cert-bank

Guides

이 글을 한국어로 읽기

AZ-305 Study Guide: Read the Constraint First, Because One Phrase Decides Most Architect Questions

AZ-305 rarely asks whether you know a service. It asks whether you can pick between two or three services that would all work, and the stem nearly always contains one phrase that rules the others out: "without changing the connection string", "with no data loss if a zone fails", "at the lowest cost". This guide is built around those phrases. Learn to spot the constraint, and know which design each one points to.

Study guide · Microsoft Azure Solutions Architect Expert (AZ-305)

AZ-305 Study Guide: Read the Constraint First, Because One Phrase Decides Most Architect Questions

The exam

Microsoft Learn's exam-duration page gives associate and expert exams 100 minutes without labs and 120 minutes when they may contain labs; you see your exact time when you register. Passing score is 700 of 1,000. Microsoft does not publish a fixed number of questions, so pace yourself by the clock. Expect case studies: a long scenario describing the current environment, the requirements and the constraints, followed by several questions. Once you take a break you cannot return to questions you have already seen, including those in a case study. You can open Microsoft Learn inside the exam, but no extra time is added and the clock keeps running. The certification itself also requires the Azure Administrator Associate certification (AZ-104).

Skill areas: Design identity, governance, and monitoring solutions (25–30%), Design data storage solutions (20–25%), Design business continuity solutions (15–20%), Design infrastructure solutions (30–35%).

How the questions are built

Among the 251 AZ-305 questions we wrote for this question bank, 57% are use-case questions ("which service should you recommend?") and another 21% are explicit trade-offs. About one stem in six carries a visible constraint word such as without, minimize, lowest or fewest, and many more hide the constraint in a requirement such as an RTO. Nearly one question in five asks for two answers. So the skill being tested is elimination: every wrong option is a real, working design that breaks one stated requirement.

The constraint phrases and where they point

Phrase in the stemWhat it usually decides
"Keep the same connection string after a regional failover"A failover group (its listeners keep their names), not active geo-replication alone and not geo-restore, which both change the endpoint.
"RPO of minutes"Replication: Azure Site Recovery for VMs, a replica or failover group for databases. Azure Backup of a VM gives an RPO of hours (every 4 hours at best with the Enhanced policy).
"Survive a zone failure with automatic failover and no data loss"Zone redundancy (or, for SQL Server on VMs, an Always On availability group with synchronous replicas in different zones). Site Recovery to another zone is disaster recovery, not this.
"SQL Server Agent, cross-database queries, CLR" plus "no operating system to patch"Azure SQL Managed Instance. If an agent must run on the server's OS, or FILESTREAM is used, only SQL Server on a VM fits.
"One database, used now and then, no usage history"Serverless compute in Azure SQL Database. Many such databases: an elastic pool.
"Reachable from on-premises over private connectivity"A private endpoint. A service endpoint keeps the public endpoint and cannot be used from on-premises.
"Allowed regions", "allowed VM sizes", "required tag"Azure Policy, assigned as high in the management group tree as the rule applies. RBAC decides who may act, not what may exist.
"Sign-in must keep working if the datacentre is down"Password hash synchronization, which validates sign-ins in the cloud. Pass-through authentication depends on on-premises agents.
"Orders must be processed in order per customer, with duplicates dropped"Service Bus, with sessions and duplicate detection (business messages). Event Grid pushes discrete events; Event Hubs ingests high-volume streams.
"Global HTTP app" versus "any protocol, regional"Front Door is global Layer 7; Application Gateway is regional Layer 7; Load Balancer is Layer 4; Traffic Manager is DNS-based.
Going deeper: why a 30-minute RTO rules out the Microsoft-managed failover policy

A failover group has a failover policy. With the Microsoft-managed policy, Microsoft fails over all such groups in a region only during a widespread outage, and only after a grace period that cannot be set below one hour. That design protects you from failing over for a short blip, but it means the database can stay unavailable for at least an hour before anything happens. If the stem gives an RTO under an hour, or says the company wants to decide when and which group fails over, the answer is the customer-managed policy. Replication is asynchronous either way, so a forced failover can still lose the latest transactions; "no data loss" across regions is not something a failover group promises.

The four areas, in order of weight

Infrastructure (the largest)

Compute choices (VMs, scale sets, App Service, AKS, Container Apps, Functions plans, Azure Virtual Desktop), messaging, API Management, migrations with Azure Migrate and the Cloud Adoption Framework, and networking: hub-spoke, VPN Gateway versus ExpressRoute, Virtual WAN, Azure Firewall SKUs, private endpoints and load balancing. Treat retired options as automatically wrong: Basic Load Balancer and Basic public IPs support no zones and are retired.

Identity, governance and monitoring

Hybrid sign-in (PHS, PTA, federation), Conditional Access, PIM, managed identities, Key Vault, the management group hierarchy, Azure Policy and locks, and Log Analytics workspace design. A common trap: adding workspaces because of data volume. Microsoft's guidance is to start with one workspace and add more only for a concrete reason, such as data that must stay in a region or different retention needs.

Data storage

The three ways to run SQL Server, the vCore tiers (General Purpose, Business Critical, Hyperscale), serverless and elastic pools, TDE versus Always Encrypted, Cosmos DB APIs and consistency levels, blob tiers and redundancy, and data integration with Data Factory and Fabric. Azure SQL Database is the most frequent service in our question set (one question in ten), followed by Blob Storage, Azure Backup and Cosmos DB.

Business continuity

Start every question from RPO and RTO. Backup keeps recovery points for restore; Site Recovery replicates for failover; database replicas and failover groups cover the data tier. Then match the failure scope: an availability set protects against a rack failure, availability zones against losing a datacentre, a second region against losing the region.

Beginner tip: draw one small app on paper: an App Service plan, an Azure SQL database and a storage account. Beside each, write three lines: what survives a rack failure, a zone failure and a region failure, and the exact setting that changes the answer (a Premium plan with at least two instances and zone redundancy, a zone-redundant database plus a failover group, ZRS or GZRS). When you can fill in all nine cells without notes, most continuity questions become lookups.

Study order

  1. Data storage first: it gives you the vocabulary (tiers, replicas, redundancy) that the continuity questions reuse.
  2. Business continuity, straight after, while the data options are fresh.
  3. Infrastructure: compute, then messaging, then networking. It is the largest area, so give it the most days.
  4. Identity, governance and monitoring.
  5. Case-study practice: before reading the questions, write each requirement and constraint as a list, then answer against the list.
Going deeper: why "we back it up to the paired region" is not a disaster recovery plan

A geo-redundant Recovery Services vault does copy backup data to the paired region, but without Cross Region Restore that copy is used only if Microsoft declares a disaster for the primary region. With Cross Region Restore you can restore there whenever you choose, yet the data is still a scheduled backup (daily with the Standard policy) that is then replicated, so Learn notes the secondary-region RPO for VM backups can reach 36 hours, and every disk must be rebuilt from the vault before the VM runs. That fits a server that can lose a day. When the stem gives an RPO of minutes, only replication, through Site Recovery or a database replica, meets it.

AZ-305: a study order where each area reuses the one before
1
Data storage
SQL options and vCore tiers, Cosmos DB, blob tiers and redundancy: the vocabulary continuity reuses
2
Business continuity
Straight after, while the data options are fresh: RPO and RTO first, then the failure scope
3
Infrastructure
Compute, then messaging, then networking. The largest area, so give it the most days
4
Identity, governance and monitoring
Hybrid sign-in, Conditional Access, management groups, Azure Policy, Log Analytics
5
Case-study practice
List each requirement and constraint before reading the questions, then answer against the list
Work left to right: each block uses terms the block before it introduced.

Practise

Practise in the question bank: every question explains each option, right and wrong. Start with the domain you just studied, then take a timed mock exam.

Skills measured as listed in Microsoft's AZ-305 study guide. Exam length, question types and service features change; check the current exam page and Microsoft Learn before you book.