you often get three numbers. Each team wrote its own SQL: one forgot returns, one counted cancelled orders, one joined a table that doubled rows. A semantic model fixes this by writing each business number once, with its recipe, and making every dashboard, spreadsheet and AI agent ask for the number by name.
This section teaches the idea first, with no product in it. Then it shows three products that build the idea, Databricks metric views, the dbt Semantic Layer and the Power BI semantic model, and where each falls short. Then it goes deep on metric views, and ends with one example built end to end.
The semantic model is the menu between the kitchen (tables) and the diners (tools and people).
The path
Ten pages, in this order
- 1 · The idea: What a semantic model is, before any product
The semantic model with no product in it - 2 · Three products: One idea, three products: metric views, dbt, Power BI
How Unity Catalog metric views, the dbt Semantic Layer (MetricFlow) and the Power BI semantic model each build the ten parts of a semantic model, with the same metric written in all three, and where Cube, LookML, Snowflake semantic views and Apache Ossie fit. - 3 · Limits: Limits of each, seen from eight seats
The limits of metric views, the dbt Semantic Layer and the Power BI semantic model from the seat of the author, business user, BI tool, AI agent, security, performance, cost and change, with a Pursue/Drop verdict for common setups. - 4 · Parts: The parts of a metric view
Every part of a Unity Catalog metric view - 5 · Grain: Grain, fan-out and the kinds of measure
Granularity for metric views - 6 · Row-level security: Row-level security with metric views
Where row-level security lives for metric views - 7 · Speed: Speed: materialization, pre-aggregation and layout
How a metric view query runs and what makes it fast - 8 · Production: Metric views in production: challenges, anti-patterns, best practices
Running Unity Catalog metric views in production - 9 · Example: One example end to end: tyre sales to dealers
A complete worked example
Words used
A small glossary
- Semantic model
- A layer of business meaning on top of tables: which numbers exist, how each is calculated, how tables join, and how you may slice them. Also called a semantic layer or metrics layer.
- Metric view
- Databricks' product for a semantic model: a Unity Catalog object, written in YAML, that holds measures and dimensions and is queried with MEASURE().
- Grain
- What one row of a table means, such as one invoice line or one dealer per day.
- Dimension (field)
- A way to slice a number: region, month, product segment. Metric views call them fields.
- Measure
- An aggregation such as SUM(net_amount). It has no fixed grouping; the query decides it.
- Composed measure
- A measure built from other measures, such as margin = gross margin / net revenue.
- Fan-out
- When a join repeats rows, so sums come out too big.
- Row-level security (RLS)
- Rules that decide which rows each person sees.
- Materialization
- Pre-computed results that the engine reads instead of the raw tables when a query fits.
- MetricFlow
- dbt's open-source engine that turns a metric request into SQL.
- DAX
- Power BI's formula language for measures.
Related
Where it connects
The semantic layer (key concept) · Power BI vs Databricks AI/BI · Genie and metric views · Gold: shaped for the business · All data engineering
FAQ
Questions people ask
What is a Databricks metric view?
A Unity Catalog object that defines business measures and dimensions once, in YAML, over a source table and its joins. You query it with SELECT field, MEASURE(measure) ... GROUP BY ALL, and Unity Catalog turns that into SQL on the source tables.
Is a metric view the same as a normal view?
No. A normal view fixes the grouping when you create it. A metric view keeps measures unaggregated until query time, so the same measure can be grouped by any field the question needs.
Do I still need gold tables if I have metric views?
Yes. Metric views sit on top of clean, atomic gold tables. They hold the meaning and the recipes, not the data cleaning.
Sources
Checked against (11 Nov 2026)
- Databricks: Unity Catalog metric views and the YAML reference
- Materialization · Advanced techniques · Relationships · BI tools · Feature availability
- Row filters and column masks · ABAC evaluation through views (KB)
- dbt semantic models · MetricFlow open-sourced · Apache Ossie
Metric views change every few months. Confirm runtime minimums and preview status before a design depends on them.