Home AZ-305 Design Prep
Azure Solutions Architect Expert

The AZ-305 design guide

AZ-104 asked how to configure a service. AZ-305 asks which service to choose. Most questions give you a set of requirements and four plausible designs, and only one meets every requirement at the lowest cost or effort. This guide is built around those choices: what each service is for, where it stops being the right answer, and the requirement words that point to it.

01 · Format

What you're walking into

700passing score out of 1000 (scaled, not a %)
40–60questions, typically about 50
~100 minfor questions; plan on ~2 hours total
AZ-104Administrator Associate is required to earn the Expert certification
  • Question types: single choice, multi-select ("choose two"), hot-area (pick a value in each dropdown), drag-and-drop, Yes/No series, and at least one case study.
  • The case study is heavier than on AZ-104. Expect a long scenario with several tabs (overview, existing environment, requirements, technical constraints). Read the requirements first. Most wrong answers break a single requirement buried in one bullet.
  • Yes/No series: the same goal with a different proposed solution each time. You can't go back. Judge each solution on its own.
  • Microsoft Learn is available during the exam (open book). It's slow, so use it only to check a specific limit or tier feature.
  • No hands-on labs. Everything is design. You won't be asked for CLI syntax, but you will be asked which SKU or tier a feature needs.
02 · Weighting

Domain map

Weights are from the published skills outline (updated April 2026). The "after AZ-104" column is an estimate of how much new material each domain holds for someone who just passed the administrator exam.

DomainWeightAfter AZ-104What's new
Design identity, governance & monitoring25–30%MediumEntra Connect sign-in methods, PIM and Conditional Access licensing, Key Vault tiers, Log Analytics workspace design, where to route logs.
Design data storage20–25%HighAlmost all new: SQL Database vs Managed Instance vs SQL on VM, service tiers, Cosmos DB APIs and consistency, data integration services.
Design business continuity15–20%MediumRPO/RTO matching, SQL geo-replication vs failover groups, composite SLAs, Cosmos DB multi-region.
Design infrastructure30–35%HighCompute and messaging pickers, global load balancing, ExpressRoute and Virtual WAN, API Management, migration tooling.
03 · Schedule

A two-week study plan

Day 1

Baseline

Take the full practice exam cold, under timed conditions. Don't study first. The domain bars on the results page tell you where to spend the next two weeks.

Days 2–5

Data and infrastructure

These are 50–60% of the exam and the most new material. Read the cheat sheets and the service pickers, then practice those two domains until you score 80%.

Days 6–9

Identity and continuity

Same loop for the other two domains. Drill flashcards for 15 minutes a day across all domains.

Days 10–13

Full exams

A full timed exam every other day, then review every miss. Use "Retry missed" between exams.

Day 14

Light review

Numbers to memorize and classic traps only. Stop early and sleep.

04 · Identity, governance & monitoring · 25–30%

Identity, governance & monitoring cheat sheet

Hybrid sign-inPick it whenWatch out
Password hash sync (PHS)Default choice. Sign-in must keep working if on-prem is down. You want leaked credential detection.On-prem account policies (logon hours, lockout) aren't checked at sign-in.
Pass-through auth (PTA)Passwords must not leave on-prem in any form, or on-prem policies such as logon hours and disabled accounts must apply immediately.Needs agents on-prem. If they're all down, cloud sign-in fails (unless PHS is also on as a backup).
Federation (AD FS)On-prem smart-card sign-in or a third-party MFA server that Entra can't do natively.Most infrastructure and effort. Rarely right unless a requirement forces it.

Entra ID features & licenses

  • Conditional Access: P1. Require MFA, compliant device, or location based on conditions.
  • Privileged Identity Management (PIM): P2. Just-in-time, time-bound role activation with approval and justification. Works for Entra roles and Azure RBAC roles.
  • Identity Protection (user and sign-in risk policies): P2.
  • Access reviews and entitlement management (access packages): P2 or Entra ID Governance.
  • Entra Connect Sync vs Cloud Sync: Cloud Sync uses lightweight agents and suits multiple disconnected forests. Connect Sync supports more advanced scenarios such as device writeback.

External identities

  • B2B collaboration: partners sign in with their own work identity as guests in your tenant.
  • Entra External ID (external tenant): customer-facing apps with self-service sign-up, social identity providers, and custom branding. It replaces Azure AD B2C for new customers.
  • Cross-tenant access settings control inbound and outbound B2B trust, such as trusting the partner's MFA.

App identity & secrets

  • Managed identity wherever the Azure resource supports it. No secrets to store or rotate.
  • System-assigned: lifecycle tied to one resource. User-assigned: separate resource, shareable across many resources, can be created before them.
  • Key Vault Standard: software keys. Premium: HSM-protected keys. Managed HSM: single-tenant, FIPS 140-3 Level 3 validated, you control the security domain.
  • Use the Azure RBAC permission model for Key Vault (the access policy model is legacy). Turn on soft delete (7–90 days) and purge protection for customer-managed keys.
  • Store app settings in App Configuration, with Key Vault references for secrets.

Governance

  • Management groups → subscriptions → resource groups. Assign policy and RBAC as high as possible so new subscriptions inherit it.
  • Azure landing zones (Cloud Adoption Framework): a standard management group hierarchy with platform (identity, management, connectivity) and application landing zones.
  • Azure Policy for guardrails, initiatives to group policies. Use DeployIfNotExists to auto-deploy diagnostics or agents.
  • Azure Blueprints is retired. Use template specs plus deployment stacks (which can deny changes to the resources they manage) and policy.
  • Separate environments by subscription for billing, quota, and blast-radius isolation. Use resource groups for lifecycle grouping.
  • Defender for Cloud: secure score and CSPM recommendations, plus workload protections (Defender for Servers, SQL, Storage, Containers).

Logging design

  • Log Analytics: query and alert with KQL. Storage account: cheapest long-term archive. Event Hubs: stream to a third-party SIEM or other tool.
  • Prefer one workspace per region or per tenant unless data residency, separate billing, or access isolation forces more. Use resource-context RBAC or table-level RBAC to limit what people see.
  • Analytics retention: 30 days included (90 with Sentinel), interactive up to 2 years, total (with long-term retention) up to 12 years.
  • Basic and Auxiliary table plans cut ingestion cost for high-volume, rarely queried logs.
  • Microsoft Sentinel (SIEM/SOAR) runs on top of a Log Analytics workspace.

Monitoring tools

  • Application Insights (workspace-based): app performance, dependencies, failures, distributed tracing.
  • Network Watcher: connection monitor, IP flow verify, next hop, VNet flow logs (NSG flow logs are retired).
  • Azure Monitor Agent plus data collection rules for VM logs and performance data.
  • Azure Resource Graph queries inventory across subscriptions. Workbooks build dashboards from logs and metrics.
05 · Data storage · 20–25%

Data storage cheat sheet

Relational optionPick it whenGives up
Azure SQL DatabaseNew cloud apps, single databases, SaaS with many databases (elastic pools). Least admin effort.No SQL Agent, no cross-database queries, no CLR, no Service Broker, no instance-level features.
Azure SQL Managed InstanceLift-and-shift with minimal changes: SQL Agent jobs, cross-database queries, CLR, Service Broker, Database Mail, linked servers. Deployed into your VNet.Some OS-level access, SSRS/SSIS/SSAS (use Power BI, Data Factory SSIS runtime, Analysis Services).
SQL Server on Azure VMsYou need OS access, a SQL version Microsoft doesn't run as PaaS, SSRS/SSAS on the same server, or third-party agents.You patch, back up, and run HA yourself (with automation helpers).
Azure Database for PostgreSQL / MySQL (flexible server)Open-source engines as PaaS. Zone-redundant HA and read replicas available.N/A

SQL tiers & purchasing

  • DTU (bundled) vs vCore (choose compute and storage separately). Azure Hybrid Benefit and reserved capacity need vCore.
  • General Purpose: remote storage, budget-friendly. Business Critical: local SSD, lowest latency, In-Memory OLTP, a free readable secondary, built-in zone redundancy option.
  • Hyperscale: up to 128 TB, near-instant backups, fast restore, up to 4 HA replicas and named replicas for read scale-out.
  • Serverless compute: auto-scales and auto-pauses (General Purpose), billed per second. Good for intermittent, unpredictable use.
  • Elastic pool: many databases with varying, non-overlapping peaks share resources. Classic multi-tenant SaaS pattern.

SQL security

  • TDE: encryption at rest, on by default. Customer-managed key from Key Vault if required.
  • Always Encrypted: encrypted in the client driver, so DBAs and Microsoft can't see plaintext. Secure enclaves allow richer queries.
  • Dynamic data masking: hides values in query results for non-privileged users. Not encryption.
  • Row-level security: filters rows per user, for example per tenant or per sales region.
  • Ledger: tamper-evident tables. Auditing to Log Analytics, Storage, or Event Hubs. Microsoft Entra-only authentication disables SQL logins.

Cosmos DB

  • APIs: NoSQL (native, newest features), MongoDB, Apache Cassandra, Apache Gremlin (graph), Table. Pick the API that matches existing code to avoid rewrites.
  • Consistency, strongest to weakest: Strong → Bounded staleness → Session (default) → Consistent prefix → Eventual. Stronger means higher latency and lower availability across regions.
  • Strong consistency isn't available with multi-region writes.
  • Partition key: high cardinality, spreads requests evenly, and appears in most query filters. Logical partition max 20 GB.
  • Throughput: provisioned (steady), autoscale (scales between 10% and 100% of max RU/s), serverless (spiky, low average, single region).
  • Global distribution: add regions with a click. Multi-region writes give 99.999% availability SLA.

Storage accounts

  • Standard general-purpose v2: default for blobs, files, queues, tables.
  • Premium block blob: low-latency, high-transaction blobs. Premium file shares: NFS or low-latency SMB. Premium page blobs: unmanaged disks.
  • Data Lake Storage = general-purpose v2 with hierarchical namespace: directory operations and POSIX ACLs for analytics.
  • Azure NetApp Files: enterprise NFS/SMB with very high throughput and low latency (SAP HANA, HPC, large file workloads).
  • Immutability: time-based retention (WORM, lockable for compliance) or legal hold. Can be set at container or version level.

Data integration

  • Azure Data Factory: ETL/ELT orchestration with 90+ connectors. A self-hosted integration runtime reaches on-prem sources. Runs existing SSIS packages with the Azure-SSIS IR.
  • Azure Databricks: Spark-based data engineering and ML. Synapse Analytics: data warehouse and Spark. Microsoft Fabric: unified SaaS analytics on OneLake.
  • Stream Analytics: SQL-like real-time queries over Event Hubs or IoT Hub streams.
  • Data Box: offline bulk transfer when the network is too slow for the data volume.

Data shape → service

  • Relational with transactions and joins → SQL family.
  • Semi-structured JSON, global scale, single-digit ms latency → Cosmos DB.
  • Simple key-attribute, cheapest → Table storage. Need global distribution or guaranteed latency → Cosmos DB for Table.
  • Files, images, video, backups → Blob. Shared SMB/NFS file system → Azure Files or NetApp Files.
  • Graph relationships → Cosmos DB for Gremlin. Time-series telemetry analytics → Azure Data Explorer.
  • Session state or caching → Azure Managed Redis (successor to Azure Cache for Redis).
06 · Business continuity · 15–20%

Business continuity cheat sheet

RPO is how much data you can lose (time since the last good copy). RTO is how long you can be down. Every continuity question gives you one or both. Pick the cheapest option that meets them.

Azure SQL DR optionScopeFailoverNotes
Geo-restoreDatabaseManual restoreUses geo-replicated backups. Cheapest. RPO about 1 hour, RTO hours.
Active geo-replicationSingle databaseManualUp to 4 readable secondaries in any region. Connection string changes on failover. SQL Database only.
Failover groupsGroup of databasesAutomatic or manualListener endpoints that don't change after failover. One secondary region. Works for SQL Database and Managed Instance.
Zone redundancyDatabaseAutomaticProtects against a zone failure only, not a region failure.

Backup design

  • Azure Backup for VMs, Files, Blobs, Disks, SQL/SAP HANA in VMs, AKS, PostgreSQL. Restores from points in time.
  • Enhanced policy for VMs: multiple backups per day (every 4 hours or more often), required for Trusted Launch VMs.
  • Cross Region Restore needs a GRS vault. Set redundancy before the first backup.
  • Ransomware protection: immutable vaults, soft delete, and multi-user authorization with a Resource Guard in a separate tenant or subscription.
  • Azure SQL automated backups: PITR 1–35 days (default 7). Long-term retention up to 10 years.

Site Recovery

  • Continuous replication of VMs (Azure-to-Azure, VMware, Hyper-V, physical) to a secondary region. RPO in seconds to minutes. RTO is the time to fail over and boot.
  • Recovery plans group VMs into boot order and run scripts or runbooks during failover.
  • Run test failovers into an isolated VNet regularly. They don't affect production.
  • Backup ≠ DR. Backup protects against deletion and corruption. Site Recovery protects against a site outage. Many designs need both.

Compute HA

  • Availability zones (99.99%) beat availability sets (99.95%). Use zone-redundant services where available.
  • VM Scale Sets (Flexible) spread instances across zones and fault domains and autoscale.
  • App Service and Functions (Premium) can be zone redundant with 2+ instances.
  • Multi-region active-active or active-passive needs a global load balancer: Front Door, Traffic Manager, or cross-region Load Balancer.

Storage & Cosmos DB continuity

  • Storage: GZRS / RA-GZRS for zone and region failure. Customer-managed account failover promotes the secondary region. RA- variants let apps read from the secondary without failing over.
  • Object replication copies blobs asynchronously to another account, which can be in another region.
  • Cosmos DB: add a region and enable service-managed failover. Multi-region writes for active-active.
  • Cosmos DB continuous backup: point-in-time restore within 7 or 30 days. Periodic backup is the default alternative.

Composite SLA

  • Services in series: multiply their SLAs. App Service 99.95% × SQL Database 99.99% ≈ 99.94%.
  • Redundant paths in parallel: 1 − (1 − A)(1 − B). Two independent regions at 99.94% each ≈ 99.9999%, before the global router's own SLA.
  • Adding a dependency always lowers the composite SLA. Adding redundancy raises it.
07 · Infrastructure · 30–35%

Infrastructure cheat sheet

Compute

  • VMs / Scale Sets: full OS control, lift-and-shift, licensed software that needs a server.
  • App Service: web apps and APIs, deployment slots (Standard+), autoscale. App Service Environment v3: single-tenant, fully isolated in your VNet, very high scale.
  • Functions: event-driven code. Consumption (5-min default timeout, 10 max). Flex Consumption (VNet integration, always-ready instances). Premium (no cold start, VNet, longer runs). Durable Functions for stateful orchestration.
  • Container Apps: serverless containers and microservices, scale to zero, KEDA event scaling, Dapr. No Kubernetes to manage.
  • AKS: full Kubernetes API, custom operators, node pools, service mesh.
  • Container Instances: single containers or short jobs, fastest to start, no orchestration.
  • Azure Batch: large-scale parallel and HPC jobs with job scheduling.
  • Azure Virtual Desktop: hosted Windows desktops and apps. Azure VMware Solution: run vSphere workloads on dedicated Azure hosts without converting them.

Messaging & events

  • Service Bus: enterprise messaging. FIFO with sessions, transactions, duplicate detection, dead-lettering, topics for pub/sub. Messages to 256 KB (Standard) or 100 MB (Premium).
  • Storage queues: simple, cheap, very large queue (TBs), 64 KB messages, server-side logs of every transaction.
  • Event Grid: reactive push of discrete events ("a blob was created") to handlers. Pay per event.
  • Event Hubs: high-throughput streaming and telemetry (millions of events per second), partitions, Capture to storage, Kafka endpoint.
  • Message = the sender expects work done. Event = a notification that something happened.

Integration & APIs

  • API Management: a single gateway for many backends. Policies for throttling, caching, transforms, JWT validation, versioning, and a developer portal.
  • APIM Premium for multi-region gateways, availability zones, and full VNet injection. Developer tier has no SLA. Consumption tier is serverless.
  • Logic Apps: low-code workflows with hundreds of connectors (SaaS, on-prem through the data gateway). Standard plan runs single-tenant with VNet integration.

Hybrid connectivity

  • Site-to-site VPN: encrypted over the internet. Quick and cheap, up to about 10 Gbps aggregate on the largest SKUs.
  • ExpressRoute: private circuit through a provider, predictable latency, SLA. Not encrypted by default (add MACsec on Direct or IPsec over it).
  • Private peering reaches VNets. Microsoft peering reaches Microsoft 365 and Azure public endpoints.
  • Global Reach: connects two on-prem sites through their ExpressRoute circuits over Microsoft's backbone. FastPath: bypasses the gateway for the data path. ExpressRoute Direct: 10 or 100 Gbps ports straight into Microsoft.
  • Use a VPN as a failover path for ExpressRoute, or two circuits in different peering locations for maximum resiliency.
  • Virtual WAN: Microsoft-managed hubs with any-to-any transit between branches (VPN, SD-WAN), users (P2S), ExpressRoute, and VNets. Pick it for many branches or regions.

Network security

  • Private endpoint: a PaaS service gets a private IP in your VNet. Reachable from peered VNets and on-prem. Needs private DNS zones.
  • On-prem clients resolving private endpoint names → Azure DNS Private Resolver inbound endpoint plus conditional forwarders on-prem.
  • Azure Firewall: central L3–L7 filtering in the hub with FQDN rules and threat intelligence. Premium adds TLS inspection, IDPS, and URL filtering.
  • WAF on Application Gateway (regional) or Front Door (global) for OWASP protection.
  • DDoS Network Protection covers a VNet's public IPs with cost protection and rapid response. DDoS IP Protection is per public IP.
  • Azure Bastion for RDP/SSH without public IPs on VMs.

Migration

  • Cloud Adoption Framework phases: Strategy → Plan → Ready → Adopt (migrate or innovate) → Govern, Secure, Manage.
  • Azure Migrate: discovery, dependency analysis, assessment (right-sizing and cost), and server migration (agentless for VMware).
  • Database Migration Service: online migrations (minimal downtime) or offline migrations to SQL Database, Managed Instance, or SQL on VMs.
  • The "R"s: rehost (lift and shift), refactor (small changes to use PaaS), rearchitect, rebuild, replace (SaaS).
  • Well-Architected Framework pillars: Reliability, Security, Cost Optimization, Operational Excellence, Performance Efficiency.
08 · Decisions

Service pickers

Most AZ-305 questions come down to one of these choices. Learn the deciding requirement in each row.

Load balancing

ServiceScopeLayerDeciding requirement
Front DoorGlobal7 (HTTP/S)Global web app, edge caching, WAF, instant failover, path routing
Traffic ManagerGlobalDNSGlobal routing for non-HTTP traffic, or endpoints outside Azure. Failover is limited by DNS TTL.
Application GatewayRegional7Path or host routing, TLS offload, WAF, inside a VNet
Load BalancerRegional4Any TCP/UDP, very low latency, internal or public
Cross-region Load BalancerGlobal4Global anycast IP in front of regional Standard Load Balancers, TCP/UDP

Messaging

Requirement wordsAnswer
Ordered (FIFO), transactions, duplicate detection, dead-letter, pub/sub to multiple subscribersService Bus (sessions, topics)
Simple queue, lowest cost, more than 80 GB of messages, audit log of every operationStorage queue
React to a state change (blob created, resource deleted), push, fan-out to handlersEvent Grid
Telemetry, clickstream, millions of events per second, replay, Kafka clientsEvent Hubs

Compute

Requirement wordsAnswer
Full OS control, custom drivers, legacy app that can't changeVMs (Scale Sets to scale)
Web app or API, minimize management, deployment slotsApp Service
Web app isolated on dedicated hardware in your VNet, no shared front endsApp Service Environment v3
Runs on a trigger, short, pay per executionFunctions (Consumption or Flex)
Containerized microservices, scale to zero, no cluster managementContainer Apps
Need the Kubernetes API, custom controllers, full cluster controlAKS
Thousands of parallel compute tasks, HPC, renderingAzure Batch
VMware workloads, no re-platforming, keep vSphere toolsAzure VMware Solution

Managed disks

DiskUseNote
Ultra DiskTop-tier IOPS and throughput, sub-millisecond latency (SAP HANA, top-tier SQL)Data disks only. Adjust performance without detaching.
Premium SSD v2Sub-ms latency, performance set independently of size, cheaper than UltraData disks only.
Premium SSDProduction and performance-sensitive workloadsCan be the OS disk. Single-VM 99.9% SLA.
Standard SSD / HDDWeb servers, dev/test / backups and infrequent accessN/A
09 · Recall

Numbers to memorize

ItemValue
VM SLA: zones / availability set / single Premium SSD99.99 / 99.95 / 99.9%
Cosmos DB availability: single region / multi-region writes99.99 / 99.999%
Cosmos DB consistency levels5 (Session is default)
Cosmos DB logical partition / item size max20 GB / 2 MB
Azure SQL point-in-time restore1–35 days (default 7)
Azure SQL long-term retentionup to 10 years
Active geo-replication readable secondariesup to 4
Hyperscale max database size128 TB
Service Bus max message: Standard / Premium256 KB / 100 MB
Storage queue max message64 KB
Functions Consumption timeout: default / max5 / 10 min
Log Analytics interactive / total retention2 years / 12 years
Key Vault soft-delete retention7–90 days
Activity log retention90 days
Management group depth (excluding root)6 levels
10 · Traps

Classic traps

"Minimize cost" and "minimize administrative effort" are tie-breakers. Several answers usually work. Pick the cheapest or most managed one that meets every requirement.
SQL Agent, cross-database queries, CLR, or Service Broker rule out Azure SQL Database. Managed Instance is the minimal-change PaaS target.
Managed Instance doesn't support active geo-replication. Use a failover group.
"Connection string must not change after failover" means failover group listeners, not active geo-replication.
Zone redundancy doesn't survive a region outage. Read whether the requirement says zone or region.
Traffic Manager is DNS. It never sees the traffic, so it can't do WAF, TLS offload, or path routing.
Strong consistency rules out multi-region writes in Cosmos DB.
Leaked credential detection needs password hash sync, even if you also use PTA or federation.
PIM and Identity Protection need P2. Conditional Access alone needs P1.
Dynamic data masking isn't encryption. To hide data from DBAs, use Always Encrypted.
Backup isn't disaster recovery. A one-hour RTO for a region failure needs Site Recovery or an active secondary, not a restore from backup.
ExpressRoute isn't encrypted by default, and its private peering doesn't reach Microsoft 365.
Ultra Disk and Premium SSD v2 can't be OS disks.
Azure Blueprints and Azure AD B2C (for new customers) are retired. If either appears, prefer deployment stacks / template specs, or Entra External ID.