Interview Flashcards
Sixty questions, grouped by topic. Read the question, say the answer out loud, click the card to check. The fastest way to spot the gaps in your own story.
These aren’t trivia. Each answer is the short, honest version a senior engineer would give in an interview — the same ones the course chapters teach in depth. If a card feels uncomfortable, that’s where to spend a study afternoon.
Foundations.
Q01
In one sentence, what is DevOps?
Click to flip
A cultural and technical agreement that the same team builds, ships, and operates software — using version control, automation, and feedback loops to keep changes small and safe.
Q02
Name the four DORA metrics.
Click to flip
Deployment frequency, lead time for changes, change failure rate, and mean time to restore. They quantify "elite" delivery performance.
Q03
Why does small batch size matter?
Click to flip
Small changes are easier to review, test and roll back. Risk grows non-linearly with diff size, so 10 ten-line PRs are safer than one 100-line PR.
Q04
DevOps vs Agile?
Click to flip
Agile fixed dev's cadence. DevOps extends that cadence into operations, so finished work actually reaches users at the same speed it's produced.
Q05
What is a blameless postmortem?
Click to flip
An incident review focused on the system that allowed the failure, not the person at the keyboard. Goal is durable fixes; people speak honestly when they aren't on trial.
Q06
What is "you build it, you run it"?
Click to flip
The team that ships a service is also on its pager. Operational pain shows up in the team's backlog, where it can be fixed.
Q07
Define "shift left" and "shift right".
Click to flip
Shift left moves security and testing earlier (in the PR). Shift right embraces production techniques — canaries, feature flags, chaos — that learn from real traffic.
Q08
Symptom of a "DevOps team" anti-pattern?
Click to flip
A separate group writes pipelines and Terraform for developers, recreating the dev/ops wall. Healthier model: a platform team builds golden paths, product teams self-serve.
Pipelines.
Q09
Continuous Delivery vs Continuous Deployment?
Click to flip
Delivery: every green build is ready to ship; a human clicks. Deployment: every green build does ship, automatically.
Q10
List the six stages of a healthy pipeline.
Click to flip
Source, build, test, scan, deploy, verify. Anything missing is a hole in your safety net.
Q11
Compare rolling, blue/green, and canary deploys.
Click to flip
Rolling replaces instances in waves (cheap, slow rollback). Blue/Green runs two full stacks and flips (fast rollback, double cost). Canary routes a slice to the new version and ramps (safest, needs metrics).
Q12
Why decouple deploy from release?
Click to flip
Code can ship "off" behind a feature flag. Marketing turns it on per cohort. Deploys become routine; releases become product decisions.
Q13
Pipeline takes 45 minutes. How do you speed it up?
Click to flip
Parallelise independent stages, cache dependencies by lockfile hash, run only affected tests on PRs (test impact analysis), split slow integration tests into nightly jobs.
Q14
What is an immutable artifact and why does it matter?
Click to flip
A single tagged build (e.g. a Docker image) promoted unchanged from dev to prod. Guarantees "the bits that passed tests are the bits in production".
Q15
What is "the deployment pipeline" pattern?
Click to flip
An automated path from commit to production with progressively stricter quality gates — the same path for every change, including hotfixes.
Cloud as files.
Q16
Declarative vs imperative IaC?
Click to flip
Declarative ("here is the end state") — Terraform, K8s manifests. Imperative ("do these steps") — shell, classic Ansible playbooks. Declarative is idempotent by design.
Q17
Provisioning vs configuration management?
Click to flip
Provisioning creates the box (VM, network, DB) — Terraform's domain. Configuration management paints it (packages, files, services) — Ansible's domain. Containers absorb much of the latter today.
Q18
Why never commit
terraform.tfstate to Git?Click to flip
It contains plaintext secrets and resource IDs, has no locking, and two parallel applies will corrupt it. Use a remote backend (S3 + DynamoDB lock, GCS, Terraform Cloud).
Q19
Define drift detection.
Click to flip
A scheduled
terraform plan that alerts when real cloud state diverges from the code — usually someone changed a setting in the console "just this once".Q20
When are Terraform workspaces a bad idea?
Click to flip
For long-lived environments. They share one backend and one set of variables, so blast radius leaks across envs. Use separate directories or repos for dev/stage/prod.
Q21
What's
terraform import for?Click to flip
Bringing existing cloud resources under Terraform management. You write the matching config, then import the live ID into state.
Ship the machine with the app.
Q22
Container vs VM in one sentence.
Click to flip
A VM virtualises hardware and ships a full guest OS; a container virtualises the OS and shares the host kernel — orders of magnitude smaller and faster.
Q23
Why a multi-stage Dockerfile?
Click to flip
Build with a fat toolchain image, copy only the final artifacts into a slim runtime image. Smaller surface, faster pulls, fewer CVEs.
Q24
Why ordering matters:
COPY package.json before COPY .Click to flip
Each instruction is a cached layer. Copying manifests first lets
npm install stay cached when only application source changes.Q25
Avoid
latest tag — why?Click to flip
It's a moving target. The image you ran in CI yesterday is not the one you ran in prod today. Always pin a specific tag or digest.
Q26
Is "Docker secret in ENV" safe?
Click to flip
No. ENV values are visible in
docker inspect, layer history if baked at build time, and to any process in the container. Use runtime secret injection (vault, CSI, K8s Secret with proper RBAC).Q27
Container starts and immediately exits. First steps?
Click to flip
docker logs <name> for stdout/stderr; check the entrypoint and CMD; run interactively with --entrypoint sh; confirm required env vars are present.Declare. Reconcile. Repeat.
Q28
Pod vs Deployment?
Click to flip
A Pod is one running unit; a Deployment is a declarative wrapper that keeps N pods alive, manages rollouts and rollbacks via ReplicaSets.
Q29
When use StatefulSet over Deployment?
Click to flip
When pods need stable network identity and persistent per-pod storage — databases, message brokers. Pods are named
app-0, app-1, …Q30
Service vs Ingress?
Click to flip
Service gives pods a stable cluster-internal address. Ingress (or Gateway API) handles external traffic with hostnames, paths and TLS.
Q31
Readiness vs liveness probe?
Click to flip
Readiness: "should I receive traffic?" — fails ⇒ removed from Service endpoints. Liveness: "am I alive?" — fails ⇒ container restarted.
Q32
Requests vs limits?
Click to flip
Requests guarantee resources for scheduling. Limits cap usage; exceed CPU = throttling, exceed memory = OOMKill. Set both, especially memory.
Q33
Pod stuck in CrashLoopBackOff. Debug path?
Click to flip
kubectl describe pod for events, kubectl logs --previous for the dying container, exec into a debug image, check image tag, env vars, ConfigMaps and probes.Q34
What is HPA vs VPA vs Cluster Autoscaler vs Karpenter?
Click to flip
HPA: more pods. VPA: bigger pods. Cluster Autoscaler: more nodes within node groups. Karpenter: smarter, faster node provisioning that picks instance types per workload.
Q35
ConfigMap vs Secret — meaningful difference?
Click to flip
Both are key/value. Secret is Base64-encoded by convention, supports encryption at rest in etcd, and tooling treats it as sensitive (RBAC, audit). Not encryption by itself.
Production senses.
Q36
Monitoring vs observability?
Click to flip
Monitoring answers known questions with pre-built dashboards. Observability lets you ask new questions of high-cardinality data after an incident.
Q37
Name the three pillars and what each is best at.
Click to flip
Logs — "what happened?" (events). Metrics — "how much/how often?" (time-series). Traces — "where did the time go?" (per-request paths across services).
Q38
List the four golden signals.
Click to flip
Traffic, errors, latency, saturation. Put them on every service dashboard before anything fancier.
Q39
Define SLI, SLO and error budget.
Click to flip
SLI is the measurement (e.g. successful requests / total). SLO is the target (e.g. 99.9% / month). Error budget is 100% minus SLO — the permitted unreliability.
Q40
Why p99 latency instead of average?
Click to flip
Averages hide tail pain. A few users seeing 5-second responses ruin the impression. p99 (and p99.9) make tail latency visible and actionable.
Q41
What is OpenTelemetry?
Click to flip
A vendor-neutral standard (CNCF) for instrumenting traces, metrics and logs. Instrument once with the SDK, route to any backend via the collector.
Version control deep cuts.
Q42
Merge vs rebase?
Click to flip
Merge preserves real history with a merge commit. Rebase replays your commits on top of the target for a linear log. Rebase your own branch, merge into shared branches.
Q43
Trunk-based vs GitFlow?
Click to flip
Trunk-based: one branch, tiny PRs, feature flags — ideal for continuous deployment. GitFlow: develop/release/hotfix branches — suits versioned releases (mobile, SDKs).
Q44
git reset vs git revert?Click to flip
reset rewrites history — safe locally, dangerous on shared branches. revert creates a new commit that undoes another — safe public undo.Q45
What is GitOps?
Click to flip
Operating production by changing a Git repo. A controller (Argo CD, Flux) watches the repo and reconciles the cluster. Rollback = revert the commit.
Q46
When is force-push acceptable?
Click to flip
Only on personal feature branches you own, and even then use
--force-with-lease. Never on shared branches like main.CI tools.
Q47
Why declarative Jenkinsfile over scripted?
Click to flip
Structured, lint-able, predictable. Scripted Groovy gives unlimited power and unlimited rope — reserved for the rare case declarative can't express.
Q48
Should you build on the Jenkins controller?
Click to flip
No. Builds belong on agents (Docker, K8s, ephemeral). The controller hosts UI/config — protect it from compromise via build code.
Q49
GitHub Actions: hosted vs self-hosted runners?
Click to flip
Hosted: zero maintenance, fresh VM per job, included minutes. Self-hosted: your hardware, access to private networks/GPUs, you own patching and isolation.
Q50
How does GitHub Actions OIDC work?
Click to flip
A workflow requests an ID token from GitHub; the cloud's trust policy validates the issuer + claims (repo, ref) and returns short-lived credentials. No long-lived secrets stored.
Cloud-specific.
Q51
When pick Lambda vs Fargate vs EKS?
Click to flip
Lambda: tiny event-driven workloads. Fargate (ECS): single stateless container apps. EKS: real Kubernetes for multi-service systems needing ingress, mesh, custom controllers.
Q52
What is IRSA in EKS?
Click to flip
IAM Roles for Service Accounts — a pod assumes an IAM role via its K8s service account, getting AWS credentials without baking keys into the image.
Q53
Azure DevOps vs GitHub — pick one.
Click to flip
Azure DevOps for enterprise process (Boards, Test Plans, audit). GitHub for modern dev culture, OSS workflow, Copilot, where Microsoft ships new features first.
Q54
What is Azure workload identity?
Click to flip
The Azure equivalent of OIDC federation — a pipeline or pod authenticates via a federated Entra ID app, gets a short-lived token, no secret to rotate.
The senior round.
Q55
What is an SBOM?
Click to flip
Software Bill of Materials — a manifest of every component in an artifact (libraries, versions, licences). Generated by tools like Syft; required by SLSA & many regulators.
Q56
Why eBPF matters for DevOps in 2026?
Click to flip
Safe programs run inside the Linux kernel to observe and modify networking, syscalls and security — with no app changes. Powers Cilium, Pixie, Tetragon, Coroot.
Q57
When should AI agents auto-apply changes in prod?
Click to flip
Rarely — for narrow, well-bounded actions with clear rollback (e.g. scale, restart, rollback a deploy). Keep a human on novel or destructive changes.
Q58
What is an internal developer platform (IDP)?
Click to flip
A self-service layer over infra and CI that exposes "golden paths" to product teams — templates, scorecards, service catalogs. Backstage and Port are typical implementations.
Q59
A deploy goes wrong at 2am. What do you do first?
Click to flip
Stop the bleeding: roll back (or flip the feature flag), then investigate. Post-incident: blameless review and a small fix that prevents the class of failure.
Q60
If you had one week to improve a slow team's DevOps, what would you do?
Click to flip
Measure deploy frequency & lead time; pick one bottleneck (usually slow CI or manual deploy); automate that single path end-to-end; teach the team to repeat the exercise.
Interview tip: read the question, pause one second, then answer in three sentences max. Long answers are a "tell" of low confidence.
The shortcut to senior-sounding answers.