Skip to content
Tutorials Center
My learning

Topics

Lesson 56 of 57 ยท Practice and Review

Azure Interview Questions

Prepare for Azure interviews with 45+ questions and model answers by level: fundamentals, administrator and architect, plus scenarios and an AZ-900 to AZ-305 study map.

An Azure interview rarely checks whether you can recite a service list. Interviewers listen for how you reason: which service you would pick and why, what you would check first when something breaks, and which trade-off you accepted. This lesson collects more than forty questions that come up again and again, grouped by level, with answers short enough to say out loud.

Use it actively. Cover the answer, say yours, then compare. The "because" part earns the points. Everything about Azure itself is read-only here; two small stand-ins at the end let you rehearse the two things interviewers love most: picking a service for a scenario and doing availability math.

Answer in three beats: the direct answer, the reason, and the trade-off or exception. "Use a queue between the web tier and the worker, because it absorbs spikes and lets the worker fail independently; the cost is that the result is no longer immediate."

Fundamentals

1. What is cloud computing? Renting computing resources (servers, storage, databases, networking) on demand over the internet and paying for what you use, instead of buying hardware in advance. See What Is Cloud Computing?.

2. What are IaaS, PaaS and SaaS? Give an Azure example of each. IaaS gives you virtual hardware and you manage the operating system and up (Virtual Machines). PaaS gives you a platform and you manage the app and data (App Service, Azure SQL Database). SaaS is a finished application (Microsoft 365). The higher the level, the less you manage and the less control you have.

3. What is the shared responsibility model? Microsoft secures the physical datacenter, hardware and the platform layers it operates; you remain responsible for your data, identities, access and configuration. How much else is yours depends on the model: with a VM you patch the operating system, with PaaS Microsoft does.

4. Region, availability zone, region pair: what is the difference? A region is a set of datacenters in a geography. An availability zone is a physically separate location (own power, cooling and network) inside a region; spreading across zones survives a datacenter failure. A region pair links two regions in one geography for geo-redundant replication and staged platform updates. See Azure at a Glance.

5. Explain tenant, subscription, management group and resource group. A tenant is an Entra ID directory (the identity boundary). Subscriptions are billing and access boundaries linked to a tenant. Management groups organize subscriptions so you can apply policy and RBAC at scale. Resource groups hold related resources inside a subscription. See Accounts, Tenants and Subscriptions.

6. What is Azure Resource Manager? The management layer every request passes through, whether it comes from the portal, CLI, PowerShell, SDK or templates. It authenticates, authorizes (RBAC, policy, locks) and routes the request to the resource provider. That is why the same role and policy rules apply everywhere. See Resource Groups and Azure Resource Manager.

7. CapEx versus OpEx? Capital expenditure is up-front spending on assets such as servers. Operational expenditure is ongoing spending as you consume, which is how cloud billing works. The cloud turns a large fixed cost into a variable one.

8. Public, private and hybrid cloud? Public cloud runs on a provider's shared infrastructure. A private cloud is dedicated to one organization. Hybrid combines your datacenter with the public cloud, connected by VPN or ExpressRoute, often for data residency or gradual migration.

9. What happens if you delete a resource group? Everything in it is deleted, including resources whose names you may have forgotten, unless a lock stops it. That makes it a handy way to clean up a project, and a reason for locks on production groups.

10. Entra ID versus on-premises Active Directory? Entra ID is a cloud identity service built around HTTP-based protocols (OAuth 2.0, OpenID Connect, SAML) and tokens. On-premises Active Directory uses Kerberos, LDAP and group policy inside a network. They can be synchronized for hybrid identity. See Microsoft Entra ID.

11. Portal, CLI, PowerShell, Cloud Shell: when do you use which? The portal is good for exploring and one-off changes. The CLI and PowerShell are scriptable and repeatable. Cloud Shell is a browser terminal with both already authenticated. All of them call the same REST API. See Portal, Azure CLI and PowerShell.

12. How can you reduce Azure costs? Right-size and shut down idle resources, use auto-shutdown for dev VMs, buy reservations or savings plans for steady workloads, use Spot VMs for interruptible work, apply Azure Hybrid Benefit to existing licenses, use cheaper storage tiers, and set budgets and tags to see where money goes. See Pricing, Cost Management and Budgets.

13. Azure Policy versus RBAC? RBAC controls who can do what. Policy controls what resources are allowed to look like, regardless of who creates them (for example, only these regions, tags required, no public access). A user can have the right role and still be stopped by policy. See Azure Policy, Locks and Tags.

14. What is an SLA? A Microsoft commitment to a minimum monthly availability for a service, with service credits if missed. It is not a guarantee, and it applies only if you deploy as required (for example two instances or zone-redundant settings). Composite SLAs multiply.

Administrator level

15. System-assigned versus user-assigned managed identity? A system-assigned identity is created with, tied to and deleted with one resource. A user-assigned identity is its own resource that you can attach to many resources and that outlives them. Use user-assigned when several resources need the same permissions or you want to pre-grant access. See Managed Identities and Service Principals.

16. NSG versus Azure Firewall? An NSG is a free, stateful layer 3/4 filter attached to subnets or NICs, matching on IP, port and protocol. Azure Firewall is a managed, centralized service with FQDN rules, threat intelligence and logging across VNets. Use NSGs everywhere and a firewall for central egress and inspection. See Network Security Groups and Firewalls.

17. Load Balancer, Application Gateway, Front Door, Traffic Manager: how do they differ? Load Balancer is regional layer 4. Application Gateway is regional layer 7 with path routing and WAF. Front Door is global layer 7 with CDN and WAF. Traffic Manager is DNS-based global routing and never sees the traffic itself. See Load Balancer, Application Gateway and Front Door.

18. Explain storage redundancy options. LRS keeps three copies in one datacenter. ZRS spreads them across three zones. GRS adds an LRS copy in the paired region, and GZRS uses ZRS in the primary plus the paired region. The "RA" variants allow reading from the secondary. Geo-replication is asynchronous, so it has an RPO greater than zero. See Disks and Redundancy.

19. What are the blob access tiers? Hot, cool, cold and archive. Hot costs most to store and least to access; archive is cheapest to store but data is offline and must be rehydrated (hours) before reading. Lifecycle rules move data between tiers automatically. See Storage Accounts and Blob Storage.

20. What types of SAS exist and how do you make them safer? User delegation SAS (signed with Entra credentials), service SAS and account SAS. Make them safer with short expiry, least permissions, HTTPS only, IP limits, and stored access policies so you can revoke them. Prefer Entra authentication where possible. See Storage Security and SAS Tokens.

21. Availability set versus availability zones versus scale sets? Availability sets spread VMs across fault and update domains inside one datacenter. Zones spread across datacenters. Scale sets manage a group of identical VMs, with autoscale, and can span zones. Zones give the stronger protection. See Scaling and Autoscale.

22. Scale up versus scale out for App Service? Scale up moves to a bigger tier (more CPU, memory, features). Scale out adds instances of the plan behind the built-in load balancer. Web apps should be stateless so scale-out works. See App Service and Web Apps.

23. What is a deployment slot? A live staging environment of an App Service app with its own host name. You deploy and warm up there, then swap into production, which switches routing with no cold start and lets you swap back. See Deployment Slots, Blue-Green and Canary.

24. What are triggers and bindings in Functions? What is a cold start? A trigger starts a function (HTTP, timer, queue, blob). Bindings declaratively connect input and output data. A cold start is the extra latency when a new instance is created after idle time; Premium or dedicated plans keep warm instances. See Azure Functions.

25. Is VNet peering transitive? No. If A peers with B and B with C, A cannot reach C unless you add a firewall, gateway transit or direct peering. See Virtual Networks and Subnets.

26. Service endpoint versus private endpoint? A service endpoint keeps traffic on the Azure backbone and lets the service's firewall allow your subnet, but the service keeps its public address. A private endpoint gives the service a private IP in your VNet, so you can disable public access. See Private Endpoints, VPN and ExpressRoute.

27. DTU versus vCore for Azure SQL Database? DTU is a bundled measure of CPU, memory and I/O in fixed tiers, simple but coarse. vCore lets you choose compute, memory and storage separately, supports serverless and reserved capacity, and is the usual choice for new workloads. See Azure SQL Database.

28. Name the Cosmos DB consistency levels. Strong, bounded staleness, session, consistent prefix and eventual, from strictest to loosest. Session is the default and fits most apps: a client reads its own writes. Looser levels cost less latency and fewer request units. See Azure Cosmos DB.

29. Service Bus, Event Grid, Event Hubs, Storage Queues: when each? Service Bus for reliable business messages with ordering, transactions and dead-lettering. Event Grid for reactive, push-based event notification. Event Hubs for high-volume streaming telemetry. Storage Queues for simple, cheap, large queues. See Service Bus Queues and Topics and Event Grid and Event Hubs.

30. Key Vault: access policies versus RBAC? What do soft delete and purge protection do? RBAC is the recommended model: consistent roles and scopes. Access policies are the older vault-level model. Soft delete keeps deleted vaults and secrets recoverable for a retention period; purge protection stops anyone, even admins, from purging them early. See Key Vault.

31. Bicep versus Terraform? Bicep is Azure-only, compiles to ARM templates, has no state file and supports new Azure features quickly. Terraform is multi-cloud and keeps state. Both are declarative and idempotent. what-if (Bicep) and plan (Terraform) preview changes. See Bicep and ARM Templates and Terraform on Azure.

32. Backup versus Site Recovery? Azure Backup protects data with point-in-time restore points. Site Recovery replicates running workloads to another region to fail over during an outage. Backup guards against corruption and deletion; Site Recovery guards against losing a site. See High Availability, Backup and Disaster Recovery.

33. What do resource locks do? CanNotDelete allows reads and changes but blocks deletion. ReadOnly blocks changes and deletion. Locks inherit to child resources and apply even to Owners until the lock is removed.

34. Metrics, logs and traces in Azure Monitor? Metrics are lightweight numbers sampled over time. Logs are rich records queried with KQL in Log Analytics. Traces follow a request across services (Application Insights). Alerts can fire on any of them. See Azure Monitor, Log Analytics and Application Insights.

35. Which tools restrict who may reach a storage account? Entra authentication with RBAC, disabling shared key access, the storage firewall, private endpoints and public-access settings, plus policy to keep those settings in place.

36. Troubleshoot: a user cannot RDP to a VM. What do you check? Is the VM running? Does it have a public IP or are you using Bastion? Does an NSG at subnet or NIC level allow TCP 3389 from your source, and are priorities right? Is a firewall route sending traffic elsewhere? Is the guest firewall on? Use Network Watcher's connection troubleshoot and IP flow verify to find the blocking rule.

Architect level

37. Design a web app for 99.99% availability. Remove single points of failure: zone-redundant App Service (two or more instances per zone), zone-redundant database or a failover group, a global entry point such as Front Door, and a second region for regional failures. Then compute the composite SLA, because the critical chain multiplies. Add health probes, automation and a tested failover. The trade-off is cost and complexity.

38. How do you calculate a composite SLA? Multiply the availabilities of components in series. For redundant independent copies, the combined unavailability is the product of each unavailability. Shared dependencies break the independence assumption.

39. Define RPO and RTO and describe disaster recovery strategies. RPO is how much data you can lose (time). RTO is how long you can be down. Strategies from cheapest to fastest: backup and restore, pilot light (minimal core running), warm standby (scaled-down copy), and active-active. Lower targets cost more.

40. Explain hub-and-spoke. A central hub VNet holds shared services (firewall, gateway, Bastion, DNS); workload spokes peer to it. It centralizes security and connectivity and isolates workloads. Spoke-to-spoke traffic goes through the hub's firewall because peering is not transitive. See Private Endpoints, VPN and ExpressRoute.

41. Choose compute for a new containerized API with unpredictable traffic. Container Apps: scale to zero, request-based scaling, no cluster operations. AKS only if you need the Kubernetes API, custom operators or tight control. App Service containers fit steady web workloads. See Choosing a Compute Service.

42. How would you isolate customers in a multi-tenant SaaS app? Options range from a shared database with a tenant ID column (cheapest, weakest isolation) to a schema or database per tenant (elastic pools help) to a deployment per tenant (strongest, most expensive). Choose by compliance needs, noisy-neighbour risk and cost per tenant, and enforce tenant filters centrally.

43. How do you secure a three-tier app? Separate subnets per tier with NSGs allowing only the required flow, a WAF at the edge, private endpoints for data services with public access disabled, managed identities and Key Vault instead of secrets, Bastion instead of open management ports, and Defender for Cloud plus logging. See Security Best Practices and Defender for Cloud.

44. What is a landing zone? A pre-configured environment for workloads: management group hierarchy, policies, role assignments, networking (often hub-and-spoke), logging and identity, built so teams can deploy safely without redoing governance each time.

45. Request-driven versus event-driven design? Request-driven calls are synchronous: simple, but the caller waits and failures propagate. Event-driven designs use messaging, so components are decoupled, absorb bursts and fail independently, at the cost of eventual consistency and harder debugging.

46. Name the Well-Architected pillars and a trade-off between two. Reliability, security, cost optimization, operational excellence and performance efficiency. Example: multi-region active-active improves reliability and performance but raises cost and operational complexity. See The Azure Well-Architected Framework.

47. Migration approaches? Rehost (lift and shift VMs), replatform (move to managed PaaS with small changes), refactor (redesign for cloud-native), rebuild or replace. Choose per application by effort, risk and benefit rather than one strategy for everything.

Scenario questions

For scenarios, interviewers want your questions as much as your answer. Ask about traffic, data size, recovery targets and compliance before drawing a diagram.

A. Your Azure bill doubled last month. How do you investigate? Open Cost Analysis, group by service, resource group and tag, and compare with the previous month to find the change. Check for new or resized resources, orphaned disks and IPs, log ingestion spikes and egress. Contact the owner by tag, fix the cause, then add budgets, anomaly alerts and a policy so it does not repeat.

B. A function that processes orders sometimes processes the same order twice. Why, and what do you do? Queue triggers deliver at least once, and retries or lock expiry cause duplicates. Make the handler idempotent (an order ID check or upsert), use duplicate detection in Service Bus where available, and send repeated failures to a dead-letter queue.

C. The app works in the portal but a pipeline deployment fails with an authorization error. What do you check? Which identity the pipeline uses, which roles it has and at what scope, whether policy denies the resource, whether a lock blocks the change, and the activity log entry for the exact failing operation.

D. You must move 40 TB to Azure from an office with a slow link. Options? Compare network transfer time against an offline device such as Azure Data Box, or a hybrid with ExpressRoute. Calculate the transfer time from bandwidth before choosing. Then land the data in Blob Storage, probably in a cool tier.

Two stand-ins to rehearse with

These run in your browser and are not Azure. The first picks a service for a requirement set with simple rules, as a mental checklist; the second does composite SLA math for two designs.

JS Service-picking checklist

Press Run (or Ctrl + Enter) to see the console output here.

No output.

JS Composite SLA for two designs

Press Run (or Ctrl + Enter) to see the console output here.

No output.

AZ-900 to AZ-305 study map

Microsoft's role-based exams change over time, so always check the current skills outline on the exam page. This map shows which lessons here match the usual topics.

Exam Audience Lessons to study
AZ-900 Azure Fundamentals Beginners, non-technical roles Cloud basics, Azure at a Glance, Accounts and Subscriptions, Cost Management
AZ-104 Azure Administrator Operators, admins Entra ID, RBAC, Virtual Machines, Storage, Virtual Networks, Azure Monitor
AZ-204 Azure Developer Developers App Service, Azure Functions, Cosmos DB, Key Vault, Service Bus, Container Apps
AZ-305 Azure Solutions Architect Architects Well-Architected, Reliability and DR, Private Connectivity, Choosing Compute
AZ-500 Azure Security Engineer Security Security and Defender, Key Vault, Network Security, Policy

Hands-on time in a free account beats memorizing. Build a small project from the Real-World Projects chapter, break it on purpose, and fix it.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. What is the best one-sentence difference between RBAC and Azure Policy?

    RBAC is about permissions of identities. Policy evaluates resource properties and can deny or audit them regardless of the user's role.
  2. 2. A scenario asks for a queue with dead-lettering, sessions and transactions. Which service fits?

    Service Bus is the enterprise messaging service with dead-letter queues, sessions and transactions.
  3. 3. Which answer structure scores best in an interview?

    Interviewers evaluate reasoning. A direct answer with a reason and a trade-off shows you can decide, not just recite.
  4. 4. Why can a geo-redundant storage account still lose recent data in a regional disaster?

    Writes are acknowledged in the primary and copied to the secondary afterwards, so the last few moments may not have arrived. That is the RPO.

Summary

  • Interviewers test reasoning: answer directly, give the reason, and name the trade-off.
  • Know the fundamentals cold: regions and zones, scope hierarchy, shared responsibility, RBAC versus Policy.
  • For administrator roles, compare similar services (NSG vs firewall, endpoints, load balancers, redundancy options) and be able to troubleshoot step by step.
  • For architect roles, use composite SLA math, RPO and RTO, hub-and-spoke and the Well-Architected trade-offs.
  • Use the study map as a guide, keep hands-on practice central, and check the current exam outline before you book.

Next: Azure Cheat Sheet.