Power BI vs Old Tools What Changed for Cloud Engineers

Legacy BI Architecture (typical):

Source DB  →  ETL (nightly)  →  BI extract/cube  →  Reports

Characteristics:
- Warehouse query too slow for interactive use
- BI engine caches data in its own store
- Refresh window is fixed (e.g., 2 AM daily)
- Report change requires IT ticket + release

Engineer responsibilities:
- Maintain ETL servers and schedules
- Manage BI server capacity
- Handle failed extracts
- Provision AD groups and folder access

Legacy Engineer Daily Checklist:

☐ Check overnight ETL job status
☐ Investigate failed extract or cube refresh
☐ Monitor BI server CPU and memory
☐ Provision new AD group for report access
☐ Deploy report changes via change request
☐ Reconcile report numbers against source
☐ Respond to "data is stale" tickets

Time split (approximate):
- 40% ETL and refresh failures
- 25% access and provisioning
- 20% report deployment
- 15% performance tuning

Cloud BI Architecture (Power BI + Fabric):

Sources  →  OneLake / Warehouse  →  Semantic model  →  Reports

Key differences:
- Compute runs on Fabric capacity, not a separate BI server
- DirectQuery and live connections reduce extract dependency
- Semantic models are reusable and governed centrally
- Copilot enables natural-language analysis
- Capacity autoscales when load increases

Engineer responsibilities:
- Manage Fabric capacities and workspaces
- Configure gateways for on-prem sources
- Govern semantic models and RLS
- Monitor refresh, performance, and cost

Fabric Context (2026):

Power BI is the user-facing layer of Microsoft Fabric.

Fabric workloads:
- Data Factory (ingestion)
- Data Engineering (Spark, notebooks)
- Data Warehouse (T-SQL)
- Real-Time Intelligence (streaming)
- Data Science (ML)
- Power BI (reporting)

For cloud engineers, this means:
- BI is part of a larger data platform
- OneLake is the shared storage layer
- Governance spans all workloads, not just reports
- Security is identity-driven via Entra ID

Cloud BI Engineer Checklist:

Platform:
☐ Fabric capacities provisioned and sized
☐ Workspaces structured by domain/team
☐ Semantic models promoted and endorsed

Connectivity:
☐ Gateways installed for on-prem sources
☐ OIDC / Entra ID configured for auth
☐ Network ports and TLS verified

Governance:
☐ RLS roles defined and tested
☐ Sensitivity labels applied via Purview
☐ BYOK configured (if required)

Operations:
☐ Refresh schedules monitored
☐ Alerts configured for failures
☐ Cost and capacity usage reviewed weekly

Gateway Security Essentials:

- Use HTTPS for all gateway communication
- Configure OIDC with Azure AD / Entra ID
- Enable multi-factor authentication
- Restrict gateway access to required data sources
- Monitor gateway health and failover
- Rotate credentials on a schedule
- Use private endpoints where possible

RLS patterns:
- Define roles in the semantic model
- Test with "View as role" in Power BI Desktop
- Map Entra ID groups to RLS roles
- Validate in the Service after publish

-- Monitoring query: refresh history (example)
SELECT
    [DatasetName],
    [RefreshType],
    [Status],
    [StartTime],
    [EndTime],
    DATEDIFF(MINUTE, [StartTime], [EndTime]) AS DurationMin
FROM
    $SYSTEM.[RefreshHistory]
WHERE
    [StartTime] >= DATEADD(DAY, -7, GETDATE())
ORDER BY
    [StartTime] DESC;

-- What to watch:
-- 1. Repeated failures on the same dataset
-- 2. Refresh duration creeping up
-- 3. Refresh overlapping peak usage hours
-- 4. Capacity throttling events

Legacy BI vs Cloud BI — Engineer View:

Deployment:
Legacy: On-prem servers, VMs
Cloud:  Fabric capacity, SaaS service

Data movement:
Legacy: Scheduled extracts
Cloud:  DirectQuery, live, incremental refresh

Security:
Legacy: AD groups, folder permissions
Cloud:  Entra ID, Purview, RLS, BYOK

Scaling:
Legacy: Manual server upgrades
Cloud:  Capacity autoscale

Governance:
Legacy: IT-owned semantic layers
Cloud:  Governed, reusable semantic models

AI:
Legacy: None or bolt-on
Cloud:  Copilot, natural-language analysis

Migration Checklist (Legacy to Power BI):

Assessment:
☐ Inventory reports and usage
☐ Identify high-value vs low-value reports
☐ Map data sources and dependencies

Data & semantic redesign:
☐ Move logic into warehouse views or semantic models
☐ Remove redundant extracts
☐ Define reusable measures

Platform setup:
☐ Provision Fabric capacity
☐ Configure gateways and networking
☐ Set up workspaces and security groups

Governance:
☐ Define RLS and sensitivity labels
☐ Establish endorsement and promotion process
☐ Set refresh and alert policies

Decommission:
☐ Retire unused reports
☐ Document migrated reports
☐ Turn off legacy jobs

Skills Map for Cloud Engineers:

Must-have:
☐ Azure / Entra ID fundamentals
☐ Networking (gateway, private endpoints)
☐ Power BI Service administration
☐ Fabric capacity and workspace management
☐ Security (RLS, Purview, BYOK)

Good-to-have:
☐ T-SQL and data modeling
☐ PowerShell / CLI automation
☐ CI/CD for semantic models
☐ Cost optimization on Fabric

Emerging:
☐ Copilot and AI feature governance
☐ Real-time intelligence integration
☐ OneLake data governance