Kubernetes
35 questionsKubernetes architecture, Spark and Airflow on K8s, EKS, Helm charts, autoscaling, and platform engineering for data teams.
Pod Not Scheduling
kubectl describe shows your pod is Pending. Walk me through every possible cause and how you'd fix each one.
CrashLoopBackOff Investigation
A pod is stuck in CrashLoopBackOff. Walk me through your investigation step by step.
OOMKilled Debugging
Your Spark executor pod keeps getting OOMKilled. Debug it and fix it.
ImagePullBackOff
Pod shows ImagePullBackOff. What are all the reasons and how do you fix each?
Pod-to-Pod Communication Broken
Pod A can't reach Pod B's service. Walk through your debugging approach layer by layer.
Logs from Crashed Pod
Your pod crashed and restarted. How do you see what went wrong? What about centralized logging?
Node Drain & Maintenance
You need to patch a node running critical Kafka consumers. Walk through the process without losing messages.
Ingress & Port Debugging
Traffic isn't reaching your pod. The path is: browser → ingress → service → pod. Debug each hop.
Readiness Probe Hands-on
Create a Deployment with a readiness probe that checks if /tmp/ready exists. The pod shouldn't get traffic until the file is created.
Resource Requests vs Limits
What's the difference between requests and limits? What are QoS classes? What happens when a container exceeds its memory limit vs CPU limit?
Liveness vs Readiness vs Startup Probes
When do you use each probe type? What happens if readiness probe fails vs liveness probe fails?
ConfigMaps & Secrets in Practice
How do you pass database credentials to a Spark job on K8s without hardcoding? Show the YAML.
Schedule Pod on Specific Node
You have GPU nodes and CPU-only nodes. How do you ensure Spark jobs only land on CPU nodes and ML training only on GPU nodes?
Services Deep-Dive
ClusterIP vs NodePort vs LoadBalancer vs Ingress. When do you use each? Draw the traffic flow.
Spark on Kubernetes
Your team runs Spark on EMR and wants to move to EKS. spark-submit native K8s vs spark-operator — which and why? How does dynamic allocation work on K8s?
Airflow on Kubernetes
KubernetesExecutor vs CeleryExecutor vs KubernetesPodOperator. When to use each?
K8s Jobs & CronJobs for Batch
Design a nightly batch ETL on K8s using CronJobs. How do you handle failures, retries, and monitoring?
Init Containers
Your Spark job needs to download a 2GB ML model from S3 before processing starts. Design this with init containers.
Sidecar Pattern
Add structured logging and metrics to every data pipeline pod without changing application code.
HPA Autoscaling
Set up HPA for a streaming consumer. Can you autoscale Spark executors? Should you?
Pod Disruption Budgets
It's 2AM, K8s wants to drain a node running your Kafka consumer with PDB minAvailable=2 and you have 3 replicas. Walk through what happens.
Monitoring K8s Cluster
How do you monitor a K8s cluster running data workloads? What metrics matter? Prometheus + Grafana setup.
Resource Quotas & LimitRanges
One team's runaway Spark job consumed all cluster memory and crashed everyone else's pods. Prevent this.
Rolling Updates & Rollbacks
Deploy a new version of your streaming consumer with zero message loss. Walk through the strategy.
RBAC Design
Your data team and ML team share a cluster. Design the access model so they can't mess with each other's workloads.
Network Policies
Lock down a namespace so only Airflow can talk to the metadata database pod.
Secrets Management Reality
K8s Secrets are just base64-encoded, not encrypted. What do you actually use in production?
Pod Security
What security contexts do you set on production pods? Why?
EKS Architecture
Explain EKS architecture. Managed node groups vs self-managed vs Fargate. When to use each for data workloads?
IRSA (IAM Roles for Service Accounts)
How do your pods access S3, Glue, Redshift without baking AWS credentials into containers?
EKS Networking Deep-Dive
VPC CNI plugin gives each pod a real VPC IP. What are the implications? What about security groups for pods?
EKS vs EMR for Spark
When should you run Spark on EKS vs EMR? What are the real tradeoffs?
Cost Optimization
Your EKS cluster costs $50K/month for data workloads. Cut it by 40%.
GitOps for Data Pipelines
Implement GitOps for 20+ data pipelines on EKS. ArgoCD or Flux? How does it change deployment?
What Happens When You Kill a Pod?
You run kubectl delete pod on a pod managed by a Deployment. What happens internally? What's the impact on service traffic?