Skip to main content
sqb dbt diff compares the relations your dbt build produces against a production baseline, so you can confirm dev matches prod before you promote. It resolves the production-shaped relations the same way reuse does (by compiling your dbt project at a configured git ref), then compares each model. Every run requires exactly one comparison mode: 
sqb dbt diff --schema-only
sqb dbt diff --full
sqb dbt diff --bounded <value>
Running without a mode is an error.

​
Modes

​
--schema-only

Compares column names and types per model, without reading row data. Fast, needs no key, and works on any model: 
     Model  stg_orders
       Key  <not used>
Comparison  schema-only

Schemas
- - - - -
No schema differences.
Use this as a cheap structural check: it catches added, dropped, or retyped columns across the build without scanning rows.

​
--full

Compares rows. Full row comparison needs a key to align rows between the two sides, so each compared model must define config.unique_key in its dbt config: 
error[C341]: dbt diff requires model 'agg_daily_revenue' to define config.unique_key for row comparison
  = help: Add unique_key to the dbt model config, or run sqb dbt diff --schema-only.
Add unique_key to the model config to enable a full row diff, or fall back to --schema-only for models without a natural key.

​
--bounded

Compares only a recent window of rows instead of the whole table, using the model’s SQLBuild cursor metadata. This keeps the comparison cheap on large relations. The value depends on the cursor kind:
  • Timestamp cursor: a duration like 30d, 12h, or 15m - compares the last N of time.
  • Integer cursor: an integer bound.
sqb dbt diff --bounded 30d
The model must define SQLBuild cursor metadata for --bounded to apply; without it, SQLBuild reports a configuration error pointing you to add cursor metadata or use another mode.

​
On this topic

  • Reuse from production - how the production baseline is resolved and reused.
  • Clone - copy or clone production relations into a target.
  • Testing - unit tests against dbt models.