D2 · Data Storage

Data Integration and Analytics

AZ-305 tests whether you can assemble the right pipeline: batch or streaming? visual ETL or Spark code? data warehouse or data lake? It doesn't ask for implementation details — it asks you to pick the right service for each piece.

Contents

The data services landscape
Batch: Data Factory, Synapse, Databricks
Streaming: Event Hubs and Stream Analytics
Analytical stores and Fabric
Decision table

The data services landscape

Role in the pipeline	Service	In one line
Batch ingestion / orchestration	Azure Data Factory	No-code visual ETL/ELT, 100+ connectors, scheduled triggers
Streaming ingestion	Event Hubs	Millions of events per second, Kafka-compatible
Streaming processing	Stream Analytics	SQL queries over real-time streams, time windows
Big data processing (code)	Azure Databricks	Managed Spark: notebooks, ML, Delta Lake
Unified analytics	Synapse Analytics	SQL pools (DW) + Spark + pipelines in one workspace
Telemetry/log analytics (interactive)	Azure Data Explorer	KQL over billions of records with second-level latency; ingests Event Hubs streams
Unified SaaS platform	Microsoft Fabric	Lakehouse (OneLake) + Power BI + everything above as SaaS
Lake storage	ADLS Gen2	The substrate underneath all of the above

Batch: Data Factory vs Synapse vs Databricks

Azure Data Factory

•No-code orchestration and data movement (mapping data flows for transforms)
•Self-hosted integration runtime for on-premises sources
•Signal: "copy data from 30 sources every night", "visual ETL"
•Not an analytics engine — it orchestrates and moves data

Azure Databricks

•Spark first-class: complex transformations IN CODE (Python/Scala/SQL)
•Machine learning, Delta Lake, structured streaming
•Signal: "data engineers use notebooks/Spark", "ML feature engineering"
•More control and power, more expertise required

Synapse Analytics

•Unified workspace: dedicated SQL pools (MPP warehouse), serverless SQL (query the lake), Spark pools and pipelines (embedded ADF)
•Signal: "petabyte-scale data warehouse" → dedicated SQL pool
•"Query parquet in the lake with T-SQL with nothing to provision" → serverless SQL pool

Streaming: Event Hubs + Stream Analytics

The canonical streaming pipeline (comes up on the exam)

1Devices/apps emit events → Event Hubs (or IoT Hub if device management, certificate-based auth or bidirectional cloud-to-device commands are needed)

2Stream Analytics consumes the stream with SQL queries and windows (tumbling, hopping, sliding, session)

3Outputs: Power BI (live dashboard), SQL/Cosmos (serving), ADLS (archive), Functions (actions)

Event Hubs — design

• Partitions are fixed at creation — size them for consumer parallelism
• Consumer groups: each consumer reads the stream independently
• Capture: automatic no-code dump to Blob/ADLS
• Tiers: Standard / Premium / Dedicated based on throughput and isolation
• Kafka endpoint: migrate Kafka producers/consumers with no code changes

Processing alternatives

• Stream Analytics: declarative SQL, no cluster — the "simple and managed" option
• Databricks Structured Streaming: complex logic in code
• Azure Data Explorer: ingestion + interactive query over massive telemetry/logs (KQL)
• Functions: react event by event (not windowed aggregations)

Analytical stores and Fabric

Need	Option
Relational MPP data warehouse at scale	Synapse dedicated SQL pool (or Fabric Warehouse)
Lakehouse: Delta files + SQL + integrated BI	Microsoft Fabric (OneLake) / Databricks + Delta Lake
Ad-hoc queries over lake files with no cluster	Synapse serverless SQL pool
Telemetry/log analytics with interactive latency	Azure Data Explorer (KQL)
HTAP: analytics over Cosmos without touching RUs	Synapse Link / mirroring in Fabric
BI and visualization	Power BI (DirectLake with Fabric)

Decision table

The scenario says...	Design with...
Copy nightly data from 20 sources (including on-prem) into the lake, no code	Data Factory + self-hosted IR
Power BI dashboard updated live from sensors	Event Hubs → Stream Analytics → Power BI
Detect anomalies in 5-minute windows over the stream	Stream Analytics (tumbling window)
Complex Spark transformations and ML models	Azure Databricks
Petabyte-scale data warehouse for corporate reporting	Synapse dedicated SQL pool / Fabric Warehouse
Analysts query lake parquet with T-SQL, pay per query	Synapse serverless SQL pool
Automatically archive the raw stream into the lake	Event Hubs Capture → ADLS Gen2
Existing Kafka apps with no rewrite	Event Hubs with Kafka endpoint
Interactive exploration of billions of logs	Azure Data Explorer

Did you get this topic?

Test what you just learned

A factory emits 100,000 sensor events per second. They need: (1) compute per-machine averages in 1-minute windows and show them on a live dashboard, (2) store every raw event in the data lake for historical analysis, (3) minimal infrastructure administration. What pipeline do you design?

Ready to test "Data Integration and Analytics"?

Put what you learned to the test with original AZ-305 exam questions.

Start AZ-305 mock exam

Unlimited mock exams with the Pro plan

PreviousStorage Accounts DesignData Storage NextData Protection and EncryptionData Storage