Architecture decisions compound. A star schema with a stable grain saves years of ad-hoc joins; a “wide table for everything” forces two rewrites. These cheat sheets cover modeling, change-data patterns, and platform features that are really architecture choices wearing product names: streams, dynamic tables, Iceberg, replication.
Read them before you pick Type 2 vs Type 1, before you put CDC on every source “just in case,” and before you copy a lakehouse blog post into a warehouse that does not need it. Cost problems after year one are usually grain, partitioning, and refresh semantics — not the logo on the invoice.
Default to a star schema for analytics. Use Type 2 only on dimensions that the business queries historically (customer, account, product). Type 1 is correct for labels that nobody needs as-of. Do not invent a vault until you have many sources and a team that will maintain hubs and satellites.
Size compute with the warehouse sizing tool after the model is stable, not before. A common failure mode is picking Iceberg or Dynamic Tables because a blog post used them, then spending a quarter on catalog permissions and refresh lag that a nightly MERGE would have avoided. Start from grain and freshness requirements; pick the feature second.
Related: cloud platforms, interview sheets, library.
← Back to Home