Read this the night before or on exam morning. It assumes you’ve studied the notes. It refreshes, it doesn’t teach.
Exam day
- Microsoft doesn’t publish a fixed question count. Expect roughly 40 to 60 questions in about 100 minutes, in several formats: multiple choice, multiple response, drag and drop, dropdown (hot area) and case studies (a scenario with several questions, which you can’t revisit once you leave the section).
- Pass mark 700 out of 1000 (a scaled score, not a percentage). Wrong answers aren’t penalised, so answer everything.
- Domain weights (skills measured from 17 April 2026): identity, governance and monitoring 25-30%, data storage 20-25%, business continuity 15-20%, infrastructure 30-35%.
- The certification (Azure Solutions Architect Expert) renews each year through a free online assessment on Microsoft Learn.
How to read an AZ-305 question
Read the last line first, then list the constraints: cost, availability, RPO/RTO, compatibility, operational effort, security.
| The question says | Think |
|---|---|
| Minimize administrative effort | PaaS or serverless over VMs and custom scripts |
| Minimize cost | Reservations, Spot, serverless, lower tiers, lifecycle tiering. Only if requirements still hold |
| Highest availability | Zones then multi-region, zone-redundant tiers, failover groups |
| Without changing application code | Compatibility option (Managed Instance, same protocol, lift and shift) |
| Must not use credentials or secrets | Managed identity, workload identity federation |
| Least privilege / just in time | RBAC at narrow scope, PIM eligible assignment |
| Across all subscriptions | Management group plus Policy initiative |
| Private, not over the internet | Private endpoint, ExpressRoute |
| Minimal downtime migration | Online migration, continuous replication |
Answers that add a custom VM-based or script-based build where a managed feature exists are usually wrong. Watch for region and SKU limits hiding in the text (data residency, tier that lacks a feature).
Fast comparisons
| Compare | Rule |
|---|---|
| Site Recovery vs Backup | ASR keeps workloads running after region loss; Backup restores data after deletion or corruption |
| Availability set vs zones | Set 99.95%, zones 99.99%. Prefer zones |
| SQL Database vs Managed Instance vs VM | Modern PaaS vs near-full SQL Server compatibility vs OS control |
| Failover group vs geo-replication | Group gives stable listener endpoints and group failover |
| Hyperscale / Business Critical / General Purpose | Large and elastic / lowest latency / balanced |
| Serverless SQL vs provisioned | Auto-pause for intermittent vs always-on |
| Cosmos DB consistency | Strong > Bounded staleness > Session (default) > Consistent prefix > Eventual |
| Storage redundancy | LRS < ZRS < GRS < GZRS; RA = read secondary |
| Blob tiers | Hot > Cool (30 d) > Cold (90 d) > Archive (180 d, offline) |
| Service Bus vs Queue Storage vs Event Hubs vs Event Grid | Reliable broker / simple queue / streaming / reactive events |
| AKS vs Container Apps vs ACI | Full Kubernetes / serverless containers / single container |
| Front Door vs Application Gateway vs Load Balancer vs Traffic Manager | Global L7 / regional L7 + WAF / L4 / DNS |
| VPN vs ExpressRoute | Internet encrypted, cheap vs private, predictable, costly |
| Private endpoint vs service endpoint | Private IP in your VNet vs subnet allow-list to a public IP |
| Key Vault vs Managed HSM | Standard shared vs single-tenant FIPS 140-3 Level 3 |
| B2B vs External ID | Partner guests vs customers and citizens |
| PHS vs PTA vs federation | Default / passwords stay on-premises / special requirements |
| Policy vs RBAC vs locks | What resources look like / who can act / prevent delete or change |
Key numbers
| Fact | Value |
|---|---|
| Management group depth | 6 levels below root |
| Activity log retention | 90 days |
| Azure Monitor metrics retention | 93 days |
| SQL PITR | 1-35 days, default 7. LTR up to 10 years |
| Cosmos DB backup | Periodic (4 h, 2 copies) or Continuous (7 or 30 days) |
| Cosmos DB SLA | 99.999% multi-region |
| Data Box sizes | Disk 35 TB, Data Box 80 TB, Heavy 770 TB |
| ExpressRoute SLA | 99.95% |
| Soft delete for vaults | 14 days default |
| Hyperscale max size | 128 TB |
| Tags per resource | 50 |
Retired or renamed (don’t pick for new designs)
- Azure AD is Microsoft Entra ID; Azure AD B2C is replaced by Entra External ID for new customers.
- Azure Blueprints is retiring: use Template Specs, deployment stacks and Policy.
- Azure Cache for Redis is moving to Azure Managed Redis.
- SQL Database, MySQL and PostgreSQL Single Server are retired in favour of Flexible Server (PostgreSQL and MySQL).
- Basic Load Balancer, Basic public IP, default outbound access are retired or being retired.
- Log Analytics (MMA) agent retired in favour of the Azure Monitor Agent.
- Dedicated HSM isn’t for new customers: Managed HSM.
Design reflexes
- Design for the worst stated failure: zone, region, deletion, malicious admin.
- Cheapest design that meets every requirement wins; one unmet requirement disqualifies an answer.
- Prefer managed identities, private endpoints, RBAC, Policy over keys, public IPs and manual process.
- For every dependency in a request path, remember the composite SLA goes down.
- Case study: read the requirements and constraints list once, note them, then answer. Don’t re-read the whole scenario each question.