LEARN. PRACTICE. GROW.Concept and mechanism
An analytical flow separates source, ingestion, transformation, storage, and consumption. ETL transforms before loading into the destination under consideration; ELT loads and uses the destination to transform. A data lake can retain files of different formats; a warehouse organizes data for analytical queries. A lakehouse combines lake patterns with table-management capabilities. Microsoft Fabric integrates SaaS experiences such as Data Factory, engineering, warehousing, Real-Time Intelligence, and Power BI over OneLake. Azure Databricks offers a platform for engineering, analytics, and other workloads, with Spark and notebooks among its capabilities. No product name removes the need to define contracts and quality.
Guided application
In a fictional scenario, daily reporting can accept batch processing while failure alerts need events processed within minutes. Streaming handles data as it arrives and can aggregate by windows. Define the time basis, acceptable lag, and late-event handling; refreshing the screen does not create that processing. Azure Stream Analytics and Fabric Real-Time Intelligence are candidates to assess around design and integrations. For each stage, record counts, timestamps, errors, and resumption ownership. A green pipeline can contain rejections or still not feed the consumed model. Acceptance should demonstrate when data becomes usable, reconciled, and sufficiently current for the decision.
Counting failures every five minutes requires defining windows; summing everything since startup is insufficient.
Common pitfalls
ELT as no transformation; lake as automatic governance; visual refresh as streaming; green job as business SLA.
Related topics: Data, workloads, and responsibilities · Relational modeling, integrity, and SQL · Relational services and compatibility
Measure the complete path from event to decision.
Reference: ETL versus ELT · DP-900 skills measured 2026-07-21