Building Gold-layer views and tables for BI and analytics

Gold layer objects expose curated, business-ready data to BI and analytics consumers in Unity Catalog. The main object types are managed tables, views, materialized views, and streaming tables, each differing in refresh behavior, storage, and cost tradeoffs. Choosing the right type depends on freshness requirements, query patterns, and compute cost tolerance.

1 · Learn the must-know

• 1A view is a virtual object with no stored data; it re-executes its defining query on every access, so it is always fresh but adds compute cost per query.
• 2A table (managed or external) stores materialized data physically and must be explicitly refreshed via write/merge/overwrite operations or a pipeline; queries against it are fast since no recomputation happens at read time.
• 3A materialized view precomputes and stores query results, refreshing incrementally where possible, and is created and managed through Lakeflow Declarative Pipelines (formerly Delta Live Tables), reducing query latency for repeated BI dashboard queries.
• 4A streaming table is designed for incremental, append-only or CDC-style ingestion using Structured Streaming semantics within a pipeline, continuously or incrementally processing new data rather than reprocessing the full dataset.
• 5Materialized views and streaming tables are pipeline-managed objects and cannot be directly modified with arbitrary DML like a normal managed table; they are updated by re-running or triggering the owning pipeline.
• 6Gold tables typically apply business logic, aggregations, and joins on top of Silver layer data, and access is governed through Unity Catalog permissions (GRANT SELECT) so BI tools connect only to these curated objects.

2 · Check your understanding