DevOps is the one discipline where the certifications and the day job overlap almost perfectly. A Kubernetes administrator who can pass the CKA can genuinely operate a cluster; an engineer who clears AZ-400 has, by necessity, designed real pipelines, branching strategies, and release gates. That is the good news. The awkward news is that “DevOps certification” is not one exam but a constellation — a foundation cert from the DevOps Institute, three rival cloud professional exams, three Kubernetes credentials, a Terraform exam, and a clutch of tool-vendor certs from GitHub and GitLab. Choosing the wrong order, or treating them as interchangeable, wastes months. Treated as a deliberate ladder, they compound: the Terraform Associate feeds every cloud DevOps exam, the cloud DevOps exam you sit feeds your interview story, and the Kubernetes certs are the differentiator that turns “I know CI/CD” into “I run the platform”.
This lesson is the capstone of the KloudVin DevOps Zero-to-Hero course: the consolidated, exam-by-exam prep kit. It is not a re-teach of DevOps from scratch — the rest of the course does that, from DevOps fundamentals through pipelines, deployment strategies, GitOps, and DevSecOps. It is the layer on top: for each credential, the objective-domain checklist with realistic weightings so your study time lands on the marks; a bank of practice scenario questions written in each exam’s idiom, with the reasoning for every right answer and why the tempting wrong ones are wrong; the topics candidates most reliably confuse across exams; and — the part most “cram guides” skip — a study order and plan that sequences the credentials so each one makes the next cheaper. The goal is to pass each on the first attempt and walk out genuinely more capable, not merely badged.
In a nutshell
If the certification names in the title read like alphabet soup — DOP-C02, AZ-400, CKA, CKAD, CKS, Terraform Associate — here is the plain-language map before any exam detail. A DevOps certification is simply a third-party exam that vouches you can do a slice of the job: build pipelines, run a Kubernetes cluster, write infrastructure-as-code, or operate reliably on one specific cloud. Treat them less like school diplomas and more like belts in a martial art — you earn them roughly in order, each one assumes the muscle you built for the last, and the belt matters far less than the training that earned it.
Here is what each one is, and who it’s for, in a line each:
- DevOps Foundation (DevOps Institute) — the vocabulary belt. Multiple choice, no prerequisite. For anyone who wants the shared language (CALMS, DORA, the Three Ways) before touching tools. The gentlest on-ramp.
- Terraform Associate (HashiCorp) — the infrastructure-as-code belt. Short, recall-style. For anyone who writes (or wants to write) Terraform; it underpins the IaC part of every cloud exam, so it’s the smart first technical cert.
- AWS DevOps Engineer – Professional (DOP-C02) — “I can automate and operate on AWS.” Scenario-heavy multiple choice; assumes AWS associate-level knowledge already.
- Azure DevOps Engineer Expert (AZ-400) — “I can run DevOps on Azure and GitHub.” Pipeline-heavy; the Expert title requires an associate cert (AZ-104 or AZ-204) alongside it.
- Google Professional Cloud DevOps Engineer — “I can run reliable services on GCP the SRE way.” Steeped in SLOs and error budgets.
- CKA / CKAD / CKS (Kubernetes, from the CNCF) — the hands-on belts. You sit at a real terminal and do the tasks: run the cluster (CKA), build workloads on it (CKAD), secure it (CKS, which needs a current CKA first). No multiple choice at all.
- GitHub Actions / GitLab CI/CD — the tool belts. Short exams proving you can author pipelines on that platform.
Which to pick, and in what order? Match the cert to the job you want, then sequence it so each rung makes the next cheaper:
- Targeting a cloud DevOps / platform role? Start with the Terraform Associate (cheap, foundational), then the one professional cloud exam for your primary cloud (DOP-C02, AZ-400, or GCP DevOps — pick one, don’t collect all three), then CKA if the role runs Kubernetes.
- Targeting a Kubernetes / platform-engineering role? Go CKAD or CKA first (build-apps vs run-cluster — whichever matches the job), then CKS once you hold a CKA.
- New to DevOps entirely? DevOps Foundation first for the language, then the Terraform Associate for your first genuinely hands-on cert.
The rest of this lesson gives you, for each exam, the objective-domain checklist with weightings, a bank of practice scenarios with explained answers, the topics people most reliably confuse, and a sequenced study plan — so you build a deliberate ladder, not a random pile of badges.
Level: Advanced (with a beginner-friendly on-ramp) · Time: ~40 min
Learning objectives
By the end of this lesson you will be able to:
- Lay out the full DevOps certification landscape — vendor-neutral, cloud-specific, Kubernetes, IaC, and tool certs — and judge which belong on your ladder.
- Recite the objective domains and weightings for the headline exams (DOP-C02, AZ-400, GCP DevOps, CKA/CKAD/CKS, Terraform Associate, DevOps Foundation) and prioritise study where the marks cluster.
- Answer scenario-style questions in each exam’s idiom and reason about why distractors are wrong, not just recognise the right answer.
- Separate the commonly-confused topics that cut across exams — blue/green vs canary, CKA vs CKAD vs CKS, AZ-400’s branch policies vs environments, DOP-C02’s CodeDeploy vs CodePipeline, declarative vs imperative kubectl.
- Distinguish the performance-based Kubernetes exams (hands-on terminal) from the multiple-choice/scenario cloud exams, and prepare for each format correctly.
- Sequence a credential path with a realistic timeline so each certification reinforces the next instead of starting cold.
Prerequisites
This is the final lesson of the KloudVin DevOps Zero-to-Hero course, so it assumes you have worked through (or already know) the practice it certifies: CI/CD pipeline design, deployment strategies, GitOps, observability and DORA, and DevSecOps. If terms like quality gate, canary, OIDC service connection, GitOps reconciliation, or SBOM are unfamiliar, study the matching course lesson first — every exam here rewards hands-on muscle memory over flashcards, and a fortnight of building beats a month of reading. You should also have sat, or be comfortable with, the HashiCorp Terraform Associate prep kit from the IaC course, because Terraform fluency underpins the IaC domain of every cloud DevOps exam. A working knowledge of Git, containers, and at least one cloud is the floor; everything above that this lesson helps you certify.
The DevOps certification landscape: what exists and what it’s for
Before choosing, see the whole board. DevOps certs fall into five families, and they answer different questions an employer or interviewer might ask.
| Family | Certifications | What it proves | Format |
|---|---|---|---|
| Vendor-neutral foundation | DevOps Institute: DevOps Foundation, DevOps Leader, DevSecOps Practitioner, SRE Foundation | You understand DevOps culture, practices, and vocabulary (CALMS, DORA, value-stream, three ways) | Multiple choice |
| Cloud DevOps (professional) | AWS Certified DevOps Engineer – Professional (DOP-C02), Microsoft Certified: DevOps Engineer Expert (AZ-400), Google Cloud Professional Cloud DevOps Engineer | You can design and run CI/CD, IaC, monitoring, and incident response on that cloud | Multiple choice / multiple response / case-style |
| Kubernetes (CNCF) | CKA (Administrator), CKAD (Application Developer), CKS (Security Specialist), plus KCNA/KCSA entry-level | You can operate / develop on / secure Kubernetes — hands-on | Performance-based (live terminal) |
| Infrastructure as Code | HashiCorp Certified: Terraform Associate (003), Terraform Authoring & Operations Professional | You can write and operate Terraform | Multiple choice / fill-in |
| Tooling | GitHub Actions, GitHub Foundations, GitHub Advanced Security; GitLab Certified CI/CD Associate / Associate / Professional Services Engineer | You can build pipelines and use the platform’s features | Multiple choice (GitHub) / multiple choice + hands-on (GitLab) |
Two structural facts drive the whole study strategy. First, the cloud DevOps exams assume you already hold the associate-level cert for that cloud — DOP-C02 expects SysOps/Developer Associate knowledge, AZ-400 requires an associate prerequisite (AZ-104 or the Developer AZ-204) to award the Expert title, and the GCP DevOps exam assumes Associate Cloud Engineer fluency. You do not bolt a professional DevOps cert onto zero cloud knowledge. Second, the Kubernetes exams are a different sport: they are live, hands-on, kubectl-at-a-terminal exams with no multiple choice at all, so they need a fundamentally different prep style (drilling commands against a real cluster) than the recognition-and-recall cloud exams.
How to read the per-exam tables
For each exam below you get the objective domains with a weight band (a realistic “where the marks cluster” — published percentages where the vendor gives them, well-established bands where it doesn’t), the format and logistics, and the highest-leverage focus. Treat weightings as priorities for revision time, not gospel. The practice-question bank that follows is cross-cutting and ordered by exam.
DevOps Institute: DevOps Foundation & DevSecOps Practitioner
These vendor-neutral exams certify vocabulary and principles, not tooling. They are the easiest entry point and the right first rung if your background is light on DevOps theory, and their language (CALMS, the three ways, DORA, value-stream mapping) appears verbatim in interviews.
DevOps Foundation — objective areas and weight:
| Objective area | What it covers | Weight band |
|---|---|---|
| DevOps principles & the Three Ways | Flow, feedback, continual learning; systems thinking | High |
| Culture & CALMS | Culture, Automation, Lean, Measurement, Sharing | High |
| DevOps practices | CI/CD, IaC, ChatOps, Kanban, value-stream mapping | Medium |
| Business & technology frameworks | Agile, Lean, ITSM, safety culture, Theory of Constraints | Medium |
| Measurement & metrics | DORA’s four keys, the importance of feedback | Medium |
| People, teams & automation | Topologies, shared ownership, toolchains | Low |
Format: ~40 multiple-choice questions, ~60 minutes, ~65% pass mark, no formal prerequisite. Focus: the Three Ways (flow, feedback, continual learning) and CALMS are the spine — most questions trace back to one of them.
DevSecOps Practitioner layers security into that model: shifting security left, threat modelling, secure pipelines, SAST/DAST/SCA, supply-chain security, secrets management, and continuous compliance. It maps directly onto the DevSecOps pipeline lesson — if you have built a pipeline with SAST/DAST/SCA gates, you have seen most of the exam.
AWS Certified DevOps Engineer – Professional (DOP-C02)
The heavyweight of the AWS path: a professional-level, scenario-dense exam that assumes you already know AWS to associate depth and tests whether you can automate, deploy, monitor, and recover on it. AWS publishes the domain weightings, which makes prioritising easy.
| # | Domain | Weight | What it really tests |
|---|---|---|---|
| 1 | SDLC automation | 22% | CodePipeline/CodeBuild/CodeDeploy, CI/CD patterns, artifact management, testing in pipelines |
| 2 | Configuration management & IaC | 17% | CloudFormation (and Terraform/CDK), drift, reusable templates, Systems Manager |
| 3 | Resilient cloud solutions | 15% | Multi-AZ/Region, auto scaling, self-healing, deployment strategies (blue/green, canary) |
| 4 | Monitoring & logging | 15% | CloudWatch, X-Ray, EventBridge, log aggregation, metrics & alarms |
| 5 | Incident & event response | 14% | EventBridge rules, automated remediation, on-call, runbooks/SSM Automation |
| 6 | Security & compliance | 17% | IAM, secrets (Secrets Manager/SSM), Config rules, GuardDuty, automated compliance |
Format: ~75 questions (multiple choice + multiple response), 180 minutes, scaled score 750/1000 to pass. Prerequisite knowledge: SysOps Administrator or Developer Associate depth (no longer a hard requirement to sit, but assumed). Focus: SDLC automation (22%) is the biggest single domain — own the AWS Code* suite cold (especially the CodeDeploy deployment configurations: in-place vs blue/green, AllAtOnce/HalfAtATime/OneAtATime/Canary/Linear). The exam loves “choose the most operationally efficient / least operational overhead” — when two answers both work, pick the more managed, less custom one.
Microsoft Certified: DevOps Engineer Expert (AZ-400)
The Azure path’s capstone. Unusually, it is an Expert cert that requires a prerequisite associate — you must also hold AZ-104 (Administrator) or AZ-204 (Developer) for the Expert title to be awarded. AZ-400 itself is heavily weighted toward CI/CD and source control, and it is comfortably tool-aware of both Azure DevOps and GitHub (Microsoft owns both).
| Objective domain | Weight band | What it covers |
|---|---|---|
| Design & implement processes & communications | 10–15% | Traceability, work-item tracking, dashboards/metrics, stakeholder integration, ChatOps |
| Design & implement a source control strategy | 15–20% | Git branching (trunk-based, GitFlow), branch policies, pull requests, monorepo vs multi-repo, large files |
| Design & implement build & release pipelines | 40–45% | YAML multi-stage pipelines, agents/runners, templates, artifacts, environments, approvals & gates, deployment strategies, IaC (Bicep/ARM/Terraform), database deployment, secrets |
| Develop a security & compliance plan | 10–15% | Secrets (Key Vault), dependency/credential scanning, GitHub Advanced Security, governance/policy |
| Implement an instrumentation strategy | 5–10% | App Insights, Azure Monitor/Log Analytics, alerts, feedback loops |
Format: 40–60 questions, ~100 minutes, scaled 700/1000 to pass, often including a case study and drag-and-drop ordering items. Focus: pipelines dominate (~40%+) — be fluent in multi-stage YAML pipelines with environments and approvals, the difference between branch policies (enforced on the PR — required reviewers, build validation, linked work items) and environment approvals/gates (enforced at deploy time), and the self-hosted vs Microsoft-hosted agent trade-offs. Source-control strategy is the second pillar and easy marks if you know the branching models.
Google Cloud Professional Cloud DevOps Engineer
Google’s exam is distinctive: it is steeped in Site Reliability Engineering (SRE) — SLIs/SLOs/error budgets, toil reduction, blameless postmortems — far more than the AWS or Azure equivalents. If you can speak SRE fluently you are halfway there.
| Objective area | Weight band | What it covers |
|---|---|---|
| Bootstrapping & maintaining a DevOps culture (SRE) | High | SLIs/SLOs/error budgets, toil, blameless postmortems, reducing organisational friction |
| Building & implementing CI/CD pipelines | High | Cloud Build, Artifact Registry, Cloud Deploy, deployment strategies, GKE/Cloud Run targets |
| Applying SRE practices to services | High | Capacity, autoscaling, incident response, on-call, deployment safety |
| Implementing observability | Medium | Cloud Monitoring/Logging/Trace/Profiler, the four golden signals, dashboards & alerting |
| Optimising service performance & managing incidents | Medium | Debugging, root cause, mitigations, runbooks |
Format: ~50–60 multiple-choice/multiple-select questions, 120 minutes, no published pass percentage (≈70% typical). Focus: error budgets and SLO maths are the signature topic — know that an SLO is a target, an SLI is the measurement, and the error budget (1 − SLO) governs release pace (budget left → ship; budget spent → freeze and stabilise). The four golden signals (latency, traffic, errors, saturation) and Google’s specific tools (Cloud Build, Cloud Deploy, Artifact Registry, Cloud Monitoring) round out the marks.
Kubernetes: CKA, CKAD & CKS (the performance-based trio)
These CNCF exams are a different category entirely: fully hands-on, sat in a browser terminal against real clusters, with kubectl and the open Kubernetes documentation available. There is no multiple choice. Speed and command muscle memory decide passes, not recognition. All three are open-book (you may use kubernetes.io/docs), two hours each, ~66% to pass, with one free retake included.
| Exam | Audience | Signature domains | The line that separates it |
|---|---|---|---|
| CKA (Administrator) | Cluster operators | Cluster architecture/install/upgrade (kubeadm), etcd backup/restore, workloads & scheduling, services & networking, troubleshooting (~30%), storage, RBAC | Run and fix the cluster |
| CKAD (Application Developer) | App developers | Application design & build (multi-container pods, init/sidecar), deployment (rollouts/strategies), config (ConfigMaps/Secrets), observability (probes), services & networking | Build and configure workloads on a cluster |
| CKS (Security Specialist) | Security engineers | Cluster hardening, system hardening, minimise microservice vulnerabilities, supply-chain security, monitoring/logging/runtime (Falco), network policies, Pod Security | Secure the cluster (requires a current CKA to sit) |
CKA domain weightings (the published blueprint): Troubleshooting 30%, Cluster architecture/installation/configuration 25%, Workloads & scheduling 15%, Services & networking 20%, Storage 10%. Read that off: troubleshooting is the single biggest domain — a CKA who cannot diagnose a CrashLoopBackOff, a failed node, or a broken kubelet under time pressure will not pass.
The non-negotiable prep tactics for all three:
- Set up aliases and context immediately.
alias k=kubectl,export do="--dry-run=client -o yaml"so you can generate manifests fast:k run nginx --image=nginx $do > pod.yaml. - Generate, don’t hand-write, YAML. Use
kubectl create/runwith--dry-run=client -o yamlthen edit — typing manifests from scratch is too slow. - Master
kubectl explainfor field names you forget, and the docs search for the rest (the exam expects you to use the docs). - For CKA: drill
etcdctl snapshot save/restoreand akubeadm upgradeuntil automatic — these are guaranteed, high-mark, all-or-nothing tasks. - Always
kubectl config use-context <ctx>at the start of each question — every question may be on a different cluster, and solving the right task on the wrong cluster scores zero.
HashiCorp Terraform Associate (003)
Covered in full in its own prep kit, so here is the executive summary in the context of the ladder: it is the canonical IaC credential, it sits underneath every cloud DevOps exam (all three test IaC), and it is genuinely useful. The marks cluster in the mechanics — state (local vs remote, locking, plaintext secrets), the configuration constructs (count vs for_each, variables, functions, the dependency graph), and the outside-core-workflow commands (import, state mv/rm, -replace, -refresh-only). Format: ~57 questions, 1 hour, recognition/recall (no labs). Sit this first on the cloud-DevOps path — it pays off in the IaC domains of DOP-C02, AZ-400, and the GCP exam.
GitHub & GitLab tool certifications
The lightest-weight family, but increasingly asked for. GitHub Actions certifies pipeline authorship: workflow syntax, events/triggers, jobs/steps, runners (hosted vs self-hosted), reusable workflows and composite actions, secrets/variables and environments, the marketplace, and OIDC for keyless cloud auth (see the OIDC lesson). GitHub also offers Foundations (the platform basics) and Advanced Security (code/secret/dependency scanning). GitLab offers a tiered path — Certified CI/CD Associate (.gitlab-ci.yml, stages/jobs, runners, artifacts, environments, CI/CD variables) up through professional and services-engineer credentials, several including a hands-on component. These are short multiple-choice exams (GitHub) you can clear in a focused week if you already build pipelines daily.
Practice question bank (scenarios, with explained answers)
Written in each exam’s idiom and ordered by exam. Commit to an answer before reading the explanation, and pay as much attention to why the wrong answers are wrong as to the right one — that habit is what every one of these exams rewards.
Q1 — DevOps Foundation (culture)
A team measures deployment frequency and change failure rate but ignores lead time and MTTR. According to DORA, what is the risk?
- A. They are measuring too much and should drop two metrics.
- B. They have an incomplete picture — the four keys balance throughput (frequency, lead time) against stability (change failure rate, MTTR); ignoring lead time and MTTR hides whether speed is costing reliability.
- C. The four metrics are independent, so two are sufficient.
- D. DORA only requires deployment frequency.
Answer: B. DORA’s four keys come as two pairs — throughput (deployment frequency, lead time for changes) and stability (change failure rate, time to restore/MTTR). The whole point is the balance: optimising throughput while ignoring stability (or vice versa) produces a distorted, gameable picture. A inverts the principle — the metrics are deliberately minimal. C is wrong because the pairs are explicitly meant to be read together. D is simply false. The exam wants you to know the four keys and why there are four.
Q2 — AWS DOP-C02 (SDLC automation)
A web app on EC2 behind an ALB must deploy with zero downtime and the ability to instantly roll back to the previous version if errors spike. Which CodeDeploy configuration fits best?
- A. In-place deployment,
AllAtOnce. - B. Blue/green deployment, shifting traffic to a new fleet, with automatic rollback on CloudWatch alarm.
- C. In-place deployment,
OneAtATime. - D. Blue/green with
AllAtOnceand no alarms.
Answer: B. Blue/green stands up a new (green) fleet, shifts the ALB’s traffic to it, and lets you roll back instantly by shifting traffic back to the still-running blue fleet — exactly “zero downtime + instant rollback”. Wiring a CloudWatch alarm to trigger automatic rollback closes the loop. A takes the app down (AllAtOnce in place) — fails zero-downtime. C is zero-downtime-ish but in-place rollback means redeploying the old version (slow), not an instant switch. D is the trap: blue/green is right, but AllAtOnce with no alarm gives you no automatic rollback signal. The distinction the exam draws: blue/green = instant rollback via traffic shift; in-place = slower redeploy rollback.
Q3 — AWS DOP-C02 (incident response)
You need to automatically remediate any S3 bucket that becomes publicly readable, with the least operational overhead. What do you build?
- A. A nightly Lambda that scans all buckets.
- B. An EventBridge rule on the Config rule’s non-compliance event that triggers an SSM Automation document (or Lambda) to remediate.
- C. A CloudWatch dashboard for an engineer to watch.
- D. A GuardDuty finding emailed to the team.
Answer: B. The low-overhead, event-driven pattern is AWS Config (managed rule detects the public bucket) → EventBridge (fires on the non-compliance event) → SSM Automation/Lambda (remediates). It is real-time and fully automated. A works but is polling (a nightly scan), which is higher overhead and slower — the exam penalises polling when an event exists. C and D are detection/notification, not remediation — a human still has to act. The exam’s recurring lesson: prefer event-driven, managed, automatic over polling or manual.
Q4 — Azure AZ-400 (source control)
Your team must ensure every change to main is reviewed by two people, passes a build, and is linked to a work item, with no direct pushes. Where do you configure this?
- A. An environment approval check.
- B. A branch policy on
main(require minimum reviewers = 2, build validation, linked work items, no direct commits). - C. A YAML pipeline gate.
- D. A pull request template.
Answer: B. These are all properties of a branch policy in Azure Repos: minimum number of reviewers, build validation, check for linked work items, and blocking direct pushes to the protected branch — enforced on the pull request before merge. A is the trap that catches people: environment approvals/gates govern deployments at release time, not merges to a branch. C confuses a pipeline-level deploy gate with a repo policy. D is just boilerplate — a template does not enforce anything. AZ-400 leans hard on branch policy (merge-time) vs environment approval (deploy-time) — keep them separate.
Q5 — Azure AZ-400 (pipelines)
You want the same deployment steps reused across ten microservice pipelines, parameterised per service, with one place to update. What Azure Pipelines feature?
- A. Copy the YAML into each repo.
- B. A pipeline template (
extends/templatewith parameters) referenced by each pipeline. - C. A variable group.
- D. A separate classic release pipeline per service.
Answer: B. Templates (a shared YAML file with parameters, consumed via template: or extends:) are the DRY mechanism — one definition, parameterised per service, single point of update. A is the anti-pattern the question is steering you away from. C (variable groups) shares values, not steps — useful but doesn’t solve reuse of logic. D is the legacy classic approach AZ-400 expects you to move away from in favour of YAML-as-code. The exam rewards pipeline-as-code and templating over duplication.
Q6 — GCP DevOps (SRE)
A service has a 99.9% availability SLO over 30 days. Three weeks in, it has already consumed 90% of its error budget. What is the SRE-correct response?
- A. Raise the SLO to 99.95% to create more budget.
- B. Slow or freeze risky releases and prioritise reliability work until the budget recovers; resume normal release pace once it does.
- C. Ignore it — SLOs are advisory.
- D. Immediately page the entire team.
Answer: B. The error budget governs release pace: budget remaining means you can spend it on shipping features; budget nearly exhausted means you slow down, freeze risky changes, and invest in reliability until reliability recovers. That feedback loop is the core of Google’s SRE model and the heart of the exam. A is gaming the metric — changing the target to dodge the signal defeats the purpose. C misunderstands SLOs as advisory; the budget is meant to drive decisions. D is disproportionate — budget burn is a prioritisation signal, not necessarily an active incident. Know the chain: SLI (measure) → SLO (target) → error budget (1 − SLO, governs release decisions).
Q7 — CKA (etcd — performance-based, described)
You must take a snapshot of etcd and then restore the cluster from it. Which command takes the snapshot?
- A.
kubectl etcd snapshot save snapshot.db - B.
etcdctl snapshot save snapshot.db(with the CA cert, cert, and key flags, andETCDCTL_API=3) - C.
kubeadm etcd backup - D.
etcdctl backup --data-dir
Answer: B. etcd is backed up with etcdctl snapshot save <file>, supplying --endpoints, --cacert, --cert, and --key (and ensuring API v3). Restore is etcdctl snapshot restore <file> --data-dir <new-dir>, then point etcd’s static pod manifest at the restored data dir. A is not a real kubectl subcommand — etcd is managed with etcdctl, not kubectl. C is invented. D uses the deprecated v2 backup form, not the snapshot mechanism the exam tests. This is a near-guaranteed CKA task; the marks are all-or-nothing, so drill the exact flags until automatic.
Q8 — CKAD vs CKA (which exam — concept)
You need to certify that an engineer can build a multi-container pod with an init container and a sidecar, configure it via a ConfigMap, and add readiness/liveness probes — but not that they can upgrade a cluster or restore etcd. Which exam fits?
- A. CKA
- B. CKAD
- C. CKS
- D. KCNA
Answer: B. That description — designing/building and configuring workloads (multi-container pods, init/sidecar, ConfigMaps/Secrets, probes/observability) — is squarely CKAD (Application Developer). A (CKA) centres on operating the cluster (install/upgrade, etcd, nodes, troubleshooting) — explicitly excluded by “not upgrade or restore etcd”. C (CKS) is securing the cluster (and requires a current CKA). D (KCNA) is the entry-level multiple-choice associate, not a hands-on workload-building exam. The load-bearing distinction: CKA = run the cluster, CKAD = build workloads on it, CKS = secure it.
Q9 — CKS (security — concept)
You must block every pod in a namespace from running as root and from using privileged containers, using built-in Kubernetes mechanisms (no third-party admission controller). What do you apply?
- A. A NetworkPolicy.
- B. Pod Security Admission set to the
restrictedstandard on the namespace (labelpod-security.kubernetes.io/enforce: restricted). - C. An RBAC Role.
- D. A ResourceQuota.
Answer: B. Built-in Pod Security Admission enforces the Pod Security Standards (privileged/baseline/restricted); labelling a namespace pod-security.kubernetes.io/enforce: restricted blocks running as root, privileged containers, host namespaces, and more — no external controller needed (this replaced the removed PodSecurityPolicy). A controls network traffic, not pod privileges. C controls who can call the API, not what a pod may run. D caps resource consumption, not security context. CKS expects you to know PSA/PSS as the built-in replacement for the deprecated PodSecurityPolicy.
Q10 — Terraform Associate (state)
Three engineers share a Terraform config. Why is a remote backend strongly preferred over local state? (See the Terraform prep kit for the full set.)
- A. It makes
applyfaster. - B. It provides a shared source of truth and state locking, preventing concurrent writes from corrupting state.
- C. It is required to use modules.
- D. Local state can’t store outputs.
Answer: B. The defining reasons are a shared state and locking so two applies can’t corrupt the file. A is false (remote state can add latency). C is false (modules work with local state). D is false. The exam wants “team → remote + locking” without being tempted by plausible-but-wrong benefits. The Terraform Associate prep kit has a full bank of these.
Q11 — GitHub Actions (OIDC)
Your workflow deploys to AWS. To avoid long-lived secrets, how should it authenticate?
- A. Store an IAM access key and secret in repository secrets.
- B. Use OIDC: configure an IAM role trusting GitHub’s OIDC provider, and have the workflow request a short-lived token via
permissions: id-token: writeand the configure-credentials action. - C. Hard-code credentials in the workflow file.
- D. Use a personal access token.
Answer: B. GitHub Actions can present an OIDC token to the cloud; AWS (or Azure/GCP) trusts GitHub’s OIDC provider via a role/federation, and the job assumes the role for short-lived, keyless credentials — no static secrets to leak or rotate. The job needs permissions: id-token: write. A is the old, riskier pattern OIDC replaces. C is a serious anti-pattern (and a likely “obviously wrong” distractor). D is for Git/API auth, not cloud deploys. See the keyless deploys lesson for the full mechanism.
Q12 — Cross-cutting (deployment strategy)
You want to release a new version to 5% of users, watch error and latency metrics, and automatically promote or roll back based on those metrics. Which strategy is this?
- A. Blue/green.
- B. Recreate.
- C. Canary (progressive delivery) with automated metric analysis.
- D. Rolling update.
Answer: C. Routing a small percentage to the new version, evaluating metrics, and automatically promoting or rolling back is canary / progressive delivery (e.g. Argo Rollouts or Flagger). A (blue/green) shifts all traffic at once after validating the green environment — not a 5% subset with metric-driven promotion. B (recreate) tears down old then starts new (downtime). D (rolling) gradually replaces instances but without the metric-gated percentage control of a canary. The exam’s distinction: canary = small % + metric analysis + auto promote/rollback; blue/green = full switch with instant rollback.
That is a representative cross-section spanning every exam family. If you can answer these and explain the distractors, you are tracking well across the board. For each exam you actually sit, do at least one full-length timed practice paper (and for the Kubernetes exams, time yourself against a real cluster — pacing is everything).
Commonly-confused topics (the marks candidates drop across exams)
A small set of distinctions cut across multiple DevOps exams. Internalise these and you reclaim marks on every paper.
CKA vs CKAD vs CKS
| CKA | CKAD | CKS | |
|---|---|---|---|
| Role | Cluster administrator | Application developer | Security specialist |
| Centre of gravity | Install/upgrade, etcd, nodes, troubleshooting, RBAC | Building/configuring workloads, probes, ConfigMaps | Hardening, PSA/PSS, network policy, supply chain, runtime (Falco) |
| Prerequisite | None | None | Current CKA required |
| Signature task | etcd backup/restore, kubeadm upgrade |
Multi-container pod, rollout strategy | Restrict a pod, network policy, image scanning |
Mnemonic: Administer, build Deployments (apps), Secure. You must hold a current CKA to sit CKS — a frequently missed prerequisite.
Blue/green vs canary vs rolling
| Blue/green | Canary | Rolling | |
|---|---|---|---|
| Traffic | 100% switch after green is verified | Small % first, increase gradually | Replace instances batch by batch |
| Rollback | Instant (switch back to blue) | Shift % back / abort | Re-roll previous version (slower) |
| Cost | Two full environments briefly | One extra version, partial | No extra environment |
| Metric-gated promotion | No (manual cutover) | Yes (the defining feature) | Not inherently |
Every cloud DevOps exam tests this. The two facts that decide questions: blue/green = instant rollback via traffic switch, canary = small percentage with metric-driven auto-promote/rollback. Don’t conflate them.
AWS Code* services (DOP-C02)
| Service | Does | Confused with |
|---|---|---|
| CodeCommit | Managed Git repos | (being retired in favour of third-party Git) |
| CodeBuild | Build & test (compile, unit tests, package) | CodeDeploy |
| CodeDeploy | Deploy to EC2/Lambda/ECS (in-place, blue/green) | CodeBuild |
| CodePipeline | Orchestrates the stages end-to-end | CodeDeploy |
| CodeArtifact | Artifact/package repository | S3 |
The classic trap: CodePipeline orchestrates; CodeDeploy does the deployment; CodeBuild builds. Know which one owns deployment configurations (CodeDeploy) and which one sequences the workflow (CodePipeline).
AZ-400: branch policy vs environment approval
- Branch policy — enforced on the pull request, before merge: required reviewers, build validation, linked work items, comment resolution, blocking direct pushes. It protects the source.
- Environment approval / check / gate — enforced at deploy time: pre-deployment approvals, business-hours gates, query-a-monitor gates, exclusive lock. It protects the target.
Expect at least one AZ-400 question that hinges on putting the control in the right place (merge vs deploy).
SLI vs SLO vs error budget (GCP)
- SLI — the measurement (e.g. % of requests under 300 ms).
- SLO — the target for that SLI (e.g. 99.9%).
- Error budget — 1 − SLO (e.g. 0.1%), the allowed unreliability that governs release pace: budget left → ship; budget spent → freeze and stabilise.
If you can recite this chain and what the budget drives, you own the GCP exam’s signature topic.
Declarative vs imperative kubectl (CKAD/CKA)
- Imperative —
kubectl run,kubectl create,kubectl expose,kubectl scale: fast, one-off, great under exam time pressure (especially with--dry-run=client -o yamlto generate a manifest). - Declarative —
kubectl apply -fagainst version-controlled manifests: the production-correct, idempotent, GitOps-friendly approach.
The exam-savvy move: use imperative commands with --dry-run=client -o yaml to generate YAML quickly, then apply it — best of both, and the fastest way to score the performance-based exams.
Read the map bottom-up: the vendor-neutral DevOps Foundation sits at the base, the Terraform Associate feeds the three cloud DevOps professional exams (DOP-C02, AZ-400, GCP DevOps), and the performance-based Kubernetes trio (CKA → CKS, with CKAD alongside) forms its own track — the prerequisite arrows decide what you can sit when (you may sit an exam only once you hold what its arrow points from), so trace your own path along them instead of chasing badges in a random order.
Going deeper: advanced exam strategy
Knowing the domains is table stakes; passing on the first attempt — and walking out genuinely more capable — comes down to how you prepare and how you sit the exam. This is the meta-layer most cram guides skip.
Build first, certify second. Every credential here is easier to earn from the far side of having done the work. Two weeks building a real multi-stage pipeline teach AZ-400/DOP-C02’s pipeline domains more durably than a month of videos; a week operating a real cluster teaches CKA better than any slide deck. Structure prep as a build-then-certify loop: build the thing a domain describes, use the practice bank to expose the gaps, then drill only the gaps. Courses and videos are reference material, not the plan.
The two exam species demand opposite prep. Sort every exam into one of two buckets and prepare accordingly:
- Recognition / recall (DOP-C02, AZ-400, GCP DevOps, Terraform Associate, GitHub/GitLab) — you choose the best answer. Prep is reading trade-offs, sitting timed practice papers, and learning to spot the qualifier in the stem. Flag-and-return is your friend.
- Performance-based (CKA, CKAD, CKS — and the hands-on Terraform Authoring & Operations Professional, a step above the recall-style Associate) — you do the task at a live terminal, graded on the result. Prep is muscle memory against a real cluster: aliases,
--dry-rungeneration, and the guaranteed tasks drilled to reflex. Reading aboutkubectlwill not pass these.
killer.sh — use the simulator that ships with the exam. CKA/CKAD/CKS registration includes two free sessions of killer.sh, the official simulator, each active for 36 hours. It is deliberately harder and longer than the real exam and covers more ground, so treat your killer.sh score as a pessimistic proxy — a 60% there often maps to a comfortable real-exam pass. The value is in the worked solutions: complete every question, then study the provided explanation until you can reproduce the solution cold. Save your two activations for the final week.
Read case studies and long stems backwards. AZ-400 and the cloud pro exams lean on case studies (one shared scenario feeding several linked questions) and long multi-paragraph stems. Read the final sentence first — the actual ask (“most cost-effective”, “least operational overhead”, “fastest recovery”, “highest availability”) is the decision key — then scan the stem only for the constraints that separate the options. Inside a case-study block, answer all of its questions before moving on; some formats won’t let you return to a case study once you leave it.
Pace by marks, not by clock alone. Budget a rough time per question and hold to it. On the recall exams, flag-and-return: bank every quick certain answer first, then spend the leftover time on the flagged hard ones. On the performance-based exams the tactic is sharper because each task shows its percentage weight — triage ruthlessly: bank the high-weight, low-effort tasks, skip the all-or-nothing hard ones until the end, and never sink twenty minutes into a 4% task while 7% tasks sit untouched.
In the K8s exams, the docs are open — exploit that. You may use kubernetes.io/docs during CKA/CKAD/CKS. Know the exact pages for the tasks you’ll hit (etcd backup/restore, kubeadm upgrade, NetworkPolicy, RBAC, Pod Security Admission) and practise searching the docs fast. Pair that with kubectl explain for field names you forget. The exam is designed assuming you use the docs — not memorising every field is fine; being slow to find it is not.
Know the pass bar and bank the certain marks. You do not need a perfect score. AWS scales to 750/1000, Azure to 700/1000, the CNCF exams sit around 66%, and Terraform is pass/fail on a recall paper. Because the bar is well below 100%, the winning strategy is to secure the marks you’re sure of and not over-invest in one stubborn domain — but also not to under-study the “boring” high-weight domains (monitoring, incident response) that quietly decide the result.
Predict readiness before you book. Use timed full-length practice papers (recall exams) and killer.sh / a real cluster under the clock (K8s) to forecast your score, and only book when the forecast clears the bar with margin. The CNCF exams include one free retake; the cloud vendors impose waiting periods and re-sit fees — so the simulator is cheaper than a failed attempt. Sit the exam at the same time of day as your practice to remove day-of surprises.
Practice challenges
Six fresh, exam-style scenarios that span the ladder, escalating from “which cert” to advanced hands-on judgement. They deliberately don’t repeat the practice-question bank above. Commit to an answer before opening the solution — and, as always, be able to say why the tempting wrong choice is wrong.
Challenge 1 — Plan the ladder (beginner). You’re a backend developer moving into a platform role at a mostly-AWS shop that runs its workloads on EKS. You have about six months. Which certifications, in which order, and why?
<details> <summary><strong>Model answer</strong></summary>
Terraform Associate → AWS DevOps Engineer Professional (DOP-C02) → CKA (add CKS later if the role turns security-heavy). Terraform first because it’s cheap and its IaC skills feed DOP-C02’s Configuration-management/IaC domain; DOP-C02 next because it proves the AWS platform-automation skills the role needs; CKA last of the three because EKS is Kubernetes and CKA proves you can run the cluster.
Why: sequence so each rung feeds the next and matches the shop’s real stack (AWS + EKS) — not a random pile of badges. </details>
Challenge 2 — The DORA claim (beginner–intermediate). A manager says: “We deploy 20 times a day, so we’re an elite DORA performer.” What’s missing from that claim?
<details> <summary><strong>Model answer</strong></summary>
Deployment frequency is one of four keys. Without lead time for changes, change failure rate, and time to restore (MTTR), you can’t claim elite — 20 deploys a day with a high change-failure-rate is fragile, not elite. DORA pairs throughput (frequency, lead time) with stability (change failure rate, MTTR) precisely so speed can’t be claimed in isolation.
Why: DORA is throughput and stability, balanced; a single throughput metric proves nothing on its own. </details>
Challenge 3 — Terraform state collision (intermediate). Two engineers share a Terraform config with local state. Both run terraform apply at overlapping times and the state ends up inconsistent. Which two features would have prevented this, and what’s the one-word reason the exam wants?
<details> <summary><strong>Model answer</strong></summary>
A remote backend (shared state) plus state locking (e.g. S3 with a DynamoDB lock table, or the locking a managed backend provides). The remote backend gives one shared source of truth; the lock stops a second apply from writing while the first holds the lock.
Why: concurrency — a lock serialises applies so two writers can’t corrupt the state file. </details>
Challenge 4 — AZ-400 secret at deploy time (intermediate). An AZ-400 scenario: a YAML pipeline must fetch a database password only for the production stage, and the secret must never appear in the repository or the build logs. What do you use?
<details> <summary><strong>Model answer</strong></summary>
Store the secret in Azure Key Vault and surface it through a variable group linked to Key Vault (or the AzureKeyVault task) scoped to the production environment, so it’s pulled at run time and masked in logs. Not a plaintext pipeline variable, not a committed .env, not a secret checked into source.
Why: secrets belong in a managed vault, retrieved at deploy time, scoped to the stage that needs them, and masked — never in code or logs. </details>
Challenge 5 — CKA node NotReady (advanced, performance-based). On a CKA task, a worker node shows NotReady and its pods are stuck. Give the exam-fast diagnosis order.
<details> <summary><strong>Model answer</strong></summary>
kubectl config use-context <ctx> first, then kubectl get nodes → kubectl describe node <node> (read the conditions/events) → SSH to the node → systemctl status kubelet → journalctl -u kubelet → check the kubelet config, its certs, and the container runtime (systemctl status containerd); fix the failing component, then confirm the node returns to Ready.
Why: a NotReady node is almost always a kubelet or container-runtime problem on that node — go node-side fast instead of burning time in kubectl alone.
</details>
Challenge 6 — Green canary, red SLO (advanced, cross-cutting). A GCP service is burning its error budget fast during a canary rollout that auto-promotes on metrics. The canary’s own metrics look healthy, yet the overall SLO is degrading. What’s the SRE-correct read and action?
<details> <summary><strong>Model answer</strong></summary>
The canary’s local metrics can look green while aggregate reliability degrades — often because the analysis metric isn’t the SLI the SLO actually tracks, or the canary is shifting load/latency onto the baseline. Halt promotion, freeze risky releases per the error-budget policy, and re-align the canary’s analysis metric with the SLO’s SLI before resuming.
Why: the error budget governs release pace regardless of a green-looking canary — measure what the SLO measures, and let the budget, not the dashboard, gate the rollout. </details>
Hands-on lab: build your study tracker and a free Kubernetes drill
You cannot practise the multiple-choice cloud exams in a lab (they are recognition/recall), but two things move the needle measurably and cost nothing: a structured study tracker so your prep is gap-driven, and a local Kubernetes cluster so the performance-based exams get the only practice that works — drilling commands against a real cluster.
Step 1 — a self-assessment tracker (any cloud exam). Create a simple table (a spreadsheet or a Markdown file) with one row per objective domain of the exam you’re targeting and columns for baseline score, target, and current. Populate the domains from the tables above. This turns “study DevOps” into “raise domain 5 from 60% to 85%”, which is the difference between drifting and passing.
Step 2 — a free local Kubernetes cluster (for CKA/CKAD/CKS). Install a local cluster with kind (Kubernetes-in-Docker) or minikube — both free, both run on your laptop:
# kind: spin up a throwaway cluster
kind create cluster --name cka-drill
# confirm and set up the exam aliases you'll rely on
kubectl cluster-info --context kind-cka-drill
alias k=kubectl
export do='--dry-run=client -o yaml'
Step 3 — drill the highest-yield CKA/CKAD patterns. Time yourself; the real exams are won on speed:
# Generate a pod manifest the fast (exam-correct) way, then edit + apply:
k run nginx --image=nginx $do > pod.yaml
# Create a deployment, scale it, expose it imperatively:
k create deployment web --image=nginx --replicas=3
k scale deployment web --replicas=5
k expose deployment web --port=80 --target-port=80
# A multi-container / probe edit (CKAD bread-and-butter): generate then add a readinessProbe
k create deployment api --image=nginx $do > api.yaml # then edit to add probes, apply
# Troubleshoot (CKA's biggest domain): inspect a broken pod
k get pods -A
k describe pod <name>
k logs <name> --previous
k get events --sort-by=.lastTimestamp
Step 4 — drill the CKA guaranteed tasks. Practise an etcd snapshot and a static-pod inspection until they are automatic (on a kind cluster the etcd pod lives in kube-system):
k -n kube-system get pod -l component=etcd
# Read the etcd flags you'd pass to etcdctl (endpoints, cacert, cert, key):
k -n kube-system describe pod -l component=etcd | grep -E 'listen-client|cert|key|trusted-ca'
Step 5 — Cleanup.
kind delete cluster --name cka-drill
Cost note: zero. kind/minikube run entirely on your machine — no cloud account, no spend. This is the only practice that prepares you for the performance-based Kubernetes exams, and it is free; keep a cluster around for the weeks before the exam and drill daily.
Common beginner mistakes
These are the misconceptions that sink first-timers — distinct from the symptom→cause→fix table below. Each is the wrong mental model, why it fails, and the model to hold instead.
- “I’ll watch the video course and I’m ready.” Passive consumption feels like progress, but every exam here rewards doing: the cloud exams test trade-off judgement you only build by building, and the Kubernetes exams are pure hands-on. Right model: build a real pipeline/cluster/Terraform project, then certify; treat videos as reference, not the plan.
- “I’ll collect DOP-C02, AZ-400, and GCP DevOps to be cloud-agnostic.” Three professional cloud certs is months of duplicated study, and interviewers hire for depth on their cloud, not breadth of logos. Right model: one professional cloud cert deep; get breadth from the vendor-neutral, Kubernetes, and Terraform certs instead.
- “CKA is multiple choice like the AWS exam.” CKA/CKAD/CKS are live-terminal, task-graded, and time-brutal; recognition-style prep fails them completely. Right model: drill commands against a real cluster daily — speed, aliases, and
--dry-runreflexes decide it. - “I’ll sit CKS first — it’s the impressive one.” CKS requires an active CKA just to register, and it assumes you can already run the cluster you’re securing. Right model: schedule CKA first; CKS builds on it.
- “I’ll skip the associate cert and jump to the professional.” DOP-C02 assumes SysOps/Developer depth, AZ-400 requires AZ-104/AZ-204 for the Expert title, and GCP DevOps assumes Associate Cloud Engineer fluency. Right model: earn (or genuinely know) the associate layer first — the pro exams don’t re-teach it.
- “Both answers work, so it’s a coin toss.” On the pro exams the stem asks for the most operationally efficient / least overhead / most managed option — two answers can both “work” while only one is right. Right model: find the qualifier in the stem and pick the managed, automated, event-driven choice.
- “I’ll read every long question top to bottom.” Linear reading of multi-paragraph stems burns the time you need. Right model: read the final ask first, then scan the stem for the constraints that decide between options.
- “I scored 55% on killer.sh — I’ll never pass.” The simulator is deliberately harder and longer than the real CKA/CKAD/CKS. Right model: treat it as a pessimistic stress-test, understand every solution it shows you, then judge readiness — a middling killer.sh score often maps to a real-exam pass.
- “Pacing just means going faster.” Rushing causes the two costliest exam errors: solving on the wrong cluster and misreading the stem. Right model: budget time per question/mark, flag-and-return, verify
use-contexton every K8s task, and bank the easy marks before the hard ones.
Common mistakes & troubleshooting
| Symptom (in study or on the day) | Cause | Fix |
|---|---|---|
| Studying a cloud DevOps exam with no cloud associate knowledge | Skipping the assumed prerequisite | Hold/learn SysOps-or-Developer (AWS), AZ-104/AZ-204 (Azure), ACE (GCP) first |
| Hand-typing YAML in a CKA/CKAD exam | Treating it like a written exam | Generate with --dry-run=client -o yaml, then edit — speed is the exam |
| Solving the right task on the wrong cluster (Kubernetes) | Forgetting to switch context | kubectl config use-context <ctx> at the start of every question |
| Confusing CodePipeline and CodeDeploy (DOP-C02) | Treating the Code* suite as one thing | Pipeline orchestrates; Deploy deploys; Build builds |
| Putting a merge control in an environment gate (AZ-400) | Conflating source and deploy controls | Branch policy = PR/merge; environment approval = deploy |
| Gaming the SLO to dodge a budget burn (GCP) | Misunderstanding error budgets | The budget drives release decisions; don’t change the target to hide the signal |
| Picking blue/green for a “5% + metrics” scenario | Conflating strategies | 5% + metric-gated auto-promote = canary, not blue/green |
| Sitting CKS without a current CKA | Missing the prerequisite | CKS requires an active CKA — schedule CKA first |
| Running out of time on any exam | No pacing practice | Time full-length practice papers; flag-and-return; for K8s, time against a real cluster |
| Memorising tools but not trade-offs | Surface-level prep | The pro exams test “most efficient / least overhead” — learn why one option wins |
Best practices for passing (and for the job)
- Build, then certify. Every exam here rewards hands-on experience: a real pipeline for AZ-400/DOP-C02, a real cluster for CKA/CKAD/CKS, a real Terraform project for the Associate. The portfolio you build in Real-World DevOps Portfolio Projects doubles as exam prep and interview evidence.
- Sequence the credentials (see the study order below) so each makes the next cheaper — don’t sit them in a random order.
- For multiple-choice exams, learn trade-offs, not trivia. The professional exams phrase questions as “most operationally efficient” / “least overhead”; when two answers work, the more managed/automated one usually wins.
- For performance-based exams, drill commands against a real cluster daily in the final fortnight. Set up your aliases and
--dry-runworkflow until they’re reflex, and practise the guaranteed tasks (etcd, upgrades) to automaticity. - Time-box every practice paper. Pacing removes day-of anxiety; flag-and-return is your friend on the recall exams, and ruthless task-triage (skip the hard ones, bank the easy ones first) is your friend on the K8s exams.
- Don’t neglect the “boring” domains. Monitoring/observability and incident response are 25–30% of the cloud exams combined and easy to under-study.
Security notes
Security is now a first-class domain in every DevOps exam, and several facts are both exam-testable and job-critical:
- Prefer keyless, short-lived credentials. OIDC federation (GitHub Actions/Azure DevOps/GitLab to the cloud) replaces long-lived static keys; this appears in DOP-C02, AZ-400, the GCP exam, and the GitHub Actions cert. See keyless deploys with OIDC.
- Secrets belong in a manager, never in code or logs. Secrets Manager/SSM (AWS), Key Vault (Azure), Secret Manager (GCP), and Kubernetes Secrets backed by encryption-at-rest (CKS) — all exams test that you don’t hard-code secrets and that you scope access tightly.
- CKS expects defence in depth. Pod Security Admission (the PodSecurityPolicy replacement), NetworkPolicies (default-deny), image scanning and supply-chain provenance, runtime detection (Falco), and least-privilege RBAC — know each layer.
- Terraform state holds secrets in plaintext. Generated passwords and connection strings land in state regardless of intent; use an encrypted, access-controlled remote backend (Terraform Associate + the IaC domains of every cloud exam).
- Shift security left, but gate at deploy too. SAST/DAST/SCA in the pipeline (DevSecOps Practitioner, AZ-400’s security domain, DOP-C02’s compliance domain) plus admission/policy controls at deploy — the DevSecOps lesson covers the full chain.
Interview & exam questions
A blend of likely exam items and the interview questions that probe the same knowledge — with concise answers.
- Name DORA’s four keys and why there are four. Deployment frequency and lead time (throughput) balanced against change failure rate and time-to-restore/MTTR (stability) — the pairing prevents optimising speed at the cost of reliability.
- Blue/green vs canary — when do you pick each? Blue/green for an instant, all-at-once cutover with instant rollback (two environments); canary for a gradual, metric-gated rollout with automatic promote/rollback (progressive delivery).
- In AWS, what’s the difference between CodePipeline, CodeBuild, and CodeDeploy? CodePipeline orchestrates the stages; CodeBuild builds/tests; CodeDeploy deploys (and owns the in-place vs blue/green deployment configurations).
- AZ-400: branch policy vs environment approval? Branch policy enforces controls on the pull request/merge (reviewers, build validation, work-item links); environment approvals/gates enforce controls at deploy time.
- What governs release pace in Google’s SRE model? The error budget (1 − SLO): budget remaining means you can ship; budget exhausted means freeze risky changes and prioritise reliability.
- CKA vs CKAD vs CKS? Administer the cluster (CKA), build/configure workloads on it (CKAD), secure it (CKS, which requires a current CKA). All three are hands-on/performance-based.
- Fastest way to author a manifest in a Kubernetes exam? An imperative
kubectl create/runwith--dry-run=client -o yamlto generate the YAML, then edit andapply— far faster than hand-writing. - How do you back up and restore etcd?
etcdctl snapshot save(with cacert/cert/key, API v3) to back up;etcdctl snapshot restore --data-dirthen repoint etcd’s static-pod manifest to restore. - How should a pipeline authenticate to the cloud without long-lived secrets? OIDC federation: the platform presents a short-lived OIDC token; the cloud trusts the provider and the job assumes a role — keyless, no static keys to rotate.
- In Kubernetes, how do you block privileged/root pods without a third-party admission controller? Pod Security Admission set to the
restrictedstandard on the namespace (the built-in replacement for the removed PodSecurityPolicy). - Why are cloud DevOps professional exams “most efficient” exams? Multiple answers often work; the exam wants the most operationally efficient / least-overhead / most-managed option — so learn trade-offs, not just feature existence.
- Where does the Terraform Associate fit in a DevOps credential path? Underneath the cloud DevOps exams — Terraform/IaC is a core domain in DOP-C02, AZ-400, and the GCP exam — so it’s the efficient first IaC credential to bank.
Quick check
- Which Kubernetes certification requires you to already hold a current CKA?
- A scenario routes 5% of traffic to a new version and auto-promotes based on metrics — which deployment strategy is it?
- In Google’s SRE model, what does the error budget govern?
- In AWS, which service orchestrates the CI/CD stages, and which one performs the deployment?
- What is the fastest exam-correct way to produce a Kubernetes manifest under time pressure?
Answers
- CKS (Certified Kubernetes Security Specialist) requires an active CKA to sit.
- Canary / progressive delivery — a small percentage with metric-gated automatic promotion or rollback (blue/green is a full switch, not a 5% subset).
- The release pace: budget remaining → ship features; budget exhausted → freeze risky changes and prioritise reliability work. (Error budget = 1 − SLO.)
- CodePipeline orchestrates the stages; CodeDeploy performs the deployment (CodeBuild builds/tests).
- An imperative
kubectl create/runwith--dry-run=client -o yamlto generate the YAML, then edit andkubectl applyit.
Exercise
Plan a deliberate credential ladder rather than chasing badges, then execute the first rung:
- Pick your ladder (day 1). Choose your primary cloud and write your sequence from the study-order table below — e.g. Terraform Associate → AZ-400 → CKA → CKS. Put target dates on each.
- Self-assess the first exam (day 2). Build the domain tracker from the Lab for your first target exam and sit a cold practice paper (or, for a Kubernetes exam, attempt a timed task set on a kind cluster). Record score per domain.
- Target the gaps (week 1–2). For every domain under ~80%, do the matching KloudVin course lesson and the relevant hands-on drill. For cloud exams, focus on the heaviest-weighted domains; for K8s, focus on troubleshooting and the guaranteed tasks.
- Drill the confusions (final week). Re-derive the “commonly-confused” tables from memory: CKA/CKAD/CKS, blue/green/canary/rolling, the AWS Code* suite, branch policy vs environment approval, SLI/SLO/error budget. If you can reproduce them, you own them.
- Dress rehearsal & book. Sit a second timed practice (same time of day as your booked slot). Aim for ≥85% (cloud) or finishing the task set with time to spare (K8s). Book the exam.
Deliverable: a written credential-ladder plan with dates, plus a filled-in score-by-domain tracker for your first exam from steps 2 and 5 — proof your prep is gap-driven, not vibes-driven.
Certification mapping
This whole course leads here. The table places each credential on the DevOps ladder and notes what it certifies and its prerequisite.
| Certification | Tier | Certifies | Prerequisite (assumed or required) |
|---|---|---|---|
| DevOps Institute — DevOps Foundation | Foundation | DevOps culture, practices, DORA, the Three Ways | None |
| DevOps Institute — DevSecOps Practitioner | Foundation+ | Shift-left security, secure pipelines, supply chain | DevOps Foundation helpful |
| HashiCorp Terraform Associate (003) | Associate (IaC) | Writing & operating Terraform | None (feeds every cloud exam) |
| GitHub Actions / GitLab CI/CD Associate | Tool | Pipeline authorship on that platform | Git fluency |
| AWS Certified DevOps Engineer – Professional (DOP-C02) | Professional | CI/CD, IaC, monitoring, incident response on AWS | SysOps/Developer Associate depth |
| Microsoft Certified: DevOps Engineer Expert (AZ-400) | Expert | DevOps on Azure (+GitHub) — pipelines, source control, security | AZ-104 or AZ-204 required for the Expert title |
| Google Cloud Professional Cloud DevOps Engineer | Professional | SRE, CI/CD, observability on GCP | Associate Cloud Engineer depth |
| CKA — Certified Kubernetes Administrator | Specialist (hands-on) | Operating a Kubernetes cluster | None |
| CKAD — Certified Kubernetes Application Developer | Specialist (hands-on) | Building/configuring workloads | None |
| CKS — Certified Kubernetes Security Specialist | Specialist (hands-on) | Securing Kubernetes | Current CKA required |
For the deep, single-exam treatment of the Terraform Associate (full objective checklist, practice bank, and cheat sheet), see the Terraform Associate prep kit in the IaC course.
Glossary
- Performance-based exam — a hands-on exam sat at a live terminal (CKA/CKAD/CKS), graded on tasks completed, not multiple choice.
- Objective domain — a published area the exam blueprint is organised around, often with a percentage weighting.
- Distractor — a deliberately plausible but incorrect option; reasoning about why it’s wrong is the skill these exams test.
- DORA four keys — deployment frequency, lead time for changes, change failure rate, time to restore (MTTR).
- CALMS — Culture, Automation, Lean, Measurement, Sharing — the DevOps culture model.
- SLI / SLO / error budget — the measurement, its target, and the allowed unreliability (1 − SLO) that governs release pace.
- Blue/green — release by standing up a parallel environment and switching all traffic, enabling instant rollback.
- Canary / progressive delivery — release to a small percentage, evaluate metrics, then auto-promote or roll back.
- CKA / CKAD / CKS — Kubernetes Administrator / Application Developer / Security Specialist certifications (CKS requires a current CKA).
- etcd snapshot — a point-in-time backup of the Kubernetes control-plane datastore (
etcdctl snapshot save/restore). - Pod Security Admission (PSA) — built-in enforcement of the Pod Security Standards, the replacement for PodSecurityPolicy.
- OIDC federation — keyless authentication where a pipeline presents a short-lived token a cloud trusts, replacing static keys.
- Branch policy — Azure DevOps controls enforced on a pull request before merge (reviewers, build validation, work-item links).
- Environment approval/gate — controls enforced at deploy time (approvals, business-hours gates, monitor queries).
- Scaled score — a normalised exam score (e.g. AWS 100–1000, pass 750; Azure 700 to pass) rather than a raw percentage, so questions of differing difficulty compare fairly.
- Cut score / pass mark — the minimum score needed to pass an exam.
- Blueprint / exam guide — the vendor’s published document listing the objective domains and their weightings; the definitive study map.
- Multiple response — a question that asks you to select N correct options (usually no partial credit), as opposed to single-answer multiple choice.
- Case study — a shared scenario that feeds several linked questions, common in AZ-400; answer all of a case study’s questions before leaving it.
- killer.sh — the official CKA/CKAD/CKS simulator bundled free with registration (two 36-hour sessions), deliberately harder and longer than the real exam.
- Proctoring — supervised exam delivery (online or at a test centre, e.g. Pearson VUE/PSI) with identity verification and a monitored environment.
- Retake policy — the vendor’s rules for re-sitting; the CNCF exams include one free retake, while cloud vendors impose waiting periods and fees.
- kubeadm — the tool for bootstrapping and upgrading a Kubernetes cluster; a near-guaranteed CKA task area.
- etcdctl — the CLI for etcd, used for
snapshot save/restorein the CKA backup-and-restore task. --dry-run=client -o yaml— thekubectlflag combination that generates a manifest without creating anything — the fast, exam-correct way to author YAML.- Imperative vs declarative (kubectl) — one-off commands (
run/create/expose) versus applying version-controlled manifests (apply -f); the exam trick is to generate with the former and apply with the latter. - The Three Ways — flow, feedback, and continual learning/experimentation — the principles underpinning DevOps (from The DevOps Handbook).
- Toil — in SRE, manual, repetitive, automatable operational work that scales with the service; reducing it is a GCP-exam theme.
- Four golden signals — latency, traffic, errors, and saturation — the core SRE monitoring signals.
- SBOM — Software Bill of Materials, an inventory of the components in a build; a supply-chain/DevSecOps and CKS concern.
- SAST / DAST / SCA — static analysis, dynamic analysis, and software-composition (dependency) analysis — the pipeline security scans a DevSecOps exam expects.
- Associate / Professional / Expert / Specialist — certification tier names; professional-, expert-, and specialist-level exams assume associate-level knowledge beneath them.
Next steps
This is the final lesson of the KloudVin DevOps Zero-to-Hero course — congratulations on reaching it. From here:
- Pick your credential ladder from the study-order table and run the one-exam sprint in the Exercise. Book the first exam while the material is fresh.
- If you want a portfolio to point at in interviews — and the best possible exam practice — build the projects in Real-World DevOps Portfolio Projects.
- For the deep, single-exam treatment of the IaC credential that underpins every cloud DevOps exam, work the Terraform Associate prep kit.
- To reinforce the security domain that now appears on every exam, revisit the DevSecOps pipeline lesson and keyless OIDC deploys; for the SRE/observability themes the GCP exam loves, see DORA metrics in the pipeline.