OneLake & Delta Lake

The one copy of data for the whole tenant — and how to keep its Delta tables fast and cheap.

OneLake is a single, tenant-wide data lake built on ADLS Gen2. Every workspace gets a folder, every lakehouse and warehouse writes Delta-Parquet into it, and every engine (Spark, SQL, Power BI Direct Lake, KQL) reads the same files.

That is the promise. The cost is that badly maintained Delta tables degrade every engine at once. This section is about keeping them healthy.

Pages

Mental model

OneLake (tenant)
└── Workspace
    └── Lakehouse
        ├── Tables/           managed Delta tables (SQL endpoint + Direct Lake)
        │   └── my_table/
        │       ├── _delta_log/       JSON + checkpoint commits
        │       └── part-*.parquet    data files (+ deletion vectors)
        └── Files/            unmanaged — raw/landing zone

The _delta_log is the source of truth. File listing performance, checkpoint frequency, and the number of small Parquet files under each table are the three levers that decide how fast a cold query starts.

Stay ahead of Fabric changes

Fabric runtime changes, API updates, and deprecations. No spam, unsubscribe anytime.

On this page