What's changed: Added §4.4 "Microsoft Fabric (Unified Analytics Platform)" to fix the drift from the 2026 refresh that centers the analytics domain on Fabric/OneLake/lakehouse/Real-Time Intelligence/Data Activator. Adds Fabric alongside the existing Synapse-centric content
4.1The Analytics Pipeline, Data Warehouse, and Data Lake
Understand the typical analytics pipeline (ingest → process → store → visualize), the difference between a data warehouse and a data lake, and ETL/ELT.
Analytics flows as collect, prepare, store, and present. Understanding this pipeline shows where each Azure service fits.
4.1.1The analytics pipeline
- Ingest: collect data from various sources.
- Process: clean and transform the data.
- Store: keep it in a warehouse or lake.
- Visualize: present it in reports and dashboards.
The pipeline starts from varied data sources (operational DBs, logs, IoT, external APIs) and ends in reports and dashboards people view. In "process," cleansing fixes missing or inconsistent values and transformation reshapes data for analysis. Keep the big picture: data flows from OLTP (operational systems, Chapter 1) through this pipeline into OLAP (the analytics platform).
4.1.2Data warehouse vs. data lake
There are two stores for "keeping" data, with different aims. A data warehouse stores pre-shaped, structured data optimized for analytics/aggregation; with a defined schema, aggregation queries are fast, fitting standard reports. A data lake holds raw, varied data (structured, semi-structured, unstructured) cheaply and at scale, regardless of format—"store first, shape when used," a staging ground for ML and exploratory analysis. Combining both—raw data in the lake, shaped data analyzed in the warehouse—is the lakehouse idea.
| Aspect | Data warehouse | Data lake |
|---|---|---|
| Data stored | Shaped, structured | Raw, varied (structured–unstructured) |
| Schema | Defined on write | Interpreted on read |
| Best for | Standard reports/aggregation | Exploratory analysis, ML |
Moving data from sources to a store is ETL (Extract → Transform → Load) or ELT (Extract → Load → Transform). The difference is when transformation happens: ETL transforms before loading; ELT loads first and transforms as needed. For large cloud-scale data, ELT—load into the lake first, then process—is common.
Four levels of analytics: descriptive (what happened / aggregation, reports) → diagnostic (why it happened / root-cause) → predictive (what will happen / ML) → prescriptive (what to do / recommended actions). More advanced to the right; for DP-900, matching a question to its level deepens understanding.
Q. Warehouse or lake? Warehouse for standard aggregation reports; lake when formats vary and you want to "store first" or feed ML. The lakehouse does both. Q. ETL or ELT? ETL when shaping is needed at ingest; ELT to ingest large data fast and process flexibly later. Q. What is cleansing? Fixing missing/duplicate/inconsistent values to raise data quality.
Common points: organized structured data for analytics = data warehouse, large raw varied data = data lake, and the ETL (transform first) vs ELT (transform later) difference. Also know the four analytics levels (descriptive/diagnostic/predictive/prescriptive).
4.1.3Section summary
- Pipeline: ingest → process → store → visualize
- Warehouse (organized structured) / lake (raw varied), ETL/ELT
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. Which stores organized structured data, optimized for analytics and aggregation?
Q2. Which can store large amounts of raw, varied data from structured to unstructured?
Q3. Which is the correct general order of a data analytics pipeline?
Keep track of your progress
The full study guide is free to read. Sign up free to practice with the question bank, track what you have read, review your mistakes, and highlight passages.