Backend interview prep

Start

Reserved vs On-Demand vs Spot Instances Interview Questions (Top 25 Questions with Answers)

Master Reserved, On-Demand, and Spot Instance Interview Questions with production-ready explanations covering pricing models, Savings Plans, Reserved Instances, interruption handling, workload selection, cost optimization, Kubernetes integration, and enterprise purchasing strategies.

Module Navigation

Previous: Cost Optimization Basics QA | Parent: Cost Optimization Learning Path | Next: Storage Cost Optimization QA

Introduction

One of the biggest cloud cost optimization decisions is choosing the correct purchasing model.

Choosing the wrong pricing model can increase cloud costs by 30–80%, while choosing the right combination can significantly reduce compute expenses without sacrificing reliability.

The three primary purchasing models are:

  • On-Demand
  • Reserved (Reserved Instances / Savings Plans / Reserved Capacity)
  • Spot (Preemptible Capacity)

A mature cloud architecture usually combines all three.

                    Compute Workloads
                           │
        ┌──────────────────┼──────────────────┐
        ▼                  ▼                  ▼
    On-Demand         Reserved         Spot Instances
        │                  │                  │
        ▼                  ▼                  ▼
 Flexibility        Predictable Cost     Lowest Cost

This guide contains 25 production-focused interview questions covering pricing models, enterprise purchasing strategies, Kubernetes workloads, production architectures, and common interview scenarios.

Learning Roadmap

Cloud Pricing Models
        │
        ▼
On-Demand
        │
        ▼
Reserved Capacity
        │
        ▼
Savings Plans
        │
        ▼
Spot Instances
        │
        ▼
Mixed Purchasing Strategy
        │
        ▼
Enterprise Optimization

Purchasing Models

1. What are the major cloud compute purchasing models?

The three primary purchasing models are:

Model Best For
On-Demand Short-term or unpredictable workloads
Reserved Long-running predictable workloads
Spot Interruptible fault-tolerant workloads

Many cloud providers also offer:

  • Savings Plans
  • Reserved Capacity
  • Dedicated Hosts
  • Dedicated Instances

2. What are On-Demand instances?

On-Demand instances allow customers to pay only while the instance is running.

Characteristics:

  • No long-term commitment
  • Highest flexibility
  • No upfront payment
  • Easy scaling
  • Higher hourly cost

Example use cases:

  • Development
  • Testing
  • New applications
  • Temporary workloads
  • Unknown demand

3. What are Reserved Instances?

Reserved Instances provide discounted pricing in exchange for a long-term commitment.

Typical commitment periods:

  • 1 year
  • 3 years

Benefits:

  • Lower hourly cost
  • Predictable billing
  • Suitable for steady workloads

Trade-offs:

  • Less flexibility
  • Commitment required

4. What are Savings Plans?

Savings Plans provide pricing discounts based on a committed spend rather than committing to specific instance types.

Advantages:

  • More flexible than Reserved Instances
  • Automatically applies discounts
  • Supports changing instance families (depending on plan type)

Savings Plans are generally preferred for many modern AWS environments because they provide greater flexibility while still reducing cost.

5. What are Spot Instances?

Spot Instances use unused cloud capacity that providers can reclaim when needed.

Benefits:

  • Lowest cost
  • Large discounts
  • Excellent for fault-tolerant workloads

Trade-off:

Instances can be interrupted with short notice.

Pricing Comparison

6. Compare On-Demand, Reserved, and Spot.

Feature On-Demand Reserved Spot
Cost Highest Medium Lowest
Commitment None Required None
Flexibility Highest Medium High
Interruption Risk None None High
Best for Production Yes Yes Selected workloads
Predictable Cost Medium High Low

7. Which purchasing model is cheapest?

Typical ranking:

Spot
   │
Reserved / Savings Plan
   │
On-Demand

Approximate relationship:

Spot
↓ 60–90%

Reserved
↓ 20–70%

On-Demand
Baseline Cost

Actual savings depend on provider, region, workload, and commitment.

8. Why is On-Demand more expensive?

Cloud providers reserve capacity for immediate availability.

Benefits include:

  • No commitment
  • Immediate provisioning
  • Elastic scaling
  • No reservation planning

The higher price reflects this flexibility.

9. Why are Reserved resources cheaper?

Providers receive predictable long-term usage commitments.

Benefits for providers:

  • Better capacity planning
  • Predictable revenue
  • Lower utilization risk

Providers share some of these savings with customers through discounted pricing.

10. Why are Spot Instances so inexpensive?

Spot capacity represents unused infrastructure.

Instead of leaving hardware idle, cloud providers offer it at significantly reduced prices.

However, capacity may be reclaimed whenever required.

Workload Selection

11. Which workloads should use On-Demand instances?

Good candidates:

  • Development
  • QA
  • New applications
  • Temporary environments
  • Unpredictable traffic
  • Migration projects
  • Short-term batch jobs

Avoid long-running stable workloads where discounts are available.

12. Which workloads should use Reserved capacity?

Best suited for:

  • Production APIs
  • Databases
  • Authentication services
  • Long-running backend services
  • Kubernetes baseline nodes
  • Monitoring systems
  • Core infrastructure

These workloads usually run continuously.

13. Which workloads should use Spot Instances?

Ideal workloads include:

  • Batch processing
  • Machine learning training
  • Video rendering
  • CI/CD workers
  • Big data processing
  • Kubernetes worker nodes
  • Stateless microservices
  • Queue consumers

The application should tolerate interruptions.

14. Which workloads should never rely only on Spot?

Avoid depending exclusively on Spot for:

  • Primary databases
  • Authentication services
  • Payment processing
  • Stateful applications
  • Critical control planes
  • Low-latency services requiring guaranteed capacity

Spot may still be used alongside On-Demand or Reserved capacity.

15. How do you decide which pricing model to use?

Evaluate:

  • Business criticality
  • Runtime duration
  • Predictability
  • Fault tolerance
  • Scaling behavior
  • Interruption tolerance
  • Availability requirements
  • Recovery time
  • Budget constraints

Decision flow:

Is workload temporary?

     │
 ┌───┴────┐
 ▼        ▼
Yes      No
 │        │
 ▼        ▼
On-Demand
          │
     Predictable?
          │
    ┌─────┴─────┐
    ▼           ▼
   Yes         No
    │           │
    ▼           ▼
Reserved   Can tolerate interruption?
                    │
             ┌──────┴──────┐
             ▼             ▼
            Yes           No
             │             │
             ▼             ▼
           Spot      On-Demand

Production Strategies

16. What is a blended purchasing strategy?

Enterprise environments rarely use a single pricing model.

Example:

Production Cluster

70% Reserved
20% Spot
10% On-Demand

Benefits:

  • Lower cost
  • High availability
  • Flexible scaling
  • Reduced commitment risk

17. How should Kubernetes clusters use Spot nodes?

A common design:

Cluster

Reserved Nodes
      │
Critical Workloads

Spot Nodes
      │
Stateless Applications
Batch Jobs
Background Workers

Pods running on Spot nodes should:

  • Be stateless
  • Restart safely
  • Handle interruption
  • Use multiple replicas

18. How do Spot interruptions affect Kubernetes?

When a Spot node is reclaimed:

Spot Interruption
        │
        ▼
Node Draining
        │
        ▼
Pods Evicted
        │
        ▼
Scheduler
        │
        ▼
New Node
        │
        ▼
Pods Restart

Applications should tolerate this process.

19. How can Spot interruption risk be reduced?

Use:

  • Multiple Availability Zones
  • Multiple instance families
  • Cluster Autoscaler
  • PodDisruptionBudgets
  • Multiple replicas
  • Graceful shutdown
  • Queue-based processing
  • Retry mechanisms

Do not depend on a single Spot instance type.

20. Should production applications use only Reserved instances?

Not necessarily.

Using only Reserved capacity may:

  • Increase commitment risk
  • Reduce flexibility
  • Increase unused capacity

Many organizations combine:

  • Reserved baseline
  • On-Demand bursts
  • Spot workers

Enterprise Purchasing

21. What are the risks of purchasing too many Reserved Instances?

Risks include:

  • Paying for unused capacity
  • Reduced flexibility
  • Architecture changes
  • Region migration
  • Instance family migration
  • Business downsizing

Reservation planning should be based on long-term stable demand.

22. What is Reserved Instance utilization?

Utilization measures how much purchased Reserved capacity is actually used.

Example:

Purchased Capacity: 100 vCPUs

Used Capacity: 75 vCPUs

Utilization = 75%

Low utilization means the organization is paying for unused committed capacity.

23. What is Reserved coverage?

Coverage measures how much eligible workload is protected by Reserved pricing.

Example:

Total Production Compute:
1000 vCPUs

Reserved Coverage:
700 vCPUs

Coverage:
70%

Organizations monitor both:

  • Coverage
  • Utilization

24. What are common purchasing mistakes?

Common mistakes include:

  • Reserving unstable workloads
  • Ignoring utilization reports
  • Buying excessive commitments
  • Running production entirely On-Demand
  • Using Spot for stateful databases
  • Ignoring interruption handling
  • Purchasing without historical metrics
  • Not reviewing commitment usage
  • Choosing incorrect regions
  • Forgetting future architecture changes

A mature strategy typically follows:

Critical Always-On Workloads
        │
        ▼
Reserved Capacity

──────────────

Variable Production Load
        │
        ▼
On-Demand

──────────────

Fault-Tolerant Workloads
        │
        ▼
Spot Instances

Example allocation:

Workload Purchasing Model
Authentication Reserved
Databases Reserved
Production APIs Reserved + On-Demand
Autoscaling Burst On-Demand
CI/CD Spot
Batch Processing Spot
ML Training Spot
Development On-Demand
Temporary Testing On-Demand

This combination balances:

  • Cost
  • Availability
  • Flexibility
  • Risk

Production Scenario

Example Enterprise Architecture

Internet Users
       │
       ▼
Load Balancer
       │
       ▼
Kubernetes Cluster
       │
 ┌─────┼──────────────┐
 ▼                    ▼
Reserved Nodes     Spot Nodes
 │                  │
Critical APIs      Batch Jobs
Databases          Analytics
Monitoring         CI/CD
Auth Service       Background Workers

Result:

  • Reliable production services
  • Low compute cost
  • Automatic scaling
  • Efficient resource utilization

Pricing Decision Matrix

Workload Characteristic Recommended Model
Temporary On-Demand
Predictable Reserved
Interruptible Spot
Critical Production Reserved
Unknown Growth On-Demand
Batch Processing Spot
Machine Learning Spot
Stateful Database Reserved
Development On-Demand
CI/CD Spot

Cost Comparison Diagram

Hourly Cost

On-Demand
██████████████████████

Reserved
██████████████

Spot
████

(Illustrative comparison; actual pricing varies by provider and workload.)

Enterprise Purchasing Strategy

             Production Workloads
                     │
        ┌────────────┼────────────┐
        ▼            ▼            ▼
Baseline Load   Variable Load   Batch Jobs
        │            │            │
        ▼            ▼            ▼
Reserved      On-Demand       Spot

Comparison Table

Feature On-Demand Reserved Spot
Commitment None 1–3 Years None
Cost High Medium Lowest
Availability Guaranteed Guaranteed Interruptible
Flexibility Highest Medium High
Predictable Billing Medium High Low
Production Critical Workloads Excellent Excellent Limited
Batch Processing Good Good Excellent
CI/CD Good Good Excellent

Best Practices Checklist

✓ Measure Historical Usage
✓ Understand Business Requirements
✓ Reserve Stable Production Capacity
✓ Use Spot for Fault-Tolerant Workloads
✓ Keep On-Demand for Bursting
✓ Monitor Reservation Utilization
✓ Monitor Reservation Coverage
✓ Diversify Spot Instance Types
✓ Use Multiple Availability Zones
✓ Configure Autoscaling
✓ Test Spot Interruptions
✓ Review Purchasing Every Quarter
✓ Track Cost KPIs
✓ Validate Performance After Optimization
✓ Balance Cost with Availability

Quick Revision

Topic Key Point
On-Demand Highest flexibility
Reserved Discounted long-term commitment
Savings Plan Flexible spending commitment
Spot Lowest cost, interruptible
Coverage Percentage of workload using discounted pricing
Utilization Percentage of purchased reservations actually used
Baseline Load Reserved
Burst Traffic On-Demand
Batch Jobs Spot
Kubernetes Spot Stateless worker nodes
Critical Databases Reserved
Enterprise Strategy Blend all three purchasing models

Interview Tips

During interviews:

  • Clearly explain the differences between On-Demand, Reserved, Savings Plans, and Spot.
  • Emphasize that enterprises rarely rely on a single purchasing model.
  • Recommend Reserved for predictable production workloads, On-Demand for unpredictable demand, and Spot for fault-tolerant processing.
  • Discuss interruption handling for Spot Instances, especially in Kubernetes.
  • Mention coverage and utilization as key FinOps metrics for committed discounts.
  • Support answers with architecture diagrams and real production examples rather than pricing percentages alone.

Summary

Selecting the correct compute purchasing model is one of the most effective cloud cost optimization techniques.

A successful enterprise strategy combines:

  • On-Demand for flexibility
  • Reserved Capacity or Savings Plans for predictable workloads
  • Spot Instances for fault-tolerant workloads

By continuously monitoring coverage, utilization, and workload behavior, organizations can significantly reduce cloud costs while maintaining performance and availability.

Mastering these 25 Reserved vs On-Demand vs Spot interview questions prepares you for AWS, Azure, Google Cloud, FinOps Engineer, DevOps Engineer, Platform Engineer, Cloud Architect, Technical Lead, and Solution Architect interviews.