Preemptible & Spot VMs

Preemptible and Spot VMs offer deep discounts (60-91% off on-demand prices) in exchange for the possibility that Google Cloud may terminate (preempt) the instance when it needs the capacity back. Understanding how they work, their limitations, and how to design fault-tolerant architectures around them is key to significant cost savings.

---

What Are Spot VMs?

Spot VMs are the current recommended type of discounted, preemptible compute on Google Cloud. They replaced the legacy "Preemptible VMs" and offer the same deep discounts with additional features.

Spot vs Preemptible (Legacy)

Feature	Spot VMs	Preemptible VMs (Legacy)
Discount	60-91% off on-demand	60-91% off on-demand
Max lifetime	No maximum (can run indefinitely if not preempted)	24-hour maximum
Availability	Based on spare capacity in the zone	Based on spare capacity in the zone
Preemption notice	30-second warning via ACPI signal	30-second warning via ACPI signal
Recommended	Yes — use for all new workloads	No — legacy, use Spot VMs instead

Important: Spot VMs have replaced Preemptible VMs. Google recommends using Spot VMs for all new workloads.

---

Creating Spot VMs

# Create a Spot VM
gcloud compute instances create batch-worker-1 \
  --zone=europe-west2-a \
  --machine-type=n2-standard-8 \
  --provisioning-model=SPOT \
  --instance-termination-action=STOP \
  --image-family=debian-12 \
  --image-project=debian-cloud

Termination Actions

Action	Behaviour
STOP	VM is stopped (can be restarted later if capacity is available)
DELETE	VM is deleted permanently

Creating Spot VMs in a MIG

# Instance template with Spot provisioning
gcloud compute instance-templates create spot-worker-template \
  --machine-type=n2-standard-8 \
  --provisioning-model=SPOT \
  --instance-termination-action=DELETE \
  --image-family=debian-12 \
  --image-project=debian-cloud \
  --metadata=startup-script='#!/bin/bash
# Pull work from queue and process it
python3 /opt/worker/process.py'

# MIG with Spot instances
gcloud compute instance-groups managed create spot-worker-mig \
  --zone=europe-west2-a \
  --template=spot-worker-template \
  --size=10

---

Handling Preemption

When Google needs the capacity back, your Spot VM receives a 30-second warning. You can detect this by polling the metadata server or listening for the ACPI G2 signal.

Detecting Preemption from Within the VM

# Poll the metadata server for preemption notice
curl -s -H "Metadata-Flavor: Google" \
  http://metadata.google.internal/computeMetadata/v1/instance/preempted
# Returns: TRUE if preemption is imminent

Shutdown Script for Graceful Termination

gcloud compute instances create batch-worker \
  --zone=europe-west2-a \
  --machine-type=n2-standard-4 \
  --provisioning-model=SPOT \
  --instance-termination-action=STOP \
  --image-family=debian-12 \
  --image-project=debian-cloud \
  --metadata=shutdown-script='#!/bin/bash
echo "Preemption detected — saving checkpoint..."
python3 /opt/worker/save_checkpoint.py
echo "Checkpoint saved."'

---

Pricing