Amazon EKS — Cost Optimization Practice Questions
Test your understanding of EKS cost management strategies.
Q1. What is typically the largest cost component in an EKS cluster?
- A. EKS control plane
- B. EC2 compute (nodes)
- C. NAT Gateway
- D. EBS storage
Show Answer
Correct Answer: B
| Option | Explanation |
|---|---|
| A | Incorrect — control plane is $73/month, usually <10% of total cost |
| B | ✅ Correct — EC2 compute typically accounts for 50-70% of total EKS costs |
| C | Incorrect — NAT Gateway can be significant but usually 10-25% |
| D | Incorrect — EBS storage is typically 5-15% of total cost |
Q2. What Spot allocation strategy is recommended for EKS workloads?
- A. lowest-price
- B. capacity-optimized
- C. diversified
- D. price-capacity-optimized
Show Answer
Correct Answer: B
| Option | Explanation |
|---|---|
| A | Incorrect — lowest-price may select pools with high interruption rates |
| B | ✅ Correct — capacity-optimized selects pools with the most available capacity, reducing interruptions |
| C | Incorrect — diversified distributes evenly but doesn’t optimize for capacity |
| D | Incorrect — price-capacity-optimized is valid for EC2 Fleet but capacity-optimized is the EKS recommendation |
Q3. How does Karpenter’s consolidation feature save money?
- A. It negotiates lower EC2 prices
- B. It replaces underutilized nodes with smaller, better-fitting instances
- C. It shuts down the cluster during off-hours
- D. It applies Reserved Instance discounts
Show Answer
Correct Answer: B
| Option | Explanation |
|---|---|
| A | Incorrect — Karpenter uses standard EC2 pricing |
| B | ✅ Correct — consolidation detects underutilized nodes, migrates pods, and replaces with right-sized nodes |
| C | Incorrect — Karpenter doesn’t manage cluster schedules |
| D | Incorrect — Karpenter provisions On-Demand or Spot instances at standard prices |
Q4. Which tool provides per-namespace cost allocation on EKS?
- A. AWS Cost Explorer
- B. CloudWatch
- C. Kubecost
- D. EKS Console
Show Answer
Correct Answer: C
| Option | Explanation |
|---|---|
| A | Incorrect — Cost Explorer shows AWS service costs, not Kubernetes namespace costs |
| B | Incorrect — CloudWatch shows metrics, not cost allocation |
| C | ✅ Correct — Kubecost allocates costs to namespaces, deployments, labels, and teams |
| D | Incorrect — EKS Console shows cluster info, not cost allocation |
Q5. What is the most effective way to reduce NAT Gateway costs?
- A. Use a smaller NAT Gateway
- B. Deploy VPC endpoints for AWS services
- C. Reduce the number of AZs
- D. Disable NAT Gateway and use public subnets
Show Answer
Correct Answer: B
| Option | Explanation |
|---|---|
| A | Incorrect — NAT Gateway doesn’t have size options; it scales automatically |
| B | ✅ Correct — VPC endpoints route traffic to AWS services directly, bypassing NAT Gateway data charges |
| C | Incorrect — fewer AZs reduces availability, not NAT costs significantly |
| D | Incorrect — public subnets for worker nodes is a security risk |
Q6. What resource request practice wastes the most compute?
- A. Setting CPU limits too high
- B. Not setting memory limits
- C. Setting CPU requests far above actual usage
- D. Using Burstable QoS class
Show Answer
Correct Answer: C
| Option | Explanation |
|---|---|
| A | Incorrect — high CPU limits don’t reserve idle capacity; requests do |
| B | Incorrect — missing memory limits don’t directly waste resources |
| C | ✅ Correct — over-provisioned CPU requests reserve capacity on nodes that goes unused, blocking other pods |
| D | Incorrect — Burstable QoS is fine if requests are set correctly |
Q7. What VPA mode should you use in production to get recommendations without risk?
- A. Auto
- B. Recreate
- C. Initial
- D. Off
Show Answer
Correct Answer: D
| Option | Explanation |
|---|---|
| A | Incorrect — Auto restarts pods to apply new requests, risky in production |
| B | Incorrect — Recreate is not a valid VPA mode |
| C | Incorrect — Initial only sets requests at pod creation, still modifies workloads |
| D | ✅ Correct — Off mode only generates recommendations; doesn’t modify pods |
Q8. When is Fargate cheaper than EC2 for EKS workloads?
- A. Always
- B. For large-scale steady-state workloads
- C. For spiky or intermittent workloads with few pods
- D. For GPU workloads
Show Answer
Correct Answer: C
| Option | Explanation |
|---|---|
| A | Incorrect — Fargate is more expensive per unit at scale |
| B | Incorrect — EC2 is cheaper for steady-state workloads due to shared node capacity |
| C | ✅ Correct — Fargate’s per-pod pricing wins when pods run intermittently; no idle node costs |
| D | Incorrect — Fargate doesn’t support GPU instances |
Q9. What tool visually shows EKS node utilization in real-time?
- A. kubectl top
- B. eks-node-viewer
- C. CloudWatch Dashboard
- D. Kubecost
Show Answer
Correct Answer: B
| Option | Explanation |
|---|---|
| A | Incorrect — kubectl top shows tabular data, not a visual terminal display |
| B | ✅ Correct — eks-node-viewer provides a real-time terminal visualization of node utilization |
| C | Incorrect — CloudWatch Dashboard works but requires setup and isn’t real-time terminal based |
| D | Incorrect — Kubecost provides cost views, not real-time node utilization visualization |
Q10. How much can you save with EKS Extended Support vs. staying on Standard Support?
- A. Extended Support is free
- B. Extended Support costs $0.60/hr extra — staying on Standard saves $438/month/cluster
- C. Extended Support costs 10% more
- D. No difference
Show Answer
Correct Answer: B
| Option | Explanation |
|---|---|
| A | Incorrect — Extended Support has a significant additional charge |
| B | ✅ Correct — Extended Support costs $0.60/hr ($438/month) vs $0.10/hr for Standard Support |
| C | Incorrect — it’s a fixed hourly rate, not a percentage increase |
| D | Incorrect — there is a 6× cost difference between Standard and Extended Support |
