Дизайн data mart на собеседовании Data Engineer

Что такое data mart

Slice of DWH под specific business unit / use case.

Marketing mart: campaigns, leads, attribution.
Finance mart: revenue, costs, P&L.
Operations mart: orders, fulfillment, inventory.

Pre-modeled, optimized под queries unit.

Subject-oriented

One business subject area. Не try to cover everything.

mkt_campaigns_fact
mkt_leads_dim
mkt_attribution_funnel

Naming reveals subject. Easy для аналитика discover.

Materialization strategy

Views. Always-fresh, no extra storage. Slow для heavy queries.

Tables. Materialized, fast. Need refresh.

Incremental tables. Only update new partitions.

Materialized views. Auto-refresh, mid-ground.

В DWH (Snowflake / CH / BQ) — typically incremental tables.

Refresh strategy

Hourly / daily. Standard для marts.

On-demand. When source updated.

Mixed. Critical metrics — frequent. Less critical — daily.

Documentation. Explicit refresh schedule. Аналитики know what fresh.

Ownership

Каждый mart — owner team.

Marketing team owns Marketing mart.

• Schema decisions.
• Quality standards.
• Stakeholder communication.

В Data Mesh — это central principle.

DE provides platform, tooling. Domain teams own marts.

Связанные темы

• Inmon vs Kimball для DE
• DWH stages для DE
• Star schema vs Snowflake для DE
• Data Mesh для DE
• Подготовка к собесу Data Engineer

FAQ

Это официальная информация?

Нет. Статья основана на стандартных DWH practices.

Тренируйте Data Engineering — откройте тренажёр с 1500+ вопросами для собесов.