Lesson 4.13.1 built the star schema in a throwaway database. In production, you need a place where the data of several sources comes together, keeps its history and stands ready for reports, without anyone running servers. Microsoft's offering for that is Microsoft Fabric. This lesson shows what is in it, how a lakehouse and a warehouse differ, how the Kladde's database gets in, and what you have to consider about cost, permissions and privacy along the way.
Fabric in a few sentences
Fabric is a platform for data analytics as SaaS: you rent compute in the form of a capacity, and all the tools run on it, from data integration through Spark and a SQL warehouse to Power BI. Everything is stored in OneLake, a single data lake per organization. The common format is Delta: Parquet files with a transaction log next to them. Because all tools read the same format, no table has to be copied for another tool.
Four building blocks count for this module:
| Building block | What it is | What you work with |
|---|---|---|
| Lakehouse | Files and Delta tables in OneLake, plus a SQL analytics endpoint | write with Spark, pipelines, dataflows; read with T-SQL as well |
| Warehouse | a relational warehouse on OneLake, also stored in Delta | T-SQL with transactions across several tables, views, procedures |
| Mirrored database | a continuously updated copy of a source database in OneLake | you set it up; Fabric keeps it current |
| Semantic model and Power BI | relationships, measures and reports on top of the tables | Power BI; in Direct Lake mode it reads the Delta tables directly |