The AZ-305 database domain demands scenario-based judgment, not rote memorization. Azure SQL Database, SQL Managed Instance, Cosmos DB, PostgreSQL Flexible Server, and Synapse Analytics each have distinct conditions under which they are the right choice — and the exam tests exactly those boundaries. Here we break down the key selection criteria through patterns found in 36 real exam questions.
---
Choosing Between Azure SQL Database and Managed Instance
Azure relational database services come in three flavors: Azure SQL Database, SQL Managed Instance, and SQL Server on Azure VM. All three are built on the SQL Server engine, but they differ significantly in feature coverage and operational overhead.
Azure SQL Database is a fully managed PaaS offering. In the Serverless tier, vCore scales up and down automatically and the database pauses after an idle period, eliminating compute costs during inactivity. This makes it ideal for intermittent workloads.
SQL Managed Instance provides near 100% compatibility with SQL Server. Instance-level features that Azure SQL Database does not support — SQL Server Agent, CLR stored procedures, cross-database queries, and distributed transactions (MSDTC) — are available without code changes, making it the top choice for lift-and-shift migrations.
Tier differences within Azure SQL Database also matter. Only the General Purpose tier supports Serverless, while Business Critical delivers sub-1ms I/O using local SSDs. Hyperscale allows compute and storage to scale independently up to 100 TB, making it suitable for large-scale OLTP workloads in the tens of terabytes and beyond.
Elastic Pool lets multiple databases share a vCore pool, cutting costs by up to 50% in SaaS multi-tenant scenarios while enforcing isolation through per-database min/max limits. The vCore purchasing model is the only model that supports Azure Hybrid Benefit (up to 55% savings) and Long-Term Retention (LTR, up to 10 years of backup retention).
!Azure SQL Database versus Managed Instance
Cosmos DB and Globally Distributed NoSQL
Azure Cosmos DB is a fully managed NoSQL database that guarantees single-digit millisecond latency in any region worldwide. Two keywords accompany every Cosmos DB scenario on the exam: multi-region simultaneous writes and millisecond response SLA.
With multi-region writes enabled, every region has an independent write endpoint, so writes continue even if any single region fails (Active-active). By contrast, Azure SQL Database Active Geo-Replication keeps secondary replicas as read-only, meaning it cannot support multi-region writes.
Provisioned Throughput mode pre-allocates throughput in RU/s and contractually guarantees a P99 read latency of 10ms and write latency of 15ms. This is the only option when you need to prove latency SLAs in documentation during regulatory audits.
Choose the API based on your data model. JSON documents with SQL queries call for the NoSQL (Core SQL) API; MongoDB driver compatibility means the MongoDB API; distributed SQL over PostgreSQL uses the PostgreSQL API (Citus-based); and node/edge relationship traversal uses the Gremlin API. For multi-hop traversals like “friends of friends,” Gremlin API is the correct answer.
Synapse Link for Cosmos DB automatically syncs Cosmos DB data to a column-oriented analytical store without ETL, enabling analytics in Synapse Analytics with no impact on operational performance.
---
PostgreSQL and MySQL Flexible Server
Azure Database for PostgreSQL Flexible Server and Azure Database for MySQL Flexible Server are fully managed versions of their respective open-source relational databases. AZ-305 exam questions in this area primarily focus on high availability configurations and disaster recovery options.
There are three compute tiers: Burstable, General Purpose, and Business Critical. Zone-redundant high availability is only supported at General Purpose tier and above. The Burstable tier has the lowest cost, but because it does not support Zone-redundant HA, you must choose General Purpose or higher whenever the requirement states that service must survive a single data center failure.
It is also important to distinguish between regional disaster recovery options. Zone-redundant HA provides redundancy across availability zones within the same region, protecting against data center-level failures. Geo-redundant Backup automatically replicates backups to another Azure region, enabling recovery when an entire region goes down. If an RTO of a few hours is acceptable and manual recovery is feasible, geo-redundant backup alone can satisfy regional disaster recovery requirements cost-effectively without Zone-redundant HA.
Also keep in mind that read replicas deployed to the same region cannot serve as regional disaster recovery. If regional DR is the goal, replicas must be deployed to a different region, or geo-redundant backup must be used.
---
Analytical Workloads: Synapse and Data Warehousing
Azure Synapse Analytics delivers a data warehouse (Dedicated SQL Pool), Apache Spark-based big data processing, and Azure Data Lake Storage integration in a single platform. If you need to transform tens of terabytes of research data with Spark and analyze both structured and unstructured data together, Synapse Studio gives you a single management surface.
Dedicated SQL Pool uses an MPP architecture to support petabyte-scale data warehouses. The decision between Synapse and Databricks is straightforward: choose Synapse Analytics when you need unified management of SQL DW + Spark + Data Lake; choose Databricks when you are focused on Spark-centric advanced machine learning.
---
Service Comparison Table
| Service | Primary Use Case | Scaling | Latency | Cost Model | |---------|-----------------|---------|---------|------------| | Azure SQL Database (Serverless) | Intermittent or unpredictable workloads | vCore auto-scales 0.5–80 | General-purpose | Per-second billing; no compute charge during idle | | Azure SQL Database (Hyperscale) | Large-scale OLTP 10s of TB+ | Compute and storage scale independently up to 100 TB | General-purpose | Storage page-server architecture | | SQL Managed Instance | Full SQL Server compatibility for lift-and-shift | Vertical scaling at instance level | Sub-1ms on Business Critical tier | Per-instance-hour billing | | Cosmos DB (Provisioned) | Multi-region writes, millisecond SLA, NoSQL | Horizontal scale in RU/s | P99 under 10ms | Pre-allocated RU/s | | PostgreSQL Flexible Server | Open-source relational with Zone-redundant HA | Vertical scaling | Standard RDBMS | Per-tier-hour billing | | Synapse Analytics | SQL DW + Spark + Data Lake unified | Horizontal scale in DWU | Analytical batch processing | Per-DWU-hour billing |
---
Selection Criteria That Frequently Trip Up Exam Takers
The key to exam questions is mapping keywords precisely to services. Here are the most commonly tested patterns organized by scenario.
Relational DB Migration Scenario
Scenario: You are migrating an on-premises SQL Server and must preserve SQL Agent, CLR stored procedures, and cross-database queries exactly as they are.
The answer is SQL Managed Instance. Azure SQL Database does not support these instance-level features. SQL Server on Azure VM offers perfect feature compatibility, but as an IaaS solution it requires you to manage OS patching and backups yourself, adding significant operational overhead. When you need PaaS management combined with full SQL Server compatibility, SQL Managed Instance is the answer.
Cost Reduction Scenario
Scenario: You want to reuse existing SQL Server licenses (with Software Assurance) in Azure, or you need to retain backups for seven or more years.
Both conditions require the vCore model. Azure Hybrid Benefit and Long-Term Retention (LTR) are only available under the vCore purchasing model. Neither feature is available with the DTU model.
Cosmos DB vs SQL Database Decision
Scenario: Data must be written simultaneously from multiple regions worldwide, with guaranteed millisecond response times.
The answer is Cosmos DB with multi-region writes enabled. Azure SQL Database Active Geo-Replication keeps secondary replicas read-only and cannot support multi-region writes.
Scenario: You need contractually guaranteed SLAs for write latency and throughput, with documentation that can be presented during regulatory audits.
Again, the answer is Cosmos DB Provisioned Throughput. Azure officially documents four-dimensional SLAs — availability, read latency, write latency, and throughput — in its service agreements.
Elastic Pool vs Serverless Decision
Scenario: You manage hundreds of customer databases with highly variable usage patterns and want to reduce management overhead while optimizing costs.
The answer is Elastic Pool. Multiple databases share a pooled resource, and management is simplified to the pool level. Serverless is the right choice when you want automatic pausing for a single database with irregular workloads.
---
Practical Architecture Tips
Here are combination patterns commonly applied in real-world architecture design.
Cosmos DB + Synapse analytics: Using Synapse Link for Cosmos DB, you can sync operational data to the analytical store in near real time without ETL, without consuming extra RUs from your operational database.
SQL Managed Instance hybrid migration: Use VNet integration to establish a private connection from your on-premises SQL Server, add an Azure VM as an Always On Availability Group secondary replica to validate sync, then switch over with a planned failover to minimize downtime.
SQL Database audit logs: When storing audit logs in a Storage Account, always use a Storage Account in the same region as the SQL server. Reusing a Storage Account in another region in an environment with data sovereignty regulations is a compliance violation.
When to use Hyperscale: Reserve Hyperscale for databases tha