Kubernetes for Application Engineers
Definition
Kubernetes (K8s) is the industry-standard container orchestration platform: it schedules containers across a cluster, keeps the declared number of replicas running, wires up networking and service discovery, and rolls out updates safely. For application engineers, it is the runtime your Docker images ship to in most mid-to-large companies.
Why it matters for AI jobs
AI companies run inference services, RAG backends, and GPU training jobs on Kubernetes; "deploy on K8s" appears in most senior backend and platform JDs. Interviewers rarely expect cluster administration — they expect you to package an app with Docker, write the manifests or Helm chart, understand resource requests and autoscaling, and debug a failing pod. That level is attainable in a few focused weeks.
Key concepts
- Pods, Deployments, ReplicaSets — the declarative model: describe desired state, the controller converges.
- Services & Ingress — stable virtual IPs, load balancing, and HTTP routing into the cluster.
- ConfigMaps & Secrets — configuration and credentials injected via env or files.
- Resource requests/limits — CPU/memory scheduling, OOMKills, and (for AI) GPU resources.
- Autoscaling — HPA on CPU/custom metrics; cluster autoscaler for nodes.
- Helm — templated, versioned application packaging.
- Observability & debugging —
kubectl logs/describe/exec, liveness/readiness probes, events.
Learning path
- Run a local cluster (kind or k3d), containerize one of your apps, and deploy it with raw YAML manifests.
- Expose it via a Service + Ingress with TLS; practice a rolling update and rollback.
- Convert the manifests to a Helm chart with values for dev/prod.
- Add liveness/readiness probes, resource limits, and an HPA; break the app deliberately and debug with kubectl.
- Deploy the same chart to a managed cloud cluster (EKS/GKE/AKS) to learn the cloud-specific parts.