System design · Advanced · 50 min
Fresh dashboards without rebuilding everything
Choose materializations from query patterns, correction rules, and refresh cost.
Task
Design a warehouse serving layer for a 3 TB event history and 100 analysts. Common dashboards query recent hourly metrics; finance closes a validated daily report. Updates can correct events from the previous week. Full rebuilds are too expensive to run every few minutes.
Requirements
- Define the raw, transformed, incremental, and published layers and their grains.
- Compare views, tables, and incremental materializations from measured query and refresh cost.
- Include late corrections, source deletes, quality gates, and rollback to a known-good release.
- Explain partition pruning, scheduling/readiness, metric definitions, and how you detect stale data.
MaterializationsWarehouse servingRefresh cost
Loading exercise…