Preemptible and Spot VMs in GCP - Step-by-Step Execution

Process Flow - Preemptible and Spot VMs

User requests VM

↓

Check VM type

↓

Preemptible?

↓

Create VM

↓

Run VM

↓

Google may stop VM anytime

↓

VM stopped or deleted

↓

User can restart or create new VM

When you request a VM as preemptible or spot, Google creates it but can stop it anytime to save resources. You can restart or create new VMs as needed.

Execution Sample

GCP

gcloud compute instances create example-vm \
  --zone=us-central1-a \
  --preemptible

This command creates a preemptible VM that runs until Google stops it, usually within 24 hours.

Process Table

Step	Action	VM State	Google's Decision	User Impact
1	User requests preemptible VM	VM not created	Check resource availability	Waiting for VM creation
2	Google creates VM	VM running	VM runs normally	User can use VM
3	Google needs resources	VM running	Decides to preempt VM	VM will be stopped soon
4	Google stops VM	VM stopped	VM preempted	User loses VM, must restart or create new
5	User restarts or creates new VM	VM running	VM runs normally	User continues work
6	VM runs up to 24 hours	VM running	VM may be preempted anytime	User should save work often
7	VM reaches max runtime	VM stopped	VM automatically stopped	User must restart or create new
8	End	VM stopped or running	No further action	User manages VM lifecycle

💡 VM stops when Google reclaims resources or max runtime reached; user must manage restarts.

Status Tracker

Variable	Start	After Step 2	After Step 4	After Step 5	After Step 7	Final
VM State	Not created	Running	Stopped	Running	Stopped	Stopped or Running
Google Decision	None	Run VM	Preempt VM	Run VM	Stop VM	No action
User Impact	Waiting	Using VM	VM lost	Using VM	Must restart	Manage VM

Key Moments - 3 Insights

Why does the VM stop even if I did not ask it to?

Google can stop preemptible or spot VMs anytime to free resources, as shown in execution_table step 4 where Google preempts the VM.

Can I keep a preemptible VM running forever?

No, preemptible VMs have a max runtime (usually 24 hours) after which Google stops them automatically (see step 7).

What should I do when my preemptible VM stops?

You need to restart or create a new VM to continue work, as shown in step 5 where the user restarts the VM.

Visual Quiz - 3 Questions

Look at the execution_table at step 4, what is the VM state?

A Running

B Stopped

C Not created

D Restarting

💡 Hint

Check the 'VM State' column in row for step 4 in execution_table.

At which step does the user restart or create a new VM?

A Step 3

B Step 7

C Step 5

D Step 2

💡 Hint

Look for 'User restarts or creates new VM' in the 'Action' column.

If Google never needed resources, what would happen to the VM state in the table?

A VM would keep running until max runtime

B VM would stop at step 4

C VM would never be created

D VM would restart automatically

💡 Hint

Refer to step 6 and 7 about VM running and max runtime.

Concept Snapshot

Preemptible and Spot VMs are cheaper VMs that Google can stop anytime.
You create them like normal VMs but expect interruptions.
Google stops them to free resources or after max runtime (~24h).
You must save work often and be ready to restart or create new VMs.
Ideal for batch jobs or fault-tolerant workloads.

Full Transcript

When you create a preemptible or spot VM on Google Cloud, the VM runs normally but can be stopped by Google at any time to free resources. This stopping is called preemption. The VM can run up to about 24 hours before Google stops it automatically. When the VM stops, you lose it and must restart or create a new one to continue your work. This makes preemptible and spot VMs cheaper but less reliable. They are good for tasks that can handle interruptions. The execution table shows the VM state changes from creation, running, preemption, stopping, and restarting. The variable tracker shows how VM state, Google's decisions, and user impact change step by step. Key moments clarify why VMs stop unexpectedly and what users should do. The quiz tests understanding of VM states and lifecycle steps. Remember, always save your work often and plan for VM interruptions when using preemptible or spot VMs.

Practice

1. What is the main advantage of using Preemptible or Spot VMs in Google Cloud?

easy

A. They offer unlimited storage capacity

B. They provide guaranteed uptime for critical applications

C. They cost less but can be stopped at any time

D. They automatically scale without user input

Solution

1. Step 1: Understand the cost and availability trade-off

   Preemptible and Spot VMs are cheaper because Google can stop them anytime to reclaim resources.

2. Step 2: Identify the main benefit

   The main benefit is cost savings with the risk of interruption, not guaranteed uptime or unlimited storage.

3. Final Answer:

   They cost less but can be stopped at any time -> Option C

4. Quick Check:

   Cost savings with interruptions = D [OK]

Hint: Cheaper VMs can be stopped anytime, so cost saving is main benefit [OK]

Common Mistakes:

• Thinking they guarantee uptime
• Assuming they scale automatically
• Confusing with storage features

2. Which of the following is the correct way to specify a Spot VM in a Google Cloud VM creation command?

easy

A. gcloud compute instances create my-vm --spot

B. gcloud compute instances create my-vm --interruptible

C. gcloud compute instances create my-vm --preemptible

D. gcloud compute instances create my-vm --ephemeral

Solution

1. Step 1: Recall the flag for Spot VMs

   Spot VMs use the flag --spot in the gcloud command.

2. Step 2: Differentiate from Preemptible flag

   --preemptible is for older Preemptible VMs, --spot is the newer recommended option.

3. Final Answer:

   gcloud compute instances create my-vm --spot -> Option A

4. Quick Check:

   Spot VM flag = --spot [OK]

Hint: Spot VMs use --spot flag, not --preemptible [OK]

Common Mistakes:

• Using --preemptible for Spot VMs
• Confusing --interruptible as a flag
• Using unrelated flags like --ephemeral

3. Consider this snippet of a VM creation command:
gcloud compute instances create test-vm --zone=us-central1-a --spot --machine-type=e2-medium
What will happen if Google Cloud needs the resources back?

medium

A. The VM will automatically migrate to another zone

B. The VM will continue running without interruption

C. The VM will be stopped immediately without warning

D. The VM will receive a 30-second warning before stopping

Solution

1. Step 1: Understand Spot VM behavior on resource reclamation

   Spot VMs are interruptible but Google Cloud sends a 30-second warning before stopping them.

2. Step 2: Eliminate other options

   Immediate stop without warning is incorrect; automatic migration does not happen for Spot VMs; uninterrupted running contradicts the interruptible nature.

3. Final Answer:

   The VM will receive a 30-second warning before stopping -> Option D

4. Quick Check:

   Spot VMs get 30-second warning before stop = A [OK]

Hint: Spot VMs get 30-second warning before stopping [OK]

Common Mistakes:

• Assuming immediate stop without warning
• Thinking Spot VMs migrate automatically
• Believing Spot VMs never stop

4. You created a VM with the command:
gcloud compute instances create my-vm --preemptible --zone=us-east1-b
But you want to switch to Spot VM instead. What is the correct fix?

medium

A. Add --spot flag without removing --preemptible

B. Replace --preemptible with --spot in the command

C. Change the zone to a Spot-only zone

D. Use --interruptible flag instead of --preemptible

Solution

1. Step 1: Identify the correct flag for Spot VMs

   Spot VMs require the --spot flag, not --preemptible.

2. Step 2: Correct the command by replacing flags

   Remove --preemptible and add --spot to switch VM type.

3. Final Answer:

   Replace --preemptible with --spot in the command -> Option B

4. Quick Check:

   Switching VM type requires flag replacement = B [OK]

Hint: Remove --preemptible, add --spot to switch VM type [OK]

Common Mistakes:

• Adding --spot without removing --preemptible
• Changing zone expecting Spot-only zones
• Using non-existent --interruptible flag

5. You want to run a batch job that can tolerate interruptions and save costs. Which approach best uses Spot VMs to handle sudden stops and restarts?

hard

A. Design the job to save progress frequently and restart automatically on VM preemption

B. Use Spot VMs without any checkpointing or restart logic

C. Run the job on standard VMs to avoid interruptions

D. Use Spot VMs but disable automatic restarts

Solution

1. Step 1: Understand Spot VM interruption nature

   Spot VMs can stop anytime, so jobs must handle interruptions gracefully.

2. Step 2: Choose a strategy to handle interruptions

   Saving progress frequently and restarting automatically ensures job completion despite stops.

3. Step 3: Eliminate unsafe options

   Running without checkpointing risks data loss; standard VMs cost more; disabling restarts loses fault tolerance.

4. Final Answer:

   Design the job to save progress frequently and restart automatically on VM preemption -> Option A

5. Quick Check:

   Checkpointing + auto-restart = reliable Spot VM use [OK]

Hint: Checkpoint progress and auto-restart for Spot VM jobs [OK]

Common Mistakes:

• Ignoring checkpointing and restart logic
• Choosing standard VMs for cost savings
• Disabling automatic restarts on Spot VMs