Cloud / Domain Knowledge Oral Round (GCP / AWS / K8s)

Requirements

• The session is spoken-only with no programming tasks, though the panel might draw diagrams on the common board.
• The subjects pressed depend on the group:
  • Core K8s concepts: comparing pods, deployments, and services; how scheduling works; scaling mechanisms (HPA, VPA, Cluster Autoscaler); network layers (CNI, services, ingress); persistent storage (PV, PVC, StorageClass); and debugging via kubectl describe, event logs, typical container startup failures.
  • Cloud platform: For GCP-focused loops, GCP services get emphasis; those with AWS history get AWS-focused queries. Compute instances (EC2 / GCE), object stores (S3 / GCS), identity access (IAM), private networks (VPC), packet filters (security groups / firewall rules), traffic distributors, and hosted DBs. They want practical setup steps ('how would you really configure Y'), not theory alone.
  • CI/CD and pipelines: delivery workflows, side-by-side vs incremental rollouts, code-defined infrastructure (Terraform, Pulumi), and Git-driven operations (ArgoCD, Flux).
  • Cloud migration: asked only of those with such projects: which systems moved, the data transfer method, failures encountered, and verification approach.
  • Domain-specific architectures: example: building a data pipeline design for the GCP estate.
• Format: a streak of swift definitional questions for 15–20 minutes, then one or two expansive scenario or design asks ('construct a delivery pipeline for X', 'root-cause this event').

Notes

• K8s replies that impress:
  • Container lifecycle stages (Pending → ContainerCreating → Running → Succeeded/Failed/CrashLoopBackoff). Typical breakages at each phase.
  • How Deployment, StatefulSet, and DaemonSet differ, plus the fitting use case for each.
  • Exposure methods (ClusterIP, NodePort, LoadBalancer, ExternalName, headless service).
  • Placement: resource requests versus caps, taints and tolerations, node affinity, pod affinity and anti-affinity.
  • Scaling: HPA (horizontal per pod), VPA (vertical per pod), CA (node quantity). Understand the metric each tracks.
• Cloud replies: grasp distinction among region, zone, AZ; bucket consistency (S3 offers strong read-after-write from 2020); managed database failover behavior; IAM role chaining. On the Salesforce GCP loop, panelists directly test the cloud named on your CV—never fake familiarity with a platform you haven't run.
• Migration talk: cite exact datasets, transition tactic (dual-write, shadow read, hard cut), observability added mid-move, and what was reverted. Vague claims like 'we migrated to cloud' without detail are the pitfall.
• Pipeline design: mirrors a standard software design loop (intake → transform → delivery), yet the asker wants named cloud products.
• They may test across providers: someone with AWS background faces AWS queries graded on true depth, not GCP knowledge. Candor about your cloud work beats making things up.

Preparation

• Brush up core K8s: CKAD-like cards covering objects, controllers, networking, storage, debugging.
• For every cloud on your CV, prepare a complete walkthrough of one system you engineered fully—components, size, improvements you'd make.
• If you listed migration work, rehearse a five-minute narrative with schedule, obstacles, revert strategy, and checks.
• Review an infrastructure-as-code tool (Terraform) sufficiently: state mechanics, plan versus apply, drift correction, multi-env workspace handling.
• For the pipeline design sub-question, keep a Kafka / Pub-Sub → Flink / Dataflow → BigQuery / Snowflake blueprint on hand.