Databricks Naming Modelling Guidelines

Databricks Naming & Modelling Guidelines

These guidelines ensure consistent naming, metadata management, and modeling across Databricks within NZTA’s Data Intelligence Platform. They are intended for anyone performing data engineering or modeling in Databricks.

1. General Guidelines

• Metadata timestamps: Always use UTC for metadata columns (e.g., load time, update time) to prevent confusion with time zones or daylight savings.
• Snake case: Prefer snake_case for naming, except when object names derive directly from a source system.
• No plurals or abbreviations: Avoid plurals and nonstandard abbreviations in table and column names.

2. Metadata Columns

• Prefix metadata columns with double underscores, and use all uppercase (e.g., __LOAD_AT).
• Key columns for metadata tracking include:
  • __EXTRACT_AT (timestamp, mandatory): When data was extracted (UTC)
  • __LOAD_AT (timestamp, mandatory): When data was loaded (UTC)
  • __RECORD_SOURCE (string, mandatory): Source or filename of record
  • Others, as required: __SOURCE_FILE_PATH, __SEQUENCE_NUMBER, __EXTRACT_RUN_ID, __RESCUED_DATA

3. Table & Column Naming

Bronze Layer (Raw Source)

• Table names match the source object/file (lowercase enforced).
• If source lacks a specific name, use snake_case. Example: assessmentcontract_type.
• Keep column names as per source (including case), unless absent, then use snake_case.
• Do not use plurals/abbreviations (e.g., avoid assessments, adj_reasons).

Copper Layer (Historized Source)

• Table and column names match source. Minimal business rules/calculations.
• Track all changes (SCD2). Recommended to use Lakeflow Declarative Pipelines.
• Add metadata columns like __START_AT, __END_AT.

Silver Layer (Business Modeling)

• Schemas: Use silver_[business_area] for the schema name (e.g., silver_purchasing).
• Tables: Use snake_case table names within the silver schema; use the stg_ prefix for staging tables where needed.
• Use snake_case for both objects and columns.
• Rename columns for business clarity (e.g., contract_name not name).
• Avoid system-specific/non-standard abbreviations.
• Flags: Use boolean type, prefix with is_ or has_.
• Spelling: Use NZ English (e.g., centre, not center).

Gold Layer (Analytical)

• Star schema is preferred for analytical modeling.
• Table prefixes:
  • Dimensions: dim_
  • Facts: fact_
  • Snapshots: snapshot_
  • Summaries: summary_
• Surrogate keys: dim_<dimension>_key (e.g., dim_camera_site_key).
• Unknown row: Add default unknown row for dimensions (0 for integer, Unknown for string).

4. Column Suffixes

• For date/time columns, use clear suffixes: _date, _datetime, _time, _datetime_utc.
• Business dates/times: Store in NZ Standard Time unless specified as UTC.

5. Edge Cases

• Log files/unstructured data: Add row numbers and use metadata columns to manage duplicates or variable formats.
• Role-playing dimensions: Use views for date dimensions playing multiple roles.
• Fact-less fact tables: Use for many-to-many relationships; these have no measures.

6. Example Naming Conventions

Dimension Table:

dim_camera_site_key
site_code
site_name
stage_start_datetime
stage_end_datetime
__LOAD_AT
__UPDATE_AT
__START_AT
__END_AT
__IS_ACTIVE
__IS_LATEST
__CHANGE_HASH

Fact Table:

fact_incident_stage_history
incident_id
stage_code
stage_name
stage_start_datetime
stage_end_datetime
dim_incident_key
dim_stage_key
__LOAD_AT
__UPDATE_AT

Metadata Columns:

__EXTRACT_AT
__LOAD_AT
__RECORD_SOURCE

Date Dimension Example:

dim_date
date_key (YYYYMMDD)
date
month
year
week
__LOAD_AT

7. Source

Source page version: 11

Source: aiden-liu/confluence_page_to_instructions_skills — distributed by TomeVault.