AZ-305 domain - 22% of the exam

Design Data Storage Solutions

Design Data Storage Solutions is 22% of the AZ-305 exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

56
Practice questions
3
Objectives
36
Medium
20
Hard

Where people go wrong

  • Choosing a Cosmos DB API by familiarity

    The wrong answers pick a Cosmos DB API by habit, often API for NoSQL as the native default, when the scenario names an existing protocol or data model. Cassandra tooling with CQL drivers points to API for Apache Cassandra, Azure Table Storage client code points to API for Table, an established document codebase points to API for MongoDB, and relationship traversal points to API for Gremlin. API for NoSQL is the answer only for a new build with no driver investment.

    See it in a sample question
  • Reaching for Strong consistency by default

    The wrong answers choose Strong consistency whenever a requirement sounds demanding, when a weaker level meets it. Strong reads cost 2x the request units, and Strong adds write latency the workload does not need. A single-region app where each user must see their own writes fits Session, whose reads cost the same as Eventual. A service with one write region whose reads may lag by at most 100 versions or 5 seconds fits Bounded staleness, the near-strong choice.

    See it in a sample question
  • Pairing Archive with zone redundancy

    The wrong answers combine the Archive access tier with zone-redundant storage because the scenario asks for resilience. Archive supports only LRS, GRS, and RA-GRS, so ZRS, GZRS, and RA-GZRS are all invalid with it. When a workload wants the lowest storage cost and a rehydrate of up to 15 hours is acceptable, the answer is Archive on LRS, or on GRS or RA-GRS where geo-replication is required.

    See it in a sample question
  • Treating geo-replication and failover groups as one feature

    The wrong answers swap active geo-replication and failover groups. Active geo-replication has no stable listener, so the connection string changes after failover because the new primary lives on a different logical server. A failover group adds the read-write listener, and its customer-managed policy keeps the failover decision with people. Chaining a geo-replica is an active geo-replication technique that failover groups with multiple secondaries do not allow.

    See it in a sample question
  • Confusing Data Factory, Synapse, and Databricks

    The wrong answers swap Azure Data Factory, Azure Synapse Analytics, and Azure Databricks. Data Factory is code-free orchestration with 90-plus connectors and no analytics workspace. Synapse combines dedicated and serverless SQL, Spark pools, and the T-SQL PREDICT function in one workspace. Databricks is where MLflow, Delta Lake, and Unity Catalog live, while Synapse governance is Synapse RBAC. Synapse embeds the same Data Factory engine, but only Synapse adds SQL and Spark pools.

    See it in a sample question

Objectives in this domain

Sample questions from this domain

Free sampleDesign Data Storage Solutionsmedium

An IoT platform team already runs Apache Cassandra on-premises, uses CQL throughout their tooling, and stores wide-column time-series telemetry from millions of devices. They want a managed Azure database without rewriting their drivers. Which API should they adopt?

  • AAdopt Azure Cosmos DB API for Table for key-value telemetry workloads.
  • BAdopt Azure Cosmos DB API for NoSQL with custom CQL adapter layers.
  • CAdopt Azure Cosmos DB API for Apache Cassandra to keep CQL drivers. Correct
  • DAdopt Azure Cosmos DB API for Gremlin for device graph traversal queries.
Existing Cassandra codebases with CQL tooling and wide-column models map to Cosmos DB API for Apache Cassandra. API for Cassandra provides wire-protocol compatibility for the Cassandra Query Language, automatic global distribution, and is the documented fit for teams already invested in Cassandra tooling and wide-column or time-series data models.

Why A is wrong: API for Table is aimed at lifting and shifting Azure Table Storage applications, not at wide-column CQL workloads with Cassandra tooling.

Why B is wrong: API for NoSQL is the native API and does not speak the Cassandra wire protocol; custom adapters are not the documented path.

Why C is correct: Correct. API for Cassandra provides wire-protocol compatibility for the Cassandra Query Language, automatic global distribution, and is the documented fit for teams already invested in Cassandra tooling and wide-column or time-series data models.

Why D is wrong: API for Gremlin targets graph-traversal workloads such as social networks or fraud detection, not wide-column time-series telemetry.

Free sampleDesign Data Storage Solutionshard

A global reporting service has one write region and four read regions, and needs reads bounded to at most 100 versions or 5 seconds behind writes across those regions. Multi-region writes are explicitly not required. Which Cosmos DB consistency level fits?

  • AConfigure Strong consistency for guaranteed freshness.
  • BConfigure Session consistency for low RU cost.
  • CConfigure Bounded staleness with K and T thresholds. Correct
  • DConfigure Eventual consistency and rely on application retries.
Single-write-region globally distributed apps that need near-strong reads point to Bounded staleness with K and T. Bounded staleness guarantees that lag between any two regions is less than K versions or T time, whichever is reached first. It is documented as the best fit for globally distributed applications using a single write region with multiple regions where near-strong consistency is desired.

Why A is wrong: Strong consistency removes any lag but costs significantly higher write latency that is more than the workload needs; bounded staleness is the documented near-strong choice.

Why B is wrong: Session bounds staleness only within a session token, not across regions or sessions; it cannot meet a tenant-wide K-or-T bound.

Why C is correct: Correct. Bounded staleness guarantees that lag between any two regions is less than K versions or T time, whichever is reached first.

Why D is wrong: Eventual provides no bound on staleness; clients can read versions that are arbitrarily old relative to the latest write.

Free sampleDesign Data Storage Solutionshard

A regulator-mandated retention policy requires keeping nightly database exports for 7 years. Retrieval requests are rare and a 15-hour rehydrate is acceptable. The team wants the lowest possible storage cost and is satisfied with LRS for the archive. Which tier and redundancy combination fits?

  • AUse the Cold access tier on GZRS storage.
  • BUse the Archive access tier on ZRS storage.
  • CUse the Cool access tier on RA-GZRS storage.
  • DUse the Archive access tier on LRS storage. Correct
Archive is cheapest for long-term retention but supports only LRS, GRS, and RA-GRS, with up to 15-hour rehydrate. Archive is the lowest-cost offline tier with a 180-day minimum retention, satisfying 7-year regulatory retention. Archive supports only LRS, GRS, and RA-GRS redundancies; LRS is the cheapest of those.

Why A is wrong: Cold is more expensive than Archive and the workload is fine with hours of rehydrate latency; Archive on LRS is cheaper for 7-year retention.

Why B is wrong: Archive does not support ZRS, GZRS, or RA-GZRS; only LRS, GRS, and RA-GRS are valid redundancies for Archive.

Why C is wrong: Cool plus RA-GZRS is significantly more expensive than Archive on LRS for a workload where 15-hour rehydrate is acceptable.

Why D is correct: Correct. Archive is the lowest-cost offline tier with a 180-day minimum retention, satisfying 7-year regulatory retention.

Free sampleDesign Data Storage Solutionsmedium

A line-of-business app connects to four databases on a single logical server. After a regional failover, the app must continue to use the same connection string for read-write traffic without code changes. Which capability satisfies the requirement?

  • AAdd the four databases to a failover group and connect the app to the read-write listener. Correct
  • BConfigure active geo-replication on each of the four databases separately.
  • CFront the databases with a Traffic Manager profile in priority routing mode.
  • DUse an Application Gateway with health probes pointing at the primary databases.
Failover-group listener endpoints preserve the connection string across geo-failovers; active geo-replication does not. A failover group exposes a read-write listener DNS CNAME of the form `<fog>.database.windows.net` that always points to the current primary. After a geo-failover, the DNS record is automatically updated so the app keeps the same connection string.

Why A is correct: Correct. A failover group exposes a read-write listener DNS CNAME of the form `<fog>.database.windows.net` that always points to the current primary.

Why B is wrong: Active geo-replication does not expose a stable listener endpoint; the connection string changes after failover because the new primary lives on a different logical server.

Why C is wrong: Traffic Manager resolves names but does not coordinate the database failover or guarantee a transactionally consistent geo-secondary at the destination.

Why D is wrong: Application Gateway is an HTTP load balancer; it does not front the SQL TDS endpoint or coordinate geo-failover of databases.

Free sampleDesign Data Storage Solutionshard

A data-science team needs collaborative notebooks, MLflow experiment tracking, Delta Lake as the storage format, and Unity Catalog for fine-grained governance, all in a single managed Azure service. Which service fits?

  • ASelect Azure Synapse Analytics dedicated SQL pools.
  • BSelect Azure Data Factory mapping data flows.
  • CSelect Microsoft Fabric data warehouse.
  • DSelect Azure Databricks as the platform. Correct
Delta Lake plus MLflow plus Unity Catalog plus notebooks is the Azure Databricks combination. Azure Databricks is the documented home for Delta Lake, MLflow, Unity Catalog, and collaborative notebooks. The decision guide names Databricks as the choice for complex ML, collaborative data science, and the data lakehouse, with deeper governance via Unity Catalog and native MLflow.

Why A is wrong: Synapse offers SQL pools and Spark pools and integrates with Azure ML, but its native governance is Synapse RBAC, not Unity Catalog, and MLflow is not native.

Why B is wrong: ADF is an orchestration service without notebooks, MLflow, Delta Lake, or Unity Catalog as native features.

Why C is wrong: Fabric data warehouse is a T-SQL surface and does not provide the native MLflow plus Unity Catalog plus Delta Lake combination described.

Why D is correct: Correct. Azure Databricks is the documented home for Delta Lake, MLflow, Unity Catalog, and collaborative notebooks.

Other domains in this exam

See also the AZ-305 cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by Microsoft. Original, blueprint-aligned practice material only.