A deep dive into Direct Lake
summary
Direct Lake runs inside Microsoft Fabric and connects Power BI to a Lakehouse, but it does not truly query files live — it uses column-level caching that pulls only needed columns into memory in seconds instead of minutes. Demos include a 3 billion row set compressed down to ~60 GB and a 900 million row model that still needs seconds to warm up. Unsupported types or row‑level security fall back to DirectQuery, and cache eviction uses a per‑column “temperature” metric; auto refresh clears the cache.
AI-written summary · about
chapters (7)
- 0:00Direct Lake and Microsoft Fabric
- 8:48Direct Lake storage mode
- 14:14Direct Lake modeling best practices
- 25:43Direct Lake demo performance
- 34:28Direct Lake limitations
- 41:21Performance benchmarks and metadata insights
- 49:32Q&A
related talks
30mEmpower every BI professional to do more with Fabric
51mPower BI Composite and Hybrid Models
20m3-phased approach to delivering a lakehouse with data mesh
44mDeep-Dive into Delta Lake
18mDelta Lake – The Foundation Of Your Lakehouse
45mAzure DW and Power BI, reporting at the speed of light!
57mBillions of records: the real success story using Microsoft Power BI Embedded
01h 02mData lakes beyond clusters