DevOps Interview Questions & Answers
A curated, continuously updated set of 510 DevOps interview questions with worked answers, organised by topic and by company. Every answer carries the same four things: how to open, the substance, the follow-up the interviewer is holding in reserve, and the answer that sounds reasonable and fails the round. Written for DevOps, SRE and platform engineering loops from the graduate screen to the staff design round.
510 questions12 topics32 companiesupdated 2026
Grounded in researched DevOps, SRE and platform engineering interview loops, written to a senior-engineer editorial bar, and never padded to hit a word count.
What a DevOps interview actually tests
Five dimensions, and they are not equally weighted. Most preparation goes into the one that carries the least weight.
- Production troubleshooting
The round that decides most loops, and the one candidates prepare least. A broken system you did not build, incomplete information, and a score for how you narrow the search space rather than for the answer.
- Infrastructure design under constraint
Capacity, failure domains and blast radius, with arithmetic before architecture. When the prompt contains a number, that number is the question.
- Tool depth beneath the abstraction
Not what a Deployment is. Why a StatefulSet will not reschedule, what Terraform state actually maps, what the kernel did when the container died.
- Automation and coding
Real code in a shared editor. Idempotence, streaming a large input, and error handling matter more than algorithmic cleverness.
- Ownership and communication
An outage you commanded, a decision you lost, a risk you quantified for someone who does not write code. At staff level this is roughly half the loop.
Linux & Networking Interview Questions for DevOps
Signals and the OOM killer, cgroups v2 and namespaces, systemd, /proc, TCP states and TIME_WAIT, DNS resolution order, and the Bash/Python/Go a Google SRE or Meta PE screen actually asks you to write.
- What happens at the kernel level on SIGTERM versus SIGKILL, and why might a container ignore SIGTERM?Google · Meta · Amazon
- At what layer should this load balancer operate? Explain the trade-off between L4 and L7.Cloudflare · Google · Meta
- What happens between typing a URL and the page rendering? Go as deep as you can.Meta · Google · Cloudflare
- Load average is 30 on a 32-core box but CPU utilisation is 20%. Explain.Google · Meta · LinkedIn
- A busy proxy is failing to open new connections. TIME_WAIT is in the hundreds of thousands. What is happening?Cloudflare · Google · PhonePe
- Writes are failing with no space left on device, but df shows 40% free. What is going on?Amazon · Red Hat · TCS
- Write a script that processes a 50GB log file on a box with 512MB of RAM. What does the interviewer watch for?Google · Meta · Cloudflare
- Hard link or symlink: what actually differs, and when does the choice bite you?Red Hat · TCS · Infosys
- A freshly built VM boots and never gets an IP address. Walk me through DHCP and where it fails.TCS · Infosys · Accenture
- Walk me through a DNS lookup end to end. What is recursion actually doing, and who caches the answer?Cloudflare · Amazon & AWS · PhonePe
- Explain how a new process comes into existence on Linux. What do fork and exec each contribute?Red Hat · Google · Meta
- We resized the cloud volume but df still shows the old size. Walk me through extending the filesystem live.Infosys · Walmart Global Tech · Oracle
- Can a Linux host have two default gateways? What actually breaks, and what is the right way to do multihoming?JPMorgan Chase · Goldman Sachs · Accenture
- Your service is collecting thousands of CLOSE_WAIT sockets. What does that state mean and how do you hunt the cause?Netflix · Uber · LinkedIn
- You have written a backend service. Make it a proper systemd service. What belongs in the unit file?Red Hat · SAP · IBM
- Someone ran chmod -x on the chmod binary. Nothing can change permissions now. How do you recover?TCS · Infosys · Accenture
- How do you run tcpdump on a production host without making things worse, and what do you do with the capture?Cloudflare · Netflix · Datadog
- Walk me through an SSH connection from keystroke to shell. Why is agent forwarding considered risky?GitLab · JPMorgan Chase · Red Hat
- Explain what actually happens when cron fires a job, and where its silent failures come from.TCS · Infosys · Atlassian
- A database host shows swap used and memory climbing. Do you disable swap, tune swappiness, or leave things alone?Amazon & AWS · Databricks · Snowflake
AWS & Cloud Architecture Interview Questions
IAM least privilege, VPC design, security groups versus NACLs, why peering is not transitive, Transit Gateway routing, multi-account landing zones and multi-region failover across AWS, Azure and GCP.
- Security groups versus network ACLs: explain stateful and stateless, and give me a case where the difference bites.Amazon · Microsoft · Accenture
- Your service autoscales on CPU and still falls over during traffic spikes. Why, and what would you scale on instead?Flipkart · Uber · Amazon
- Walk me through how an AWS IAM request is evaluated. Where does an explicit deny fit?Amazon · JPMorgan Chase · Goldman Sachs
- You need to connect forty VPCs plus on-premises. Peering, Transit Gateway or PrivateLink?Amazon · JPMorgan Chase · Walmart Global Tech
- When is serverless the wrong choice? Talk me through cold starts and what they actually cost.Amazon · Microsoft · Atlassian
- An instance in a private subnet cannot reach the internet. Debug it in order.Amazon · TCS · Infosys
- What actually makes cloud computing different from running your own servers? Cut past the marketing.Amazon & AWS · Microsoft · Oracle
- An app that read from its S3 bucket yesterday gets AccessDenied today. Walk me through every gate that could have closed.Amazon & AWS · Flipkart · Walmart Global Tech
- The database failed over to its Multi-AZ standby and the application fell over anyway. What did Multi-AZ promise, what does it not?Amazon & AWS · Razorpay · JPMorgan Chase
- Everything is deployed across three AZs, yet one AZ impairment took your whole service down. How does that happen?Google · Amazon & AWS · Netflix
- A legacy workload must move off a dying data centre: lift-and-shift, re-platform or rebuild. How do you decide?Accenture · Infosys · IBM
- DNS failover with health checks is configured between two regions, yet users saw minutes of downtime. Why is it slower than it looks?Amazon & AWS · Cloudflare · Salesforce
- Storage spend doubled though traffic barely moved. Walk me through S3 storage classes and getting the bill back down.Amazon & AWS · Flipkart · Walmart Global Tech
- A nightly 40 GB export to S3 fails at 85 percent and starts over. How does multipart upload change that story?Amazon & AWS · Databricks · Flipkart
- Pick an EBS volume type for a Postgres primary with a nightly analytics batch. What decides it?Amazon & AWS · PhonePe · Goldman Sachs
- Reads are drowning our database primary. Someone suggested Multi-AZ. Does that help, and if not, what does?Razorpay · Swiggy · JPMorgan Chase
- A DynamoDB table provisioned at 6,000 writes per second throttles at 2,000. The partition key is a date shared by many items. Fix it.Amazon & AWS · Flipkart · PhonePe
- Where does Redis actually earn its keep in production, and which of its features disappoint teams?Swiggy · Razorpay · LinkedIn
- One event fans out to ten thousand Lambda invocations and the payments API behind them falls over. What is going on?Amazon & AWS · Stripe · Atlassian
- Why do mature AWS shops avoid IAM users for workloads? Where do instance profiles and IRSA fit in?Amazon & AWS · GitLab · Red Hat
Terraform & Infrastructure as Code Interview Questions
Terraform and OpenTofu state, remote backends and locking, drift detection, count versus for_each, module design, lifecycle blocks, Ansible idempotence, and the BUSL license split every interviewer now probes.
- count versus for_each in Terraform: when does list ordering cause a destroy and recreate?HashiCorp · Google · Amazon
- Name three things in the Terraform state file, and explain why it is a security artifact rather than a cache.HashiCorp · JPMorgan Chase · Accenture
- Two engineers run terraform apply against the same backend at the same time. What happens?HashiCorp · Accenture · Infosys
- Where does Terraform stop and Ansible start? Give me a case where using the wrong one hurts.Red Hat · Accenture · Infosys
- Your Ansible playbook reports changed on every run. Why is that a bug, and how do you fix it?Red Hat · IBM · Infosys
- One Terraform module tree, three environments. How do you lay out state so a production apply can never touch anything else?HashiCorp · Accenture · Infosys
- Five hundred resources were built by hand over three years and must come under Terraform. Walk me through importing them without breakage.HashiCorp · Accenture · IBM
- Resources keep changing outside Terraform and plans show surprises weeks later. How do you detect drift early, and who may change things?HashiCorp · Salesforce · Goldman Sachs
- When does duplicated Terraform code deserve extraction into a module, and how do you version modules so consumers do not break?HashiCorp · Accenture · Infosys
- A teammate asks you to approve their Terraform plan. Walk me through what you check before saying yes.JPMorgan Chase · Accenture · Infosys
- An apply ran in an order that broke the deploy. How does Terraform decide ordering, and where does depends_on fit?HashiCorp · Flipkart · Uber
- You spot a local-exec provisioner in a production Terraform module. Make the case for keeping or killing it.HashiCorp · TCS · Infosys
- Pick and defend a remote state backend for a 40-engineer platform. Compare the realistic options.HashiCorp · Goldman Sachs · Accenture
- Trace a database password from your Terraform pipeline to the resource. Where can it leak, and what closes each hole?JPMorgan Chase · Goldman Sachs · Razorpay
- Design the CI checks for a Terraform repository. What runs at each stage, and what can each check never catch?GitLab · Atlassian · Netflix
- A routine change wants to replace a production resource. How do Terraform updates become replacements, and what controls that?HashiCorp · Flipkart · Swiggy
- An apply planned to destroy and recreate the production database. Which lifecycle arguments belonged on that resource?Razorpay · Amazon & AWS · HashiCorp
- A module needs an AMI ID and three subnet IDs. Do you hardcode them, accept variables, or look them up?TCS · Accenture · Infosys
- Thirty teams ship from one Terraform repository. How do you cut it into stacks so one bad apply cannot touch everyone?Uber · LinkedIn · Walmart Global Tech
- Plans get scanned by policy-as-code before merge. Which rules earn a hard block, and who can waive one?JPMorgan Chase · Goldman Sachs · Cloudflare
Kubernetes & Docker Interview Questions
Control plane internals, the full kubectl apply to running pod path, scheduling and affinity, native sidecars, in-place pod resize, CNI and network policy, CSI and StatefulSet identity, RBAC, operators and service mesh.
- Walk me through everything that happens between kubectl apply and a running pod.Google · Meta · Red Hat
- What does a service mesh solve that Kubernetes does not, and when is the operational cost not worth paying?Uber · Meta · Red Hat
- When do you need a StatefulSet rather than a Deployment, and what does it actually guarantee?Databricks · Snowflake · Uber
- Liveness, readiness and startup probes: what does each one do when it fails, and how do you get them wrong?Uber · Shopify · Red Hat
- Explain requests versus limits, and how they decide which pod gets killed first.Google · Databricks · Uber
- How does a pod get an IP address, and what happens to traffic between two pods on different nodes?Cloudflare · Google · Red Hat
- A PersistentVolumeClaim has been Terminating for an hour. What is holding it, and how do you clear it safely?Databricks · Snowflake · Red Hat
- A service image is 1.2GB and deploys are slow. Get it under 100MB without breaking it.Red Hat · GitLab · Shopify
- Some nodes are at 90% and others are idle. Why does the scheduler allow this, and how do you fix it?Uber · Databricks · Red Hat
- Helm or Kustomize for managing manifests across three environments? Defend the choice.Red Hat · GitLab · SAP
- Why did eBPF-based networking displace iptables in Kubernetes, and what does that change for you?Cloudflare · Uber · Datadog
- HPA and VPA on the same deployment: why is that a problem, and what should you run instead?Databricks · Uber · SAP
- Lock down a namespace so pods can talk only to their own dependencies. Why does DNS break first, and how do you get it right?Red Hat · Goldman Sachs · Databricks
- What does a PodDisruptionBudget actually protect against, and why did your zero-downtime deployment drop requests anyway?Google · LinkedIn · Swiggy
- Databases will be provisioned by a Kubernetes Operator. What must the reconcile loop get right, and what when it cannot reach the API?Red Hat · Oracle · Databricks
- You need one log collector on every node, including nodes that join later. Which workload object do you reach for?Datadog · Red Hat · Amazon & AWS
- Envoy gets SIGTERM at the same instant as the app and requests drop mid-rollout. How does container ordering in a pod really work?Google · Uber · Stripe
- You updated a ConfigMap ten minutes ago and production still serves the old value. Explain how Kubernetes propagates configuration.Atlassian · GitLab · Walmart Global Tech
- Security wants rootless containers everywhere. What actually breaks, and where does the trade pay off?Red Hat · GitLab · IBM
- One container panics the host kernel and takes down forty neighbours. Where is the real isolation boundary, and what do you tighten first?Cloudflare · NVIDIA · Databricks
CI/CD, Jenkins & GitOps Interview Questions
Pipeline design and caching, artifact promotion, rolling versus blue-green versus canary and what each costs, progressive delivery with Argo Rollouts and Flagger, ArgoCD and Flux reconciliation, and the DORA metrics behind them.
- Blue-green, canary or rolling for a payments API: pick one and defend the resource and rollback trade-off.Stripe · Razorpay · Salesforce
- Explain fetch versus pull, then tell me when you would rebase rather than merge and what that does to a branch others have pulled.GitLab · Atlassian · Microsoft
- Your pipeline takes 45 minutes and developers have stopped trusting it. How do you fix it?GitLab · Atlassian · Shopify
- Ship a schema change to a live service with no downtime. Walk me through the sequence.Stripe · Shopify · GitLab
- Monorepo or many repos for forty microservices? Argue it from the operational side.Google · Meta · Uber
- Design the test strategy for a pipeline that must stay under ten minutes. What runs where?GitLab · Shopify · Atlassian
- Push or pull for deployments? Explain what pull actually removes from your threat model.GitLab · Atlassian · Stripe
- Walk me through what happens between a developer pushing a commit and a job starting on a runner.Microsoft · Accenture · Infosys
- A teammate kubectl-applied a fix during an incident and Argo CD reverted it a minute later. What happened?Red Hat · Razorpay · Salesforce
- Our Argo CD setup has forty services and grows every quarter. How should we organise Applications?Red Hat · Walmart Global Tech · Salesforce
- You deploy to production ten times a day. Does semantic versioning even mean anything for you?Atlassian · Stripe · Swiggy
- How do you land a refactor that needs three months without keeping a branch open for three months?Meta · Google · Atlassian
- It is release week and every Jenkins build queues behind another. Sort out the agent situation.TCS · Accenture · Infosys
- Why promote one artifact through dev, staging and prod instead of rebuilding for each environment?Amazon & AWS · Netflix · JPMorgan Chase
- You rolled back to the previous tag and errors continue. What is rollback actually made of?Flipkart · PhonePe · Uber
- You want forty engineers merging to main daily. How does trunk-based development survive contact with reality?Google · Meta · LinkedIn
- How do you decouple deployment from release with feature flags, and who retires the flags afterwards?Netflix · Flipkart · Swiggy
- Two engineers run the same build and get different dependency trees. Where do lockfiles fit in fixing this?GitLab · Atlassian · Stripe
- Our container registry passed ten terabytes and the bill keeps climbing. Design a retention policy.JPMorgan Chase · Goldman Sachs · Walmart Global Tech
- A stranger opens a pull request from their fork. What can that PR execute in your pipeline, and where does pull_request_target go wrong?Cloudflare · GitLab · Stripe
SRE, Prometheus & Observability Interview Questions
Metrics versus logs versus traces, PromQL, Prometheus cardinality blowups, OpenTelemetry collectors and temporality, exemplars, SLI and SLO design, error budget math, and burn-rate alerting that survives an on-call rotation.
- Define SLI, SLO and SLA, compute the monthly error budget for 99.95 percent, and tell me what changes when it is spent.Google · LinkedIn · Datadog
- Design a chaos engineering programme. What do you inject first, and how do you avoid causing the outage you were preventing?Netflix · Uber · Datadog
- Prometheus memory jumped 400% overnight. Find the cause and stop it recurring.Datadog · Uber · Grafana Labs
- On-call is drowning in alerts and starting to ignore them. Redesign the alerting.Google · LinkedIn · Netflix
- Counter, gauge, histogram, summary: which do you use for request latency, and why not the others?Datadog · Grafana Labs · LinkedIn
- A request crosses six services and you have no idea where the time goes. How does tracing fix that?Datadog · Uber · Stripe
- Logging costs more than the database and leadership wants the bill halved without losing signal. Design the pipeline.Datadog · Snowflake · Cloudflare
- Forty services each ship a different vendor agent. You are consolidating on OpenTelemetry. What improves, what breaks, what of the vendors?Datadog · GitLab · Salesforce
- You are building a new service. What do you instrument on day one so the first incident is diagnosable?Google · Stripe · LinkedIn
- How does Prometheus actually store time series? Walk me through the WAL, blocks, compaction and retention.Datadog · Cloudflare · GitLab
- Write the PromQL behind our availability SLO and tell me where these queries usually go wrong.Uber · Razorpay · PhonePe
- One Grafana serves three hundred teams and half the dashboards are abandoned. How do you organise it?Cloudflare · Walmart Global Tech · Red Hat
- Follow one log line from the application into Elasticsearch and onto the Kibana screen. Where does this path break at scale?Amazon & AWS · Flipkart · JPMorgan Chase
- Pick a trace-sampling strategy for our checkout service and defend it.Datadog · Stripe · LinkedIn
- Dashboards stayed green during an outage users definitely felt. Where does synthetic monitoring fit so this cannot recur?Salesforce · PhonePe · Swiggy
- Production is down and the war room is chaos. Who does what in a well-run incident?Google · Netflix · Goldman Sachs
- The error budget ran out. Describe exactly what happens next and what makes the policy stick.Google · LinkedIn · Goldman Sachs
- What can eBPF-based observability see that a normal agent cannot, and when is it not worth adding?Netflix · Cloudflare · Datadog
- Forty Kubernetes clusters, each monitoring itself. How do you make one observable estate out of this?Cloudflare · Uber · PhonePe
- Every nightly job shows green but analysts still get stale numbers. How do you set SLOs for a data platform?Databricks · Snowflake · Flipkart
DevOps Scenario-Based & Troubleshooting Interview Questions
The highest-signal round on the loop. CrashLoopBackOff with empty logs, exit code 137, ImagePullBackOff, stuck PVC finalizers, 503s from a healthy-looking service, p99 tails, incident command, and the blameless postmortem.
- A pod is in CrashLoopBackOff and the logs are empty. Debug it, step by step.Google · Uber · Flipkart
- A pod has been Pending for ten minutes. Walk me through your diagnosis.Flipkart · TCS · Accenture
- A service returns 503 but every pod behind it looks healthy. Debug it.Uber · Shopify · Flipkart
- Your p99 latency is three seconds while the average is two hundred milliseconds. Where do you look?Google · Datadog · Stripe
- A node goes NotReady. What happens to its pods, and on what timeline?Google · Meta · Red Hat
- Requests inside the cluster fail with DNS errors, but only sometimes. Diagnose it.Uber · Cloudflare · Shopify
- The pipeline says the deploy succeeded, but users still see the old version. Find it.GitLab · Atlassian · Meta
- You are paged at 3am: a production node is at 100% disk. What do you do, in order?Amazon · TCS · Infosys
- TCP retransmissions are spiking but the network team says there is no packet loss. Who is right?Google · Cloudflare · Meta
- Users get intermittent 504s but every backend service metric looks normal. Where is the problem?Cloudflare · Uber · Shopify
- TLS renewal is automated and has worked for months. One night every client starts failing certificate verification. What happened?Cloudflare · Amazon & AWS · Flipkart
- p99 spikes, probes time out, CPU utilisation reads 30 percent and nothing is OOMKilled. cpu.stat shows nr_throttled climbing. Explain it.Google · Uber · Datadog
- Checkout is failing with 'connection pool exhausted', but the database CPU sits at 10 percent. Where is the problem?Flipkart · Razorpay · Walmart Global Tech
- A busy node drops new connections intermittently; dmesg shows 'nf_conntrack: table full'. Walk me through what is happening.Amazon & AWS · Cloudflare · Uber
- One dependency slowed down and three services fell over completely. Walk me through the mechanics, then how you stop the next one.Google · Netflix · Uber
- Your last three incidents had nearly identical action items. What is broken about your post-mortem process?Google · LinkedIn · Goldman Sachs
- A VM rebooted after a storage change and dropped into emergency mode. Talk me through the recovery and what you check first.Microsoft · Oracle · IBM
- kubectl takes ten seconds, operator leader elections flap, and etcd fsync latency is the only red metric. What now?Google · Red Hat · Databricks
- Requests pile up at the nginx edge, one core is pegged at 100 percent, and the backends look bored. Walk me through it.Cloudflare · Flipkart · Swiggy
- After last night's firewall cleanup, calls to one payment vendor hang instead of failing fast. Find the rule that did it.Razorpay · JPMorgan Chase · Goldman Sachs
DevSecOps & Supply Chain Security Interview Questions
SAST, DAST, SCA and IAST placement in a pipeline, Trivy and Checkov gates, secrets management and rotation with Vault, image signing and SLSA provenance, zero trust in Kubernetes, and compliance as code.
- SAST, DAST, SCA and IAST: where does each belong in a pipeline, and what can none of them see?JPMorgan Chase · Goldman Sachs · Razorpay
- Implement zero trust inside a Kubernetes cluster. What does it mean beyond turning on mTLS?JPMorgan Chase · Goldman Sachs · Razorpay
- An AWS key was committed to a public repo two hours ago. Walk me through the next hour.JPMorgan Chase · Razorpay · GitLab
- Your CI stores a long-lived cloud access key. Replace it, and explain what the replacement actually proves.GitLab · Stripe · JPMorgan Chase
- SBOM, SLSA and Sigstore all address supply chain security. What does each one actually prove?Google · GitLab · JPMorgan Chase
- Sealed Secrets, External Secrets, SOPS or Vault CSI? Pick one for a regulated environment.Razorpay · JPMorgan Chase · Goldman Sachs
- You inherit 300 cloud accounts and thousands of roles. How do you get to least privilege without breaking production?Amazon & AWS · JPMorgan Chase · Goldman Sachs
- Design RBAC for a shared multi-team Kubernetes cluster. Which misconfigurations do you see most often?Red Hat · Flipkart · Razorpay
- PodSecurityPolicy was removed from Kubernetes. What replaced it, and how do you roll pod security out safely?Red Hat · GitLab · PhonePe
- A colleague says VLANs give the company network security. What is your honest take?TCS · Accenture · Infosys
- Walk me through hardening a server, from fresh install to production-ready.IBM · Accenture · TCS
- Why should a DevOps engineer care about the OWASP Top 10 if they never write the frontend?Stripe · Razorpay · Swiggy
- Should container image scanning block deploys or just warn? Defend your policy.Netflix · Uber · Walmart Global Tech
- Your images were scanned and signed, yet a pod starts behaving maliciously. How do you detect and respond?Cloudflare · Netflix · Datadog
- Describe the full lifecycle of a production secret, from creation through rotation to revocation.JPMorgan Chase · Goldman Sachs · PhonePe
- Audit season starts in eight weeks. How do you turn compliance from a document scramble into code?JPMorgan Chase · Goldman Sachs · Salesforce
- Your platform team merges infrastructure changes daily. Where does threat modelling fit without stalling delivery?Amazon & AWS · Microsoft · Goldman Sachs
- We still hand every employee a full-tunnel VPN. What does moving to ZTNA actually change?GitLab · Cloudflare · Accenture
- We put a WAF in front of the API gateway. Which attacks does it still let through?Flipkart · Razorpay · Cloudflare
- Your admins can delete the audit logs covering their own actions. How do you make the trail tamper-evident?JPMorgan Chase · Goldman Sachs · Stripe
DevOps System Design Interview Questions
Google-style NALSD with napkin math before boxes, multi-region active-active, disaster recovery under a real RTO, capacity planning, blast radius, and designing a delivery platform for hundreds of services.
- Design a disaster recovery plan for a 5PB data platform with a four-hour recovery objective.Google · Amazon · Snowflake
- Design multi-region active-active for a tier-1 API. What breaks, and what does it cost?Netflix · Stripe · Uber
- A payments monolith releases monthly and scaling means buying hardware. Design the migration.JPMorgan Chase · Razorpay · SAP
- An e-commerce platform takes 10x traffic during flash sales and the database locks up. Fix it.Flipkart · Swiggy · Amazon
- Make a Postgres database highly available. What are you actually promising, and what breaks?Amazon · Razorpay · Snowflake
- Design an authentication service for 100,000 QPS across three regions: 2KB payloads, 10ms processing, 16-core VMs. Do the math out loud.Google · Meta · Amazon & AWS
- When would you scale a system vertically rather than horizontally? Give me a case where vertical wins.Oracle · SAP · Goldman Sachs
- Design rate limiting for a public API: what algorithm, where does it live, and what happens when the limiter itself fails?Stripe · Cloudflare · Datadog
- Design caching for a product catalog read 50k times a second. How do you handle invalidation, and what happens when a hot key expires?Flipkart · Amazon & AWS · Swiggy
- Design a cron system that schedules two hundred thousand jobs across a shared fleet. What goes wrong at that scale?Google · Uber · LinkedIn
- A network partition split your database cluster in half and both sides kept serving writes. What does CAP predict about this?Google · Snowflake · JPMorgan Chase
- A payment request times out at the gateway, so the client retries. Design the API so money moves exactly once.Stripe · JPMorgan Chase · Goldman Sachs
- Design a URL shortener like bit.ly. What breaks first once the toy version works?Amazon & AWS · Flipkart · Microsoft
- Design the webhook platform that tells merchants when payments settle. Their endpoint is down. What happens?Stripe · Razorpay · GitLab
- Design a real-time leaderboard for a mobile game with ten million daily players.Flipkart · PhonePe · Swiggy
- Design notifications for a food delivery app. Order updates cannot be late; coupon blasts cannot spam.Swiggy · Uber · LinkedIn
- Design a feature flag service used by 200 services. Why keep ordinary evaluation off the network path?Atlassian · Goldman Sachs · Walmart Global Tech
- Postgres is out of headroom. How do you choose a shard key, and how do you reshard later without downtime?PhonePe · Flipkart · Uber
- Design type-ahead search for a marketplace with fifty million products. It must feel instant.Google · LinkedIn · Amazon & AWS
- Checkout calls five downstream services. Design backpressure so overload degrades instead of collapsing.Netflix · Datadog · Uber
Platform Engineering & FinOps Interview Questions
Internal developer platforms, golden paths, Backstage catalogs and templates, Crossplane, self-service without a ticket queue, plus cost allocation, showback, Karpenter, spot strategy and commitment management.
- Your cloud bill is two million dollars a year and leadership wants it cut. Where do you start?Amazon · Netflix · Snowflake
- Design an internal developer platform for fifteen product teams. How do you know if it worked?Atlassian · GitLab · SAP
- Finance wants to know what each team costs on a shared Kubernetes cluster. How do you answer them?Uber · Databricks · Snowflake
- Your cluster autoscaler adds nodes slowly and they are the wrong size. What would you change?Amazon · Datadog · Uber
- How do you decide between savings plans, reserved instances, spot and on-demand for a mixed fleet?Amazon & AWS · Netflix · Uber
- Your batch fleet runs on spot and keeps losing nodes mid-job. How do you make interruptions survivable?Amazon & AWS · Flipkart · Datadog
- Walk me through running a rightsizing programme across fifty teams, end to end.Accenture · Infosys · Walmart Global Tech
- Leadership wants proof cloud spend is efficient. Why report cost per transaction instead of percentage savings?Stripe · Uber · Netflix
- Half our resources are untagged and finance cannot allocate the bill. How do you fix cost allocation?TCS · Accenture · JPMorgan Chase
- Our multi-region architecture quietly doubled the networking bill. Where is the money going?Cloudflare · Netflix · Flipkart
- Pods request three times what they use. How do you reclaim Kubernetes waste without breaking reliability?Flipkart · Swiggy · Walmart Global Tech
- Where does serverless pricing actually bite, and when do functions beat VMs on cost?Amazon & AWS · Stripe · Razorpay
- Your Postgres bill tripled because three years of data live on one hot volume. How do you tier storage without breaking queries?Amazon & AWS · Snowflake · Flipkart
- Leadership wants a build-versus-buy comparison that ignores engineering salaries. Fix the analysis.Accenture · Goldman Sachs · Salesforce
- Who actually owns cloud cost efficiency at your company: a central team or every engineering team? Defend your operating model.Amazon & AWS · Snowflake · Salesforce
- An AI feature shipped and the model bill grew 8x in a month. How do you get control without turning it off?Stripe · Salesforce · Uber
- Your cloud provider offers a bigger discount for a three-year commitment. How do you decide?Snowflake · Uber · Netflix
- Your FOCUS cost dashboard disagrees with the invoice. How do you reconcile it without hiding commitment waste?Amazon & AWS · Microsoft · Walmart Global Tech
- Cloud spend missed forecast by 20%, but traffic grew 50%. Did engineering become less efficient?Netflix · Uber · Flipkart
MLOps & GPU Infrastructure Interview Questions
The fastest-growing DevOps sub-domain: GPU scheduling with MIG, time-slicing and DRA, vLLM paged attention and continuous batching, KServe and Ray, gang scheduling with Kueue, and the cost model behind inference.
- Device plugin versus Dynamic Resource Allocation for GPUs: why did Kubernetes need DRA?NVIDIA · Databricks · Google
- Four teams want GPUs and you have twelve A100s. Walk me through MIG, time-slicing and MPS, and how you would decide.NVIDIA · Google · Databricks
- A multi-node training job sits Pending forever while the cluster shows free GPUs. What is happening?NVIDIA · Meta · Databricks
- One node in your training fleet makes every job it touches 30 percent slower, but it passes health checks. How do you find and handle it?NVIDIA · Meta · Microsoft
- You need to upgrade GPU drivers across a live inference and training fleet. What is your plan?NVIDIA · Google · Microsoft
- Six teams share a GPU cluster. Two of them are idle most nights. Design the quota model.Databricks · Uber · Microsoft
- Training throughput is 40 percent below what the GPU count predicts. The GPUs look fine. Where do you look?NVIDIA · Meta · Microsoft
- You need 64 GPUs for the next year and the region is out of capacity. What is your procurement strategy?Microsoft · Oracle · Google
- Your cluster has three GPU generations because that is what you could buy. How do you schedule against it?NVIDIA · Microsoft · Oracle
- Design a GPU serving platform for several LLMs with autoscaling and a cost ceiling.NVIDIA · Databricks · Uber
- A vLLM pod serving an LLM OOMs at a request rate it handled fine yesterday, dropping in-flight requests. Walk me through it.Databricks · NVIDIA · Google
- How should you adapt web-service autoscaling for an LLM inference service?Databricks · NVIDIA · Meta
- How is rolling out a new model different from rolling out new code, and what does a model canary actually measure?Meta · Google · Databricks
- You have eight 80GB GPUs and a 70B model. Tensor parallel across all eight, or two replicas of four? Defend the choice.NVIDIA · Databricks · Meta
- One customer's batch job is making your interactive chat slow. Fix it at the serving layer.Databricks · Salesforce · Stripe
- Your inference replica takes nine minutes from pod scheduled to first token served. Where does the time go and what do you fix?NVIDIA · Databricks · Microsoft
- Someone proposes serving the quantised model to halve your GPU bill. What has to be true before you ship it?NVIDIA · Databricks · Microsoft
- Product wants to raise the context limit from 8k to 128k tokens. What happens to your infrastructure?Databricks · NVIDIA · Snowflake
- Long prompts are making short requests slow on the same replica. What is happening, and how would you fix it?NVIDIA · Databricks · Microsoft
- Downstream automation needs valid JSON every time. How do you get it, and what does it cost?Stripe · Salesforce · Snowflake
DevOps Behavioral & Leadership Interview Questions
Amazon Leadership Principles in STAR form, the outage story that shows ownership rather than blame, pushing back on a bad deadline, driving a migration nobody wanted, and explaining risk to people who do not write code.
- Tell me about an outage you owned.Amazon · Google · Netflix
- Tell me about a time you disagreed with a technical decision and lost.Amazon · Google · Atlassian
- What is the biggest technical mistake you have made, and what happened afterwards?Amazon · Netflix · Google
- Tell me about a time you pushed back on an unrealistic deadline. What happened?Accenture · TCS · Infosys
- It is Friday evening and your manager asks you to deploy to production to hit a customer commitment. Walk me through that conversation.TCS · Infosys · Accenture
- Two senior engineers are arguing in your design review and it is getting personal. You are running the meeting. What do you do?Google · Microsoft · LinkedIn
- Your on-call rotation is burning people out. Walk me through how you noticed and what you changed.Flipkart · Swiggy · PhonePe
- Tell me about mentoring a junior engineer who was struggling. How did you help without taking the work away?TCS · Infosys · Accenture
- Another team's service keeps degrading yours in production. How do you get them to fix it when you have no authority over them?Flipkart · Swiggy · Uber
- Tell me about a time you were blamed for an incident that was not your fault. What did you do?PhonePe · Razorpay · Swiggy
- Tell me about a time you had to tell leadership something they did not want to hear.TCS · Infosys · JPMorgan Chase
- You have just become a tech lead. How do you decide what to delegate and what to keep?Accenture · TCS · Walmart Global Tech
- Tell me about a time you said no to a stakeholder request that would have hurt reliability. How did you handle it?Amazon & AWS · Flipkart · PhonePe
- Everything on your board is marked urgent by somebody. How do you decide what actually gets done this sprint?TCS · Infosys · Accenture
- Tell me about harsh criticism you received in a performance review. What did you change as a result?Meta · Google · LinkedIn
- A teammate keeps merging changes that skip tests and rubber-stamp reviews. How do you handle it?GitLab · Atlassian · Stripe
- How do you get engineers to report their own mistakes early instead of hiding them?Netflix · Google · Cloudflare
- Tell me about an AI feature you shipped that did not work as promised. What did you do?Salesforce · Microsoft · Atlassian
- Reliability work keeps losing to features. How do you change that?Google · Stripe · Uber
DevOps interview questions by company
Every company runs DevOps, so what matters is which kind of loop you are walking into. These pages break down the process round by round, compiled from public sources, each with an explicit confidence level.
Before the interview: the role itself
DevOps interview questionsThe pillar page: every topic, every round, in one place.What is a DevOps engineer?The job behind the title, and what changed about it in 2026.DevOps vs SRE vs Platform EngineerSame tools, three accountabilities. Which loop you are in.DevOps roadmapFour levels, each with a benchmark you can actually prove.DevOps salaryReported bands by market, labelled base or total, caveats stated.The interview processFive round types, and what each one is really measuring.Who is hiringThree hiring markets, and how each one interviews differently.DevOps courseWhat formal training exists for this job, and whether it is worth buying.DevOps certificationsWhat CKA, Terraform Associate and a cloud badge actually buy.The NALSD design roundGoogle's SRE design interview, with the arithmetic worked.
DevOps interview questions: FAQ
What questions are asked in a DevOps interview?
Five kinds, in roughly this order of weight: a troubleshooting scenario in a system you did not build, an infrastructure design question with a number in the prompt, a coding or automation exercise, tool depth across Linux, Kubernetes, Terraform and CI/CD, and a behavioral round about an outage you owned. Definitions appear in the screen and almost nowhere after it.
How do I prepare for a DevOps interview in one week?
Drill troubleshooting scenarios for your target stack, memorise the error budget arithmetic (99.9 percent is about 43 minutes a month), and write three behavioral stories properly with numbers in them. Skip revising tool breadth. It is the cheapest round to lose and the most expensive to cram.
What is the hardest round in a DevOps interview?
Troubleshooting, and it is also the least prepared. You are given a broken system with incomplete information and scored on whether you narrow the search space with evidence rather than guessing. The specific failure candidates make is proposing the fix that worked last time instead of stating a hypothesis and the observation that would rule it out.
What DevOps questions are asked for experienced candidates?
The tool questions get replaced by ownership questions. Expect Terraform state and drift rather than Terraform syntax, why a StatefulSet will not reschedule rather than what a Deployment is, how you would roll a change across a fleet, how you set an SLO with a team that does not want one, and what you would cut from a cloud bill first.
Are DevOps interviews the same as SRE interviews?
They overlap heavily and differ in weight. A DevOps loop leans toward pipelines, infrastructure as code and cloud architecture. An SRE loop leans toward Linux internals, reliability arithmetic and distributed systems, and at large companies it is interviewed against the software engineering ladder, which raises the coding bar.
Do I need coding for a DevOps interview?
Yes, and treating it as a formality is a common way to fail. The round is usually Python, Bash or Go, on something practical like parsing a log stream or calling an API. Interviewers watch for whether the script is safe to run twice, handles the empty input, and streams a large file rather than loading it into memory.
Is there a DevOps interview questions PDF?
One page holding every DevOps interview question on this site, grouped by topic and by company, each with its worked answer. It is regenerated as questions are added, which is the thing a downloaded PDF cannot do. Every entry links to the full answer rather than a one-line note.
Unlock every answer, including the follow-ups
510 questions across 12 topics, ordered easy to hard, each with a worked senior-engineer answer, the probe an interviewer holds back, and the mistake that fails the round. The first questions in every track are free.
Loops get revised constantly and differ by team, by level and by region. Prepare in this direction, then ask your recruiter what the rounds actually are.