Index

dbt-bouncer runs checks against artifacts from dbt. Every check also has a unique rule code (e.g. MO021) that can be used in place of its name. These checks fall into four categories:

• Catalog checks:
  • Catalog Seeds
  • Catalog Sources
  • Columns
    • Description
    • Naming
    • Tests
• Information Schema checks (dbt 2.0+):
  • Columns
  • Grain
  • Query
  • Sources
• Manifest checks:
  • Exposures
  • Lineage
  • Macros
  • Metadata
  • Models
    • Access
    • Code
    • Columns
    • Description
    • Directories
    • Lineage
    • Meta
    • Naming
    • Tags
    • Tests
    • Versioning
  • Seeds
  • Semantic Models
  • Snapshots
  • Sources
    • Description
    • Freshness
    • Lineage
    • Loader
    • Meta
    • Naming
    • Directories
    • Tags
    • Tests
  • Tests
  • Unit Tests
• Run Results checks:
  • Run Results

Information Schema checks#

dbt 2.0 and later can write the dbt Information Schema: a set of Parquet tables that describe the project, including column-level lineage, inferred column types and the grain of each model. The JSON artifacts do not contain this data. info_schema_checks read it from info_schema/v1/ in the dbt target directory, so dbt_artifacts_dir must point at a target directory that contains it.

To generate the Information Schema, add --generate-info-schema to a dbt command. Add --static-analysis strict to include column types and column-level lineage:

dbt build --static-analysis strict --generate-info-schema

dbt-bouncer reads info_schema/v1/ and not the Parquet files under target/private/:

• dbt documents info_schema/v1/ as a contracted interface with a versioned directory. dbt-bouncer refuses an Information Schema whose schema_version it does not support.
• target/private/ is undocumented. It also holds only the data of the last dbt command, so a later command such as dbt source freshness replaces the lineage that dbt build wrote.

If info_schema_checks are configured and the directory does not exist, for example with dbt 1.x artifacts, dbt-bouncer exits with an artifact error.

Known limitations, from dbt:

• Column-level lineage needs --static-analysis strict. dbt skips a model whose SQL it cannot analyse, and records the lineage of an ephemeral model on the models that select from it.
• dbt.node_columns.meta is empty, so checks that use column meta read it from manifest.json.
• Inferred column types use Apache Arrow names, for example Int64 (dbt-labs/dbt#16515).