Skip to main content
Careers · Interviews

Junior cloud engineer interview questions on AWS and Azure, with what a good answer covers

Junior cloud and cloud support interviews test whether you understand the building blocks well enough to reason about a fault: who is responsible for what, how identity and networking decide what can talk to what, and how to keep a system up and its bill under control. Answer with the mechanism, then with how you would check it.

9 min read Sources checked 17 November 2026

Junior cloud engineer and cloud support interviews test whether you understand the building blocks well enough to reason about a fault. Expect questions on who is responsible for what, how identity and networking decide what can reach what, how a system stays up, how its bill stays under control, and how infrastructure is written as code, often followed by a small design exercise. The questions below cover AWS and Azure, with what the interviewer is checking and what a good answer includes, cited to the vendors' own documentation.

The habit that marks out a good candidate is answering with the mechanism and then with how you would check it. "Security groups are stateful" is half an answer. "Security groups are stateful, so I wouldn't need an inbound rule for the replies, but a network ACL is stateless, so I'd check the ACL allows the return traffic" is the whole one.

Who looks after what

"Explain the shared responsibility model."

This checks that you know the cloud provider does not secure everything for you. AWS puts it as security "of" the cloud versus security "in" the cloud: AWS protects the hardware, software, networking and facilities that run its services, and the customer is responsible for what they build on top (AWS shared responsibility model). How much falls to you depends on the service. For an EC2 instance you manage the guest operating system, including its patches, the software you install and the security group around it. For a managed service such as S3, AWS runs the operating system and platform, and you manage your data, its encryption options and who has permission to it.

Microsoft's version makes the same point by service type, IaaS, PaaS and SaaS, and adds one line worth quoting: whatever the deployment type, you always own your data and identities. A good answer finishes with an example: if a virtual machine is compromised through an operating system patch you never applied, that is the customer's side of the line.

Regions, zones and staying up

"What is the difference between a region and an availability zone?"

A region is a separate geographic area, and AWS designs each region to be isolated from the others. Within a region are several isolated locations called availability zones, and spreading instances across zones protects an application from the failure of a single location (AWS regions and zones). In Azure, an availability zone is a group of one or more datacentres within a region, with independent power, cooling and networking (Azure availability zones).

The Azure page adds a distinction that a stronger answer uses. A zone-redundant resource is spread across zones by the service, and Microsoft handles failover. A zonal resource sits in the one zone you pick, and if you want it to survive a zone outage you have to deploy copies in other zones and handle failover yourself.

"Isn't high availability the same as backup?"

No, and the interviewer wants to hear why. Running across zones keeps a service up when a location fails, but it faithfully copies a mistaken deletion or corrupted data everywhere. Backup is for getting data back. Azure Backup describes itself as protection against accidental deletion, corruption and ransomware, and points to Azure Site Recovery for disaster recovery and failover. On AWS, AWS Backup centralises and automates backups across services, and EBS snapshots give point-in-time copies of a volume that you can restore a new volume from. Then say the thing that matters most: a backup you have never restored is a hope, not a plan.

Identity and permissions

"What is an IAM role, and why use one instead of access keys?"

An IAM role is an identity with permissions that can be assumed by whoever needs it, and it has no long-term credentials such as a password or access keys. Assuming it gives you temporary credentials for the session (IAM roles). AWS's IAM best practices ask for workloads to use temporary credentials through roles, for human users to sign in through an identity provider, and for least-privilege permissions. A good answer gives the practical reason: an access key pasted into a script on a server lasts until someone remembers to rotate it, while a role's credentials expire on their own.

"How does access work in Azure?"

Azure RBAC grants access through a role assignment, which has three parts: a security principal (a user, group, service principal or managed identity), a role definition (a set of permissions, broad like Owner or narrow like Virtual Machine Reader) and a scope (Azure RBAC overview). Scope can be a management group, subscription, resource group or single resource, and lower levels inherit role permissions from higher ones (scope in Azure RBAC). If you can explain those three parts and why you would assign to a group at the resource group rather than to a person at the subscription, you have answered it.

"A policy allows the action, but the request is still denied. Why?"

This is a reasoning test. In AWS, every request is denied by default unless a policy allows it, and an explicit deny in any policy that applies overrides any allow (how AWS evaluates requests). So a good answer lists where else a deny can live: a resource-based policy such as an S3 bucket policy, a permissions boundary, a service control policy from AWS Organizations, or a session policy (policy types). KBY's lesson on S3 AccessDenied from an EC2 instance role works through exactly this kind of fault.

Networks, and what can reach what

"What is a VPC, and what makes a subnet public or private?"

A VPC is a virtual network in your own logically isolated part of AWS, and each subnet in it sits in a single availability zone (what is Amazon VPC). The difference between public and private is the route table, not a setting on the subnet: a public subnet has a direct route to an internet gateway, and a private subnet doesn't, so its resources need a NAT device to reach the internet (subnet types). A NAT gateway lets instances in a private subnet open connections out while stopping anything outside from opening connections in (NAT gateways). The Azure equivalent is a virtual network, whose subnets span all the zones in a region.

"Compare security groups, network ACLs and Azure NSGs."

AWS publishes the comparison this question is after:

Security groupNetwork ACL
Applies toInstances associated with itEvery instance in the associated subnets
RulesAllow onlyAllow and deny
EvaluationAll rules, then decideIn number order until one matches
Return trafficAllowed automatically (stateful)Must be allowed explicitly (stateless)

An Azure network security group sits between the two. It has allow and deny rules, processed by priority from 100 to 4096 with lower numbers first, stopping at the first match, and it is stateful. Azure's default rules can't be deleted, only overridden by rules with higher priority. An NSG can be attached to a subnet, a network interface or both, and for inbound traffic the subnet's NSG is processed before the network interface's (how NSGs filter traffic), so traffic has to get past both.

"An instance in a private subnet can't reach the internet. Where do you look?"

A good answer follows the packet, in order:

  1. The subnet's route table: is there a default route (0.0.0.0/0) to a NAT gateway? Route tables send traffic to targets by destination address (route tables).
  2. The NAT gateway: a public one sits in a public subnet, has an Elastic IP address and routes on to the internet gateway.
  3. The security group's outbound rules.
  4. The network ACLs on both subnets, remembering that they are stateless, so the reply traffic needs its own rule.

KBY's lesson on an instance with no internet access follows the same path. The Azure version of this question has a newer twist: new virtual networks now default to private subnets (the Azure portal already does, and so do API versions released after 31 March 2026), so their virtual machines can't reach the internet until you add an explicit outbound method such as a NAT gateway (default outbound access).

Storage, cost and code

"Which kind of storage would you use, and when?"

Sort the services into three kinds. Object storage holds files as objects: Amazon S3 and Azure Blob Storage. Block storage is a disk attached to a virtual machine: an EBS volume attaches to an EC2 instance and behaves like a local drive, and Azure managed disks do the same for Azure VMs. File storage is a shared file system: Amazon EFS over NFS, and Azure Files over SMB or NFS (Azure Storage introduction). Then match them to uses: a website's images in object storage, a database server's data on block storage, a team's shared folder on file storage.

"How do you stop a cloud bill running away?"

Budgets and tags, and knowing what each one does not do. AWS Budgets alerts on actual or forecast spend and can trigger an action, such as applying an IAM policy. An Azure budget sends alerts, but Microsoft is explicit that resources aren't affected and consumption isn't stopped. Its cost data typically arrives within 8 to 24 hours, so an alert is never instant.

Tags tell you whose spend it is. They are key-value pairs, Azure resources don't inherit tags from their resource group, and tags are stored as plain text, so they must never hold anything sensitive (Azure tags). In AWS, tags only appear in cost reports once you activate them as cost allocation tags. KBY's lesson on AWS Budgets and cost anomaly detection shows a working setup.

"What is infrastructure as code, and what does terraform plan do?"

Infrastructure as code means describing resources in files you can version, review and reuse, instead of clicking through a console. Terraform works in three steps, write, plan and apply, and its configuration is declarative: you describe the end state, not the steps. terraform plan shows what would be created, changed or destroyed and changes nothing itself (plan command). Terraform keeps a state file that maps the configuration to the real resources. Bicep is Microsoft's declarative language for Azure, and CloudFormation does the same job on AWS with templates that create stacks. A follow-up you can expect is "what happens if someone changes a resource in the console?", and KBY's piece on fixing Terraform drift covers it.

"How would you know something is wrong before the users tell you?"

Name the service and what it gives you: Amazon CloudWatch for metrics, alarms, dashboards and logs, and Azure Monitor, which collects logs and metrics from Azure resources. Then say what you would alert on. Alerts tied to what users feel, such as errors and slow responses, beat a page of CPU graphs nobody reads.

The design question

Panels often finish with something like "design a simple, resilient web application for a small company". It checks structure, not a perfect diagram. Start by asking questions: how many users, how much downtime they can tolerate, what data it holds, what the budget is. Then build it up in layers and say why for each:

  • a VPC or virtual network with subnets in at least two availability zones, public subnets only for the load balancer, and the application and database in private subnets;
  • application servers spread across those zones, so losing one location doesn't take the service down;
  • a managed database with a standby in another zone, such as an RDS Multi-AZ deployment, plus backups that are tested by restoring them;
  • security groups that allow only the traffic each tier needs, and roles rather than stored keys for anything that calls a cloud service;
  • tags, a budget with alerts, monitoring, and the whole thing written as code.

Say what you would leave out to save money and what that costs in resilience. Naming a trade-off shows you understand the design rather than reciting it.

Getting ready

Build the design above in a free-tier or trial account, break it and fix it. KBY's Cloud Infrastructure and Operations lessons are short practical faults of the kind interviews use, and the certification paths show which exams lead where if you want one on your CV. If the role also involves Microsoft 365 or Active Directory, the systems administrator questions are worth reading too, and the STAR guide covers the behavioural half of the interview.

Courses and practice tests

Partner links
  • Tutorials Dojo · PaidTutorials Dojo: Solutions Architect Associate practice examsPractice exams for AWS Solutions Architect Associate, which covers much of the VPC, IAM and high availability ground in this guide.Check partner offer
  • Amazon · PaidAWS Certified Solutions Architect Study Guide: Associate SAA-C03 (4th edition)The Sybex study guide for AWS Solutions Architect Associate, with practice questions, for reading round the AWS answers here.Check partner offer
  • Amazon · PaidExam Ref AZ-104 Microsoft Azure Administrator (2nd edition)Microsoft Press's exam reference for AZ-104 Azure Administrator, which covers Azure RBAC, virtual networks, storage and monitoring.Check partner offer

Partner links are sponsored: KBY may receive a commission at no extra cost to you. Whether a provider pays KBY does not decide what we recommend, and the free official material comes first. Advertising disclosure.

Common questions

Do I need to know both AWS and Azure for a junior cloud role?
Usually not in depth. Read the advert and learn the platform it names properly, then learn the equivalent names on the other one (VPC and virtual network, security group and NSG, IAM role and managed identity), because panels often ask how the two compare.
What if the design question covers something I have never built?
Say so and reason from the parts you do know. Asking about users, downtime and budget before drawing anything, and naming what you would trade off, shows more than a diagram copied from a reference architecture.

Sources

Every figure and rule in this guide comes from these pages, read on 17 November 2026. Pay rates, exams and schemes change: check the source before you rely on one.

Spotted something wrong or out of date? Tell us and we will correct it.

All careers guides