Preemptible and Spot VMs in GCP - Deep Dive

Overview - Preemptible and Spot VMs

What is it?

Preemptible and Spot VMs are special types of virtual machines offered by cloud providers like Google Cloud Platform. They are cheaper than regular VMs because they can be stopped or taken away by the cloud provider at any time. These VMs are useful for tasks that can handle interruptions, like batch jobs or testing environments. They help save money while using cloud computing resources.

Why it matters

Without Preemptible and Spot VMs, running large or flexible workloads in the cloud would be more expensive, limiting access for many users and companies. These VMs allow users to use spare cloud capacity at a lower cost, making cloud computing more affordable and efficient. This helps businesses innovate faster and manage costs better.

Where it fits

Before learning about Preemptible and Spot VMs, you should understand basic cloud computing concepts like virtual machines and how cloud providers manage resources. After this, you can learn about workload management, autoscaling, and cost optimization strategies in cloud environments.

Mental Model

Core Idea

Preemptible and Spot VMs are temporary, low-cost cloud machines that can be taken away anytime, ideal for flexible, interruptible tasks.

Think of it like...

Imagine renting a bike from a bike-sharing service that offers cheaper rides but can ask you to return the bike anytime if they need it back. You save money but must be ready to stop your ride quickly.

┌───────────────────────────────┐
│        Cloud Provider          │
│  ┌───────────────┐            │
│  │ Regular VMs   │  Always on │
│  └───────────────┘            │
│  ┌───────────────┐            │
│  │ Preemptible   │  Can stop │
│  │ / Spot VMs    │  anytime  │
│  └───────────────┘            │
└─────────────┬─────────────────┘
              │
              ▼
  ┌───────────────────────────┐
  │ User Workloads            │
  │ - Flexible tasks          │
  │ - Can handle interruptions│
  └───────────────────────────┘

Build-Up - 6 Steps

1

Foundation

What is a Virtual Machine

Concept: Introduce the basic idea of a virtual machine as a computer inside a computer.

A virtual machine (VM) is like a computer that runs inside another computer. It has its own operating system and can run programs just like a real computer. Cloud providers offer VMs so users can run software without owning physical machines.

Result

You understand that VMs let you use computers remotely and flexibly.

Knowing what a VM is helps you grasp how cloud computing offers flexible computing power.

2

Foundation

Regular vs Special VMs

Concept: Explain the difference between regular VMs and special types like Preemptible or Spot VMs.

Regular VMs run continuously until you stop them and cost more. Preemptible or Spot VMs are cheaper but can be stopped by the cloud provider anytime to free resources. They are good for tasks that don’t need to run all the time.

Result

You see that not all VMs are the same and some trade reliability for cost savings.

Understanding this tradeoff is key to using cloud resources efficiently.

3

Intermediate

How Preemptible and Spot VMs Work

🤔Before reading on: do you think Preemptible VMs can run forever if you keep restarting them? Commit to your answer.

Concept: Learn the lifecycle and behavior of Preemptible and Spot VMs in the cloud.

Preemptible and Spot VMs run on spare cloud capacity. The cloud provider can stop them at any time, usually after 24 hours or sooner if needed. You pay less because you accept this risk. If stopped, your VM is deleted and you lose any data not saved externally.

Result

You know these VMs are temporary and must be used with this limitation in mind.

Knowing the temporary nature helps you design applications that can handle interruptions gracefully.

4

Intermediate

Use Cases for Preemptible and Spot VMs

🤔Before reading on: do you think Preemptible VMs are good for running a website that must always be online? Commit to your answer.

Concept: Identify when it makes sense to use these VMs based on workload needs.

Preemptible and Spot VMs are great for batch processing, data analysis, testing, and other tasks that can pause and resume. They are not suitable for critical services that need constant uptime, like customer-facing websites.

Result

You can match workload types to VM types for cost and reliability balance.

Understanding workload tolerance to interruptions guides smart cloud cost optimization.

5

Advanced

Handling VM Preemption Gracefully

🤔Before reading on: do you think your application will keep running smoothly if a Preemptible VM is stopped suddenly? Commit to your answer.

Concept: Learn strategies to manage VM interruptions without losing work or data.

To handle preemption, save work frequently to external storage, use checkpointing, or design your system to restart tasks automatically. Google Cloud sends a short warning before stopping the VM, so you can prepare to save state.

Result

Your applications can tolerate VM shutdowns without data loss or major downtime.

Knowing how to handle preemption turns a limitation into a manageable risk.

6

Expert

Cost and Capacity Tradeoffs in Production

🤔Before reading on: do you think using only Preemptible VMs always saves the most money in production? Commit to your answer.

Concept: Explore how to balance cost savings with reliability and capacity planning in real systems.

In production, mixing Preemptible and regular VMs can optimize cost and uptime. Overusing Preemptible VMs risks frequent interruptions, while using only regular VMs increases cost. Spot VM availability depends on cloud capacity, so plan fallback strategies.

Result

You can design cloud infrastructure that balances cost, performance, and reliability.

Understanding tradeoffs helps build resilient, cost-effective cloud systems.

Under the Hood

Preemptible and Spot VMs run on spare physical servers in the cloud data centers. The cloud provider monitors demand and reclaims these servers when needed for higher priority tasks. When reclaimed, the VM is stopped and its resources freed. The VM's local disk is ephemeral and lost on stop, so persistent storage must be external. A short preemption notice is sent to the VM to allow cleanup.

Why designed this way?

Cloud providers created Preemptible and Spot VMs to maximize resource utilization and offer cost savings. Instead of leaving servers idle, they rent them cheaply with the condition of interruption. This design balances supply and demand, reduces waste, and opens cloud access to cost-sensitive users. Alternatives like reserved instances lock resources but cost more.

┌─────────────────────────────┐
│ Cloud Data Center            │
│ ┌───────────────┐           │
│ │ Physical Host │           │
│ └───────────────┘           │
│       ▲                     │
│       │ Spare Capacity       │
│       ▼                     │
│ ┌───────────────┐           │
│ │ Preemptible   │◄──────────┤
│ │ / Spot VM     │  Preemption│
│ └───────────────┘  Notice   │
└─────────────┬───────────────┘
              │
              ▼
      ┌───────────────┐
      │ User Workload │
      └───────────────┘

Myth Busters - 4 Common Misconceptions

Quick: Do Preemptible VMs guarantee to run for 24 hours every time? Commit to yes or no.

Common Belief: Preemptible VMs always run for a full 24 hours before being stopped.

Reality: Preemptible VMs can be stopped anytime before 24 hours if the cloud provider needs the resources back.

Why it matters: Assuming a full 24-hour run can cause failures if your workload depends on uninterrupted execution.

Quick: Can you save data on the local disk of a Preemptible VM and expect it to persist after shutdown? Commit to yes or no.

Common Belief: Data stored on the VM's local disk will remain safe after the VM is preempted.

Reality: Local disk data is lost when a Preemptible or Spot VM is stopped; persistent storage must be external.

Why it matters: Losing unsaved data can cause job failures and data corruption.

Quick: Are Preemptible VMs suitable for running critical, always-on services? Commit to yes or no.

Common Belief: Preemptible VMs are fine for any workload, including critical services.

Reality: They are not suitable for critical services because they can be stopped without warning, causing downtime.

Why it matters: Using them for critical services risks unexpected outages and user dissatisfaction.

Quick: Do Spot VMs always cost less than regular VMs? Commit to yes or no.

Common Belief: Spot VMs are always cheaper than regular VMs regardless of demand.

Reality: Spot VM prices can fluctuate and sometimes approach regular VM prices depending on supply and demand.

Why it matters: Assuming constant low cost can lead to budget surprises.

Expert Zone

1

Preemption notices are typically 30 seconds, requiring fast and efficient shutdown logic in applications.

2

Spot VM availability varies by region and time, so multi-region strategies improve reliability.

3

Combining Preemptible VMs with managed instance groups allows automatic replacement and scaling.

When NOT to use

Avoid Preemptible and Spot VMs for workloads requiring guaranteed uptime, persistent local storage, or real-time responsiveness. Use regular VMs or dedicated instances instead.

Production Patterns

In production, teams use mixed VM fleets with autoscaling groups that include Preemptible VMs for cost savings and regular VMs for reliability. They implement checkpointing, external storage, and preemption handling to maintain service continuity.

Connections

Autoscaling

Builds-on

Understanding Preemptible VMs helps optimize autoscaling by adding low-cost capacity that can be reclaimed without impacting critical services.

Fault Tolerance

Same pattern

Preemptible VMs require fault-tolerant design, teaching principles that apply broadly to building resilient systems.

Airline Overbooking

Similar concept

Like airlines overbooking flights expecting some no-shows, cloud providers offer Preemptible VMs using spare capacity they might reclaim, balancing utilization and risk.

Common Pitfalls

#1 Running critical services on Preemptible VMs expecting no downtime.

Wrong approach:

Deploying a customer-facing website solely on Preemptible VMs without backups.

Correct approach:

Use regular VMs for critical services and Preemptible VMs only for non-critical batch jobs.

Root cause:

Misunderstanding the temporary nature and risk of Preemptible VMs.

#2 Storing important data only on the VM's local disk.

Wrong approach:

Saving database files on the local disk of a Preemptible VM.

Correct approach:

Use persistent cloud storage services like Cloud Storage or Persistent Disks for important data.

Root cause:

Not knowing that local disk data is lost when the VM is stopped.

#3 Ignoring preemption warnings and not saving state.

Wrong approach:

Running long computations without checkpointing or saving progress on Preemptible VMs.

Correct approach:

Implement checkpointing and save progress externally when receiving preemption notices.

Root cause:

Not handling the short warning period before VM shutdown.

Key Takeaways

Preemptible and Spot VMs offer significant cost savings by using spare cloud capacity that can be reclaimed anytime.

These VMs are temporary and can be stopped with little notice, so workloads must tolerate interruptions.

They are ideal for flexible, batch, or fault-tolerant tasks but not for critical, always-on services.

Handling preemption properly requires saving state externally and designing for quick shutdowns.

Combining Preemptible VMs with regular VMs and autoscaling enables cost-effective and reliable cloud infrastructure.