Autocoder

Cloud Engineer Interview Questions and Answers

Last updated 2026-11-19

A cloud engineer interview is provider-neutral first and provider-specific second. Interviewers check that you can reason about service models, networking, identity, resilience, cost and migration, then they ask how you did it on AWS, Azure or Google Cloud. This page has 15 worked questions that apply across clouds, with a table of equivalent service names to keep your answers fluent whichever platform the job uses.

For service-level detail, use the dedicated pages for AWS, Azure and Google Cloud. For process and rounds, read how to crack a DevOps interview, which overlaps heavily with cloud hiring.

Provider-neutral themesService models, networking, identity, availability, cost, migration, security
Hands-on expectationsCLI and infrastructure as code, a scripting language, basic Linux and networking
Often paired rolesDevOps engineer, platform engineer, solutions architect

Equivalent services across the big three

Virtual machines are EC2 on AWS, Virtual Machines on Azure and Compute Engine on Google Cloud. Object storage is S3, Blob Storage and Cloud Storage. Managed Kubernetes is EKS, AKS and GKE. Serverless functions are Lambda, Azure Functions and Cloud Run functions. Identity is IAM, Microsoft Entra ID with Azure RBAC, and Google Cloud IAM. Virtual networks are VPC, VNet and VPC network, which is global on Google Cloud. Message queues are SQS, Service Bus or Storage Queues, and Pub/Sub. Monitoring is CloudWatch, Azure Monitor and Cloud Monitoring. Being able to translate lets you answer in the interviewer's cloud even when you know another best.

Cloud Engineer Interview Questions and Answers: questions with worked answers

  1. 1

    Explain IaaS, PaaS, SaaS and FaaS with examples, and the shared responsibility model.

    easy

    In IaaS the provider gives virtual machines, storage and networks, and you manage the OS upward, as with EC2. In PaaS you deploy code or containers and the provider manages the runtime and OS, as with App Service or Cloud Run. SaaS is a finished application, such as Microsoft 365. FaaS runs individual functions on events and bills per execution, as with Lambda. The higher the abstraction, the less you operate but the less you control. Under shared responsibility, the provider secures the underlying infrastructure and you secure your configuration, identities, data and in IaaS the guest OS, so misconfigured storage permissions are always the customer's problem.

  2. 2

    Compare public, private, hybrid and multi-cloud, and when each makes sense.

    easy

    Public cloud shares provider infrastructure across tenants with pay-as-you-go economics. Private cloud is dedicated, on premises or hosted, giving control for strict regulation or latency to existing systems. Hybrid connects the two, common during migration or where some data must stay local. Multi-cloud uses more than one public provider, to avoid lock-in, meet customer or regulatory requirements, or use a best-of-breed service. Multi-cloud has real costs: duplicated skills, tooling and egress charges, and lowest-common-denominator architecture. Many teams are multi-cloud by accident through acquisitions rather than by design, so be ready to talk about governance, not just ideals.

  3. 3

    Design a highly available three-tier web application in the cloud.

    medium

    Use at least two availability zones. A public load balancer in public subnets distributes to stateless application instances in private subnets, managed by an autoscaling group or container service across zones. The database is a managed relational service with a synchronous standby in a second zone and automatic failover, with backups and point-in-time recovery. Static content comes from object storage behind a CDN. Sessions and cache live in a managed in-memory store, not on instances. Outbound access uses NAT per zone, and secrets are in a secrets manager. Add health checks, autoscaling on request metrics, multi-region replication if the recovery objectives need it, and runbooks that are tested.

  4. 4

    What are RTO and RPO, and what disaster recovery strategies exist?

    medium

    RPO, recovery point objective, is how much data loss you can tolerate, measured in time since the last good copy. RTO, recovery time objective, is how long you can be down. Strategies trade cost for speed: backup and restore is cheapest with RTO of hours; pilot light keeps core data replicated and minimal compute off, scaling up in a disaster; warm standby runs a scaled-down copy that scales up; active-active runs full capacity in multiple regions with near-zero RTO and RPO but the highest cost and the hardest data consistency. Choose per workload from business impact, and test restores, because untested backups are assumptions.

  5. 5

    Explain CIDR notation and design the subnets for a VPC.

    medium

    A CIDR block such as 10.0.0.0/16 fixes the first 16 bits as the network, leaving 16 host bits, so 65,536 addresses. Each extra prefix bit halves the range: a /24 has 256 addresses, though clouds reserve a few per subnet (AWS reserves five, so a /24 gives 251 usable). Plan address space so it never overlaps with other VPCs or on-premises networks you may connect. For a /16 split into /20 subnets you get 16 subnets of 4,096 addresses, enough to place public, application and data tiers in each of three zones with room to grow. Use Python's ipaddress module to check your arithmetic.

    python
    import ipaddress
    
    vpc = ipaddress.ip_network('10.0.0.0/16')
    subnets = list(vpc.subnets(new_prefix=20))
    print(len(subnets), subnets[0], subnets[0].num_addresses)   # 16 10.0.0.0/20 4096
    print(ipaddress.ip_network('10.0.1.0/24').overlaps(ipaddress.ip_network('10.0.0.0/16')))  # True
  6. 6

    Describe the 6 Rs of cloud migration.

    medium

    Rehost (lift and shift) moves servers as they are, fast but without cloud benefits. Replatform makes small optimisations such as moving a database to a managed service. Refactor or re-architect redesigns for cloud-native patterns, the most effort and the most benefit. Repurchase replaces a system with SaaS. Retire switches off systems nobody needs, which often removes a surprising fraction of an estate. Retain keeps systems on premises for now, for latency, compliance or cost reasons. Start with discovery and dependency mapping, move low-risk workloads first to build confidence, and plan network connectivity, identity and landing-zone governance before moving anything.

  7. 7

    How do you implement least privilege and avoid long-lived credentials in the cloud?

    medium

    Give humans and workloads identities with narrowly scoped roles, grant to groups rather than individuals, and review permissions with access analysers that show unused grants. Replace static access keys with roles assumed through short-lived tokens: instance and pod identities for workloads, SSO for people, and OIDC federation for CI systems so pipelines never store cloud keys. Enforce MFA, use organisation-level guardrails such as service control policies or Azure Policy to block dangerous actions like disabling logging, require approval for privileged access and log every API call for audit. Rotate anything that must remain static.

  8. 8

    How would you reduce cloud costs without hurting reliability?

    medium

    Establish visibility first through tagging, cost allocation and budgets with alerts. Then right-size using utilisation metrics, schedule non-production environments off outside working hours, delete orphaned disks, snapshots and idle load balancers, and choose appropriate storage tiers with lifecycle rules. Buy commitments, such as savings plans, reserved instances or committed use discounts, for the steady baseline only, and use spot or preemptible capacity for stateless, interruptible work. Reduce data transfer by keeping traffic in-region and using private endpoints and CDNs. Review architecture: serverless can be cheaper for spiky load and dearer for steady heavy load. Treat cost as a metric engineers see, which is the core of FinOps.

  9. 9

    Vertical versus horizontal scaling, and what makes an application able to scale out?

    easy

    Vertical scaling increases the size of one machine; it is simple but capped, usually requires a restart and leaves a single point of failure. Horizontal scaling adds more instances behind a load balancer, giving elasticity and fault tolerance. To scale out, the application must be stateless: no local sessions or files, with state in shared stores; instances must be disposable and start quickly from an immutable image; configuration should come from the environment; and shared resources such as the database must handle the connection count, using pooling or proxies. Scale on a metric that reflects load, such as requests per target or queue depth, not only CPU.

  10. 10

    What is serverless, and what are its limitations?

    easy

    Serverless means you do not manage servers or capacity: the platform scales automatically, including to zero, and charges for usage. It covers functions, managed containers, queues, databases and API gateways. Limitations include cold start latency, execution time and memory caps, vendor-specific triggers and APIs that raise lock-in, harder local testing and debugging, concurrency limits that can overload downstream databases, and cost that can exceed always-on compute for steady high throughput. It is a strong fit for event-driven glue, spiky APIs and background jobs.

  11. 11

    Explain how DNS, load balancing and CDNs combine to serve a global user base.

    medium

    DNS resolves the hostname, and latency-based or geolocation routing sends users to the nearest healthy region. A CDN caches static and cacheable content at edge locations close to users, reducing latency and origin load, and terminates TLS near the user. Within a region a load balancer spreads requests across instances in several zones using health checks. DNS TTLs control how fast failover propagates, so keep them short for records that fail over but remember resolvers may cache longer. Cache invalidation and cache keys need design, and dynamic or personalised content must be marked non-cacheable or varied correctly.

  12. 12

    What is a landing zone and why do organisations build one?

    hard

    A landing zone is a pre-configured, secure multi-account or multi-subscription environment that new workloads are deployed into. It defines the account structure and organisation hierarchy, centralised logging and monitoring, identity federation, baseline network design (often hub and spoke), guardrail policies, security tooling and cost controls. The goal is that teams get a compliant environment fast without each reinventing it. Managed offerings include AWS Control Tower, Azure landing zone accelerators and Google Cloud's setup blueprints, usually implemented with Terraform. See the Terraform questions for how this is codified.

  13. 13

    Write a script that lists unattached disks in an AWS account across regions, to find waste.

    medium

    Use boto3 to enumerate regions with describe_regions, create a regional EC2 client, and page through describe_volumes filtered on status available, which means not attached to any instance. Sum the size for a rough monthly cost estimate. The script needs read-only permissions only, and should be run before any cleanup, with deletions done by a separate, reviewed step. The same approach works with the Azure and Google SDKs.

    python
    import boto3
    
    ec2 = boto3.client('ec2', region_name='us-east-1')
    regions = [r['RegionName'] for r in ec2.describe_regions()['Regions']]
    
    total_gib = 0
    for region in regions:
        client = boto3.client('ec2', region_name=region)
        pages = client.get_paginator('describe_volumes').paginate(
            Filters=[{'Name': 'status', 'Values': ['available']}]
        )
        for page in pages:
            for vol in page['Volumes']:
                total_gib += vol['Size']
                print(region, vol['VolumeId'], vol['Size'], 'GiB', vol['VolumeType'])
    print('Unattached total GiB:', total_gib)
  14. 14

    How do you monitor and secure a cloud environment continuously?

    medium

    Collect control-plane audit logs (CloudTrail, Azure Activity Log, Cloud Audit Logs) and network flow logs centrally into a protected account, and alert on high-risk events such as root usage, policy changes and disabled logging. Enable the native posture and threat services, such as Security Hub and GuardDuty, Defender for Cloud or Security Command Center, and scan infrastructure code and images before deployment. Track configuration against benchmarks such as CIS, and use automated remediation for clear-cut violations like public storage buckets. Metrics, logs and traces give operational visibility; both streams feed an on-call process with runbooks.

  15. 15

    What is the difference between a container, a VM and a function, and how do you choose?

    easy

    A VM virtualises hardware and carries a full OS, giving the strongest isolation and the most control, at the cost of start time and overhead. A container packages an app with its dependencies on a shared kernel, starting in seconds and packing densely. A function is code the platform runs per event, with no runtime management. Choose a VM for legacy software, special kernels or licensed appliances; containers for most microservices and portable workloads; functions for event glue and bursty tasks. Consider team skills and operational burden: a managed container platform often sits in the sweet spot. See the Docker questions.

Frequently asked questions

Which cloud should I learn first for a cloud engineer job?

AWS has the most job listings in India and globally, Azure is strong in enterprises that use Microsoft products, and Google Cloud is strong in data and Kubernetes teams. Learn one in depth and map the others by concept.

What is the difference between a cloud engineer and a DevOps engineer?

A cloud engineer designs and operates infrastructure on a provider, covering networking, identity, compute and cost. A DevOps engineer focuses on delivery pipelines, automation and reliability across that infrastructure. In practice the roles overlap and titles vary by company.

Do cloud engineer interviews include coding?

Yes, usually scripting in Python or Bash, sometimes SQL, and occasionally an easy algorithm question. The expectation is automation and API fluency rather than hard competitive programming.

How do I show cloud experience as a fresher?

Build a small project end to end with infrastructure as code: a VPC, an autoscaled app, a database, a pipeline and monitoring, then document the architecture and cost. A certification plus such a project is persuasive.

Related guides

More in Role-wise Interview Questions by Job and Tech Stack

See all in this section