The AZ-305 blueprint is split across 4 domains. Weights are the official share of the exam; see the official exam guide for the authoritative breakdown.
What you must be able to do. Given an organisation's identity, governance, and observability requirements, design the authentication and authorisation model, the management-group and policy hierarchy, and the monitoring and diagnostic topology that enforce least privilege and meet the stated residency, retention, and visibility constraints.
In one sentenceThe control-plane domain: designing who can do what, what is allowed at all, and how you see everything, using Entra and RBAC, management groups and Azure Policy, and Azure Monitor with Log Analytics.
Recall check: answer these from memory first
- When must you use a user-assigned managed identity instead of a system-assigned one, and why does sharing across resources force that choice?
- Which requirements justify creating an additional Log Analytics workspace rather than consolidating into one, and which tempting reason does not?
- What does Azure Policy plus management groups give you that an RBAC role assignment does not, and how far down does each reach?
- How many destinations of one type can a single diagnostic setting target, and how do you fan logs to two Log Analytics workspaces?
What it tests. Designing the governance and observability backbone before workloads land. Authentication and authorisation with Microsoft Entra ID and Azure RBAC (least privilege, system-assigned versus user-assigned managed identities, scoping role assignments); governance with management groups, Azure Policy, and landing zones for guardrails that apply across subscriptions; and logging and monitoring with Azure Monitor, Log Analytics workspace design (when to add a workspace for residency, retention, or billing separation), diagnostic settings, and Application Insights for application and AI-agent observability.
How to study it. Learn each control by the requirement it answers, because the exam tests the choice, not the click path. For identity, fix the managed-identity rule: system-assigned is one-per-resource and shares that resource's life cycle, so a shared identity across resources forces user-assigned. For governance, separate the two layers: RBAC grants a principal permission to act, Azure Policy plus management groups set guardrails on what is allowed at all and inherit down the hierarchy. For monitoring, drill Log Analytics workspace design as a decision (add a workspace for per-tenant residency, divergent retention in the same table, or split billing, not for cost arbitrage) and learn the diagnostic-settings constraint that one setting allows one of each destination type. Practise reading a requirement and naming the single control that fits.
Easy to confuse
- System-assigned versus user-assigned managed identity. A system-assigned identity is tied to exactly one resource and shares its life cycle, so it cannot be reused across resources; a user-assigned identity is a standalone resource you attach to many. If the scenario wants several resources to share one identity or one set of RBAC assignments, it is user-assigned.
- Azure RBAC versus Azure Policy. RBAC controls who can perform actions on a resource (grant a principal a role at a scope); Azure Policy controls what configurations are allowed at all and can deny or remediate regardless of who acts. RBAC is about permission, Policy is about compliance guardrails, and both inherit down management groups.
Worked example from the AZ-305 bank
lock_openFree sampleDesign Identity, Governance, and Monitoring Solutionsmedium
A platform team plans to attach one system-assigned managed identity to two virtual machines to consolidate RBAC assignments. A system-assigned managed identity can be associated with more than one Azure resource so multiple VMs can share a single identity. Is this statement correct?
- AYes
- BNocheck_circle Correct
System-assigned managed identities are one per resource; sharing requires user-assigned. System-assigned managed identities can only be associated with a single Azure resource and share its life cycle. To share an identity across multiple Azure resources, the documented pattern is a user-assigned managed identity.
Why A is wrong: Treating system-assigned as shareable is a common shortcut that fails at deployment time; the assignment is bound to a single parent resource and cannot be attached to a second.
Why B is correct: Correct. System-assigned managed identities can only be associated with a single Azure resource and share its life cycle.
What you must be able to do. Given an access pattern with stated consistency, compatibility, latency, and continuity requirements, select the relational, NoSQL, or unstructured store and its configuration, and choose the right integration tool, rather than the service you know best.
In one sentenceThe data-platform domain: reading the access and compatibility requirement and choosing the right store and consistency model, from Azure SQL and Managed Instance to Cosmos DB and Storage, plus the integration tooling.
Recall check: answer these from memory first
- Which three traits distinguish Azure SQL Managed Instance from Azure SQL Database, and which capability is exclusive to SQL Database?
- Why can a Cosmos DB account with multiple write regions not use Strong consistency, and what does that tell you about the trade-off?
- Which Cosmos DB API lets a team lift and shift an existing Azure Table Storage workload without rewriting client code?
- Customer-managed or Microsoft-managed failover policy when business owners must control when failover happens, and why?
What it tests. Matching storage to requirement and configuring it correctly. Relational data with Azure SQL Database and SQL Managed Instance (compatibility, private endpoints, Hyperscale, failover-group policy); semi-structured and unstructured data with Azure Cosmos DB (API selection for migration, consistency levels and their interaction with multi-region writes) and Azure Storage; and data integration with Azure Data Factory, Synapse Analytics, and Databricks for moving and transforming data at scale.
How to study it. Build a storage-selection decision tree and drill it, because this domain is largely selection plus a few configuration traps. Start from the requirement: near-100 percent SQL Server compatibility for a lift-and-shift points to SQL Managed Instance (private endpoint by default, Agent, Service Broker, cross-database queries), while Hyperscale and the cheapest single-database PaaS point to Azure SQL Database. For Cosmos DB, learn API-by-migration (Table Storage maps to API for Table, MongoDB to API for MongoDB) and the hard rule that Strong consistency is incompatible with multiple write regions. For continuity inside the data tier, know that customer-managed (manual) failover is the recommended default so the business keeps control of timing. Practise naming the one store and one setting that satisfy every named constraint.
Easy to confuse
- Azure SQL Database versus SQL Managed Instance. SQL Database is the per-database PaaS with the broadest scale features including Hyperscale up to very large databases; Managed Instance gives near-100 percent SQL Server compatibility, a private VNet endpoint by default, and instance-scoped features like SQL Agent, Service Broker, and cross-database queries. Choose MI when compatibility for a lift-and-shift is the constraint, SQL Database otherwise.
- Microsoft-managed versus customer-managed failover policy. Customer-managed (manual) failover keeps the decision and timing with the business and is the documented recommended default; Microsoft-managed (automatic) fires only on a widespread regional outage after a grace period and removes that control. If a runbook needs humans to make the call, it is customer-managed.
Worked example from the AZ-305 bank
lock_openFree sampleDesign Data Storage Solutionsmedium
Which characteristics correctly describe Azure SQL Managed Instance compared to Azure SQL Database? (Select 3 answers)
- AProvides close to 100% SQL Server compatibility for lift-and-shift workloads from on-premises.check_circle Correct
- BExposes the SQL endpoint only through a private IP address by default during provisioning.check_circle Correct
- CSupports cross-database queries, SQL Server Agent, and Service Broker natively in the engine.check_circle Correct
- DSupports the Hyperscale service tier for databases up to 128 TB capacity per instance.
- EPermits user databases up to 4 TB with native columnstore indexes on every service tier.
MI = near-100 percent SQL Server compatibility, private endpoint by default, Agent/Service Broker/cross-DB; Hyperscale is SQL Database only. Managed Instance is positioned as the high-compatibility PaaS landing zone: near-100 percent SQL Server feature compatibility, a private VNet endpoint by default, and instance-scoped features such as Agent, Service Broker, cross-database queries, and CLR. Hyperscale is exclusive to Azure SQL Database.
Why A is correct: Correct. Provides close to 100% SQL Server compatibility for lift-and-shift workloads from on-premises is one of the keyed answers. Managed Instance is positioned as the high-compatibility PaaS landing zone: near-100 percent SQL Server feature compatibility, a private VNet endpoint by default, and instance-scoped features such as Agent, Service Broker, cross-database queries, and CLR.
Why B is correct: Correct. Exposes the SQL endpoint only through a private IP address by default during provisioning is one of the keyed answers. Managed Instance is positioned as the high-compatibility PaaS landing zone: near-100 percent SQL Server feature compatibility, a private VNet endpoint by default, and instance-scoped features such as Agent, Service Broker, cross-database queries, and CLR.
Why C is correct: Correct. Supports cross-database queries, SQL Server Agent, and Service Broker natively in the engine is one of the keyed answers. Managed Instance is positioned as the high-compatibility PaaS landing zone: near-100 percent SQL Server feature compatibility, a private VNet endpoint by default, and instance-scoped features such as Agent, Service Broker, cross-database queries, and CLR.
Why D is wrong: Hyperscale is an Azure SQL Database service tier; Managed Instance offers General Purpose, Next-gen General Purpose, and Business Critical, with no Hyperscale tier.
Why E is wrong: The 4 TB cap and tier-specific columnstore behaviour described is an Azure SQL Database limit; Managed Instance has its own storage limits per tier and supports columnstore on the SQL Server engine surface.
What you must be able to do. Given recovery point and recovery time objectives with cost and scope constraints, design the backup, disaster-recovery, and high-availability solution that meets the RPO and RTO while respecting documented platform limits and replication trade-offs.
In one sentenceThe resilience domain: turning RPO and RTO targets into a backup, replication, and availability-zone or multi-region design that actually meets them within Azure's documented limits.
Recall check: answer these from memory first
- What is the documented replication split for high-priority versus lower-priority data across availability zones and regions, and why is synchronous cross-region not the default?
- What is the per-subscription limit on Recovery Services vaults, and how does a 600-vault isolation requirement comply with it?
- How often does Azure Site Recovery create crash-consistent recovery points, and can that cadence be tightened?
- Which behaviours apply to a Cosmos DB account with multiple write regions regarding consistency, downtime, and RPO?
What it tests. Designing for survival of failures. Backup and disaster recovery with Azure Backup (Recovery Services vaults and their per-subscription limits) and Azure Site Recovery (replication policy, fixed crash-consistent recovery-point cadence, RTO behaviour); and high availability with availability zones, availability sets, and multi-region architectures, including the documented replication split of synchronous across zones and asynchronous across regions, and Cosmos DB multi-region write behaviour against RPO and RTO targets.
How to study it. Anchor every answer to a recovery objective and a documented limit. Learn the WAF replication pattern as a default: synchronous replication across availability zones for high-priority data, asynchronous across regions for lower-priority data, because synchronous cross-region replication is rarely recommended due to latency. Learn the platform ceilings the exam loves: Recovery Services vaults cap at 500 per subscription (so a 600-vault isolation design must span subscriptions), and ASR crash-consistent recovery points are fixed at five minutes and cannot be tightened. For Cosmos DB continuity, tie multi-region writes to their RPO behaviour (under 15 minutes for Session, prefix, and eventual; no Strong consistency; no expected downtime when SDK region preferences are set). Practise mapping an RPO or RTO straight to the zone-versus-region and replication-mode choice.
Easy to confuse
- Availability zones versus multi-region for resilience. Availability zones spread resources across physically separate zones in one region to survive a zone failure with synchronous replication; multi-region spreads across regions to survive a whole-region outage, typically with asynchronous replication at higher cost and latency. Match the design to whether the objective is surviving a zone failure or a regional one.
- Microsoft-managed versus customer-managed ASR recovery-point cadence. Azure Site Recovery crash-consistent recovery points are fixed at every five minutes and cannot be modified; only the app-consistent snapshot cadence is configurable. A request to tighten crash-consistent RPO below five minutes must be declined and met outside ASR replication-policy settings.
Worked example from the AZ-305 bank
lock_openFree sampleDesign Business Continuity Solutionsmedium
You are designing a backup topology for a customer with 600 application teams who each want their own Recovery Services vault for isolation, with a target RPO of 24 hours and RTO of 8 hours per workload. Which design respects the documented per-subscription limit?
- ASpread the vaults across two or more subscriptions because each subscription supports a maximum of 500 Recovery Services vaults.check_circle Correct
- BCreate all 600 Recovery Services vaults inside one subscription; the per-subscription cap is documented as 1,000 vaults.
- CReplace Recovery Services vaults with Backup vaults to bypass the 500-vault cap, then attach VM and MARS workloads to the Backup vaults.
- DCreate 100 shared vaults in one subscription so each vault protects multiple teams, even though isolation will be lost.
Recovery Services vaults are capped at 500 per subscription; large-isolation designs span multiple subscriptions. Azure Backup documentation states that each Azure subscription supports up to 500 Recovery Services vaults. A 600-vault isolation requirement therefore must be split across two or more subscriptions to comply with the platform limit while keeping the workload-vault binding intact.
Why A is correct: Correct. Azure Backup documentation states that each Azure subscription supports up to 500 Recovery Services vaults.
Why B is wrong: The documented limit is 500 vaults per subscription, not 1,000, so 600 cannot fit in a single subscription.
Why C is wrong: MARS-agent and Azure VM extension backups are not supported in Backup vault, so the substitution does not satisfy the workload mix.
Why D is wrong: This sidesteps the customer's stated isolation requirement and is not driven by the documented per-subscription cap.
What you must be able to do. Given compute, application, migration, and networking requirements, choose the compute platform, application pattern, migration approach, and network topology that fit the stated scale, timeout, multi-region, and connectivity constraints at the least operational overhead.
In one sentenceThe heaviest domain: designing compute, application architecture, migration, and networking, and choosing correctly between VMs, containers, and serverless, the right messaging and API pattern, and the right hybrid topology.
Recall check: answer these from memory first
- What is the maximum timeout on the legacy Azure Functions Consumption plan, and which plans do you move to for a longer-running job?
- Which API Management tier supports active-active multi-region deployment under a single instance, and which newer tier does not?
- Which pattern fits a workload that is mostly reads where the read tier must scale independently of the write tier, and why?
- Name three signals that route a containerised workload to AKS rather than to Container Apps or ACI.
What it tests. The largest slice of the exam, designing the workload itself. Compute selection across virtual machines, containers (AKS, Container Apps, ACI), and serverless (Azure Functions hosting plans and their timeout limits); application architecture with microservices, messaging, integration patterns such as CQRS and Saga, and API Management tier selection; migration planning for moving workloads from on-premises or other clouds; and network design with hub-and-spoke topologies, hybrid connectivity, and network security.
How to study it. Spend the most time here because it is the biggest domain by weight, and most of it is matching a requirement to a platform, tier, or pattern. For compute, fix the decision tree: route to AKS when the workload needs the Kubernetes API, an existing Helm and operator ecosystem, or the large node-scale envelope, and to Container Apps or ACI for serverless containers without running Kubernetes. Learn the Functions plan timeouts (legacy Consumption tops out at ten minutes, so a longer job needs Flex Consumption, Premium, or Dedicated). For application patterns, pick CQRS when reads and writes must scale independently under a high read-to-write ratio. For API Management, remember multi-region active-active deployment is a Classic Premium exclusive, not Premium v2. For networking, design hub-and-spoke with shared services in the hub and connectivity radiating to spokes. Practise matching each constraint to its tier or pattern.
Easy to confuse
- AKS versus Container Apps or ACI. AKS is the choice when the workload needs direct Kubernetes API access, an existing Helm and operator ecosystem, or scale into thousands of nodes; Container Apps and ACI give serverless containers (event-driven scaling, single short-lived groups) without operating Kubernetes day to day. If the scenario names Kubernetes itself or its ecosystem, it is AKS.
- API Management Classic Premium versus Premium v2. Active-active multi-region deployment under one API Management instance is supported only on the Classic Premium tier; Premium v2 adds availability zones and higher scale units but does not include multi-region. When the requirement is gateways in several regions under one instance, it is Classic Premium.
Worked example from the AZ-305 bank
lock_openFree sampleDesign Infrastructure Solutionshard
A global API platform must run gateways in three Azure regions under a single API Management instance with active-active multi-region routing. Which tier supports this?
- AUse the Premium v2 tier and add regional gateways during provisioning.
- BUse the Standard v2 tier with outbound VNet integration in three regions.
- CUse the Classic Premium tier for the multi-region API Management deployment.check_circle Correct
- DUse the Consumption tier and rely on automatic regional scaling.
Multi-region API Management deployment is a Classic Premium exclusive; Premium v2 does not support it. The feature table explicitly lists multi-region deployment as supported only on Classic Premium. Premium v2 adds availability zones and 30 scale units but does not include multi-region; this is one of the exam's key tier differentiators.
Why A is wrong: Premium v2 supports availability zones and 30 scale units but does not offer multi-region deployment; that capability is reserved for Classic Premium.
Why B is wrong: Standard v2 offers outbound VNet integration and workspaces but cannot host a single multi-region API Management service.
Why C is correct: Correct. The feature table explicitly lists multi-region deployment as supported only on Classic Premium.
Why D is wrong: Consumption is a serverless gateway with no multi-region deployment, no static IP, and no built-in cache; it cannot satisfy the requirement.