Question bank · DevOps & Cloud
20 real DevOps & Cloud interview questions.
These are the prompts we put in front of platform, SRE and infrastructure engineers — verbatim, not a summary. Pick one and answer it out loud; Vocalis scores the answer on pacing, structure, filler words, tone and directness, and tells you which numbers it measured and which the coach judged. Audio stays on your device.
FinOps
4 questions
- 01
VPC & Network Cost Optimization
How do you architect a multi-account AWS or Azure environment to reduce NAT Gateway and cross-AZ data transfer costs by 90% without sacrificing high availability for outbound traffic?
- 02
Advanced Spot Orchestration
Walk through designing a Kubernetes node group strategy that seamlessly blends On-Demand and Spot instances, ensuring stateful workloads are protected while stateless microservices maximize spot discounts.
- 03
Serverless vs. Containerization Constraints
Under what specific traffic patterns, cold-start latency requirements, and cost constraints would you choose a serverless architecture (AWS Lambda/Azure Functions) over a dedicated Kubernetes cluster?
- 04
Multi-Tenant Quota Management
How do you implement and enforce strict chargeback models, namespace resource quotas, and cost-allocation tagging for disparate engineering teams sharing a single large cloud footprint?
Kubernetes
4 questions
- 05
Deep Network Troubleshooting
A pod in an EKS or AKS cluster can resolve external DNS but cannot communicate with another service in a different namespace. Walk me through your exact diagnostic steps using ip-route, CNI logs, and iptables or eBPF.
- 06
Etcd Disaster Recovery
Your Kubernetes cluster's etcd database suffers severe data corruption. Detail the exact procedure to restore the cluster state from an external backup while minimizing workload disruption.
- 07
Ingress at Extreme Scale
Compare the architectural differences and performance implications of using a traditional Nginx Ingress Controller versus an Envoy-based API Gateway (like Cilium or Istio) for handling 100,000+ requests per minute.
- 08
StatefulSet Migration
Explain how you safely migrate a running PostgreSQL StatefulSet with a 500GB Persistent Volume Claim (PVC) from one availability zone to another with near-zero downtime.
CI/CD
4 questions
- 09
Ephemeral CI Environments
How do you architect a GitHub Actions or Jenkins pipeline that automatically provisions, seeds, and tears down fully isolated testing environments for every individual pull request?
- 10
Terraform State Refactoring
A team inherits a monolithic Terraform state file that takes 20 minutes to run terraform plan. How do you safely refactor this into decoupled modules and remote state data sources without destroying and recreating the live infrastructure?
- 11
Self-Hosted Runner Architecture
What are the security, cost, and scaling trade-offs of using auto-scaling EC2/VM instances for GitHub Actions runners versus container-based runner controllers (like ARC) inside Kubernetes?
- 12
Pipeline Guardrails
How do you implement automated DevSecOps guardrails (using tools like Checkov or OPA) that physically prevent a developer from bypassing mandatory security checks or merging infrastructure with critical vulnerabilities?
AI & MLOps
4 questions
- 13
Model Serving & GPU Orchestration
How do you architect a scalable inference pipeline for serving LLMs (using frameworks like vLLM) on Kubernetes, specifically addressing GPU fractional sharing and custom metric auto-scaling?
- 14
Vector Database Trade-offs
Discuss the architectural differences and operational overhead of deploying a vector database (like Qdrant or Milvus) as a managed cloud service versus hosting and scaling it on your own persistent Kubernetes infrastructure.
- 15
ML Training Pipelines
How do you integrate machine learning model training and dataset validation into a standard CI/CD flow, ensuring that model weights are immutable and versioned alongside the infrastructure code?
- 16
GPU Resource Contention
GPUs are expensive and scarce. How do you implement advanced Kubernetes scheduling constraints (taints, tolerations, priority classes) to ensure batch training jobs utilize idle compute without starving real-time inference workloads?
SRE & Security
4 questions
- 17
Zero Trust Implementation
How do you design a Zero Trust access architecture for engineers connecting to private internal tools (like OpenCode or internal dashboards), replacing legacy VPNs with identity-aware proxies?
- 18
Automated Incident Remediation
Explain how you would design an event-driven auto-remediation system to handle a recurring infrastructure alert (e.g., DiskSpaceExhausted on a legacy worker node) without requiring human intervention.
- 19
Zero-Downtime Secret Rotation
Walk through the automated rotation lifecycle of a highly sensitive production database credential. How do you ensure the application dynamically transitions to the new credential without dropping active user connections?
- 20
Systemic Post-Mortems
Describe a scenario where a production outage was initially blamed on "human error." How do you structure blameless post-mortems to identify the systemic tooling failures and missing guardrails that allowed the human error to reach production?