roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Prepare for GitOps interview questions on delivering to Kubernetes from Git with Argo CD and Flux. Covers reconciliation and drift, repo structure and promotion, Helm and Kustomize, and secrets.
96 of 96 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 96 new. 0 cut, outside your plan.
GitOps seldom gets a round to itself. It surfaces in DevOps, SRE and platform engineering loops, usually inside the Kubernetes or delivery conversation, and from middle level up as the deployment half of a design exercise: several environments, many clusters, and the path a merged change takes to production and back out. Backend candidates meet it when they describe how their own service ships.
Most of the probing happens where Git and the cluster disagree. Someone runs kubectl edit on a Deployment that Argo CD manages: what happens next, and what changes when self-heal is off? A HorizontalPodAutoscaler and the manifest in Git both claim replicas; which wins, and how do you stop them fighting? A file is deleted from the repository; does the resource go with it?
Explaining pull against push and walking through one sync is the entry bar. From senior level up the interviewer wants the layout argued: a directory per environment rather than a branch per environment, one image digest promoted from stage to stage, a rollback made with git revert rather than a button that leaves Git describing something the cluster no longer runs, and the risk of one Argo CD instance holding credentials to every cluster it deploys to.
Argo CD and Flux take the loop every Kubernetes controller already runs and move it one step outward, with a Git commit as the spec, so the plan begins below them: declarative configuration, then desired-state reconciliation as Kubernetes performs it, the whole idea in miniature. The API object conventions after them answer the drift questions early: field management decides which writer owns replicas, and owner references decide what disappears with a pruned parent.
Core Principles states the model as a pull-based loop and raises the design question it brings at once, how the repository is laid out for promotion. Manifest Packaging precedes the tools because both engines consume rendered YAML, and who renders matters: a chart rendered by a GitOps controller is not always installed as a Helm release, while Kustomize patches plain manifests without templates.
Only then come the engines. Argo CD runs from the Application and its sync behavior to the projects that fence one team's deployments from another's; Flux spreads the same work across separate controllers and adds image update automation, which writes a new tag back to Git instead of going around it.
The last two stretches turn from mechanism to release. Deployment strategies and artifact promotion ask what a sync means for users and why one built image should travel through every environment unchanged. Secrets & Guardrails ends on the two objections a GitOps design usually meets: credentials cannot sit in Git in plain text, which external secret integration answers with a reference in the repository and the value in a vault; and a pull-based deploy leaves no pipeline step to fail, so the gate moves to the merge and to admission in the cluster.
Coming from push-based CI, where a job ran `helm upgrade` or `kubectl apply` and turned red when it failed, the habit to unlearn is the deploy as a step. Even if Kubernetes is familiar, do not skip ahead: switch study order from by frequency to by topics so the questions follow the plan from its first module, where the apply and ownership rules behind most wrong answers about drift and pruning sit. When the posting names one engine, work its module fully and read the other for contrast, since the difference between them usually comes up either way. Close to the date, importance cut to must know does most of the trimming, and moving the low end of level up to your own band drops definitional questions you will not be asked. Senior and platform candidates should put the saved hours into repository layout, multi-cluster composition and the gate, where the design round is decided. The plan takes the cluster as given. Scheduling, Services and pod networking, storage and the workload controllers belong to the [Kubernetes roadmap](/roadmaps/kubernetes); writing charts, as opposed to rendering them, to the [Helm roadmap](/roadmaps/helm).
Coming from push-based CI, where a job ran helm upgrade or kubectl apply and turned red when it failed, the habit to unlearn is the deploy as a step. Even if Kubernetes is familiar, do not skip ahead: switch study order from by frequency to by topics so the questions follow the plan from its first module, where the apply and ownership rules behind most wrong answers about drift and pruning sit. When the posting names one engine, work its module fully and read the other for contrast, since the difference between them usually comes up either way.
Close to the date, importance cut to must know does most of the trimming, and moving the low end of level up to your own band drops definitional questions you will not be asked. Senior and platform candidates should put the saved hours into repository layout, multi-cluster composition and the gate, where the design round is decided.
The plan takes the cluster as given. Scheduling, Services and pod networking, storage and the workload controllers belong to the Kubernetes roadmap; writing charts, as opposed to rendering them, to the Helm roadmap.