Power BI 和 Databricks：实用最佳实践.pdf

1、Power BI+Databricks:Pratical Best Practice 51 Best Practices in 40 MinutesHobbsJune 2025A Whirlwind AgendaFixing everything,one step at a timeoData oServingoPBI ModelingoPBI ReportsoPBI Service3Of Bite Size Best PracticesEverything that can go wrong4Root Cause AnalysisSecurityServingReportsSettingsF

2、ilesDataModelingQueryingModelingCalculationsConnectionPBI ServiceCI/CDStarting at the bottom withData OptimizationData Layer OptimizationUse Liquid Clustering(or Z-ORDER)Use Predictive Optimization(or VACUUM+OPTIMIZE)Use Automatic Statistics(or compute statistics)Avoid wide data typesAvoid high card

3、inality columnsUse materialized viewsPre-aggregate if possibleFilesData6Best Practice#1-11 Use Medallion Architecture to serve only GoldStar Schema is still kingDeclare PK/FK&use RELYGenerate surrogate keys as neededLogical ModelingNext up,SQL Warehouse ServingServing Layer OptimizationUse a SQL War

4、ehouse(not all-purpose)Use Serverless for a faster SLAConsider the Cache when making multiple warehousesSetup Autostop(consider your SLA)Large Data?Go up a t-shirt sizeMany queries?Go wide with auto-scalingSQL Server Warehouse8Best Practice#12-21Monitor warehouse events for performance Setup alertsS

5、et a minimum for scale-outMonitor query history/profile for charge-backAnd where would we be withoutModeling&QueryingPower BI Querying&ModelingConsider RLS/OLS/CLSParamaterize your PQ codePQ should only be SELECT*Limit rows/columnsTry DirectQuery first,then composite,then full importUse User Defined

6、 Aggs.1-1 or 1-*not*-*Use one way relationshipsClick Assume Ref.IntegrityWith DQ,use Native QuerySecurity&Querying10Best Practice#22-34Write efficient DAXNo calculated columns or tables(shift left)Use DAX Studio or Query History for debuggingModelingCalculatio