Stopped Compute Engine VMs whose disks and static IPs keep billing

Stopping a VM ends compute billing, not storage billing

Google’s instance lifecycle page lists when each charge applies. CPU is billed only in RUNNING and PENDING_STOP; memory adds SUSPENDING and SUSPENDED. Stopping moves a VM to TERMINATED, where neither applies. But “for attached resources like disks or external IP addresses, you incur charges while the resources exist, regardless of the compute instance state.”

A static external IP stays billable too. VPC network pricing treats an address associated with a VM as in use “whether the instance is running or stopped”, so it keeps the in-use hourly rate. A VM stopped months ago is a disk and address bill with no workload behind it.

Finding stopped VMs and what they still hold

gcloud compute instances list \
  --filter="status=TERMINATED" \
  --format="table(name,zone,machineType,lastStopTimestamp)"

For each candidate, gcloud compute instances describe VM_NAME --zone=ZONE shows its disks and their autoDelete setting. gcloud compute addresses list shows reserved addresses and whether each is in use.

Which stopped VMs qualify

1. The VM’s status is TERMINATED.
2. It is standalone. VMs that belong to a managed instance group or a GKE node pool are skipped, because their parent controls when they stop and start.
3. At least one attached persistent disk or reserved static IP has a real price.

There is no minimum stop duration, so check the stop time before acting on a VM stopped recently.

When a stopped VM produces no finding

If none of the VM’s attached disks or addresses can be priced, ZopNight raises nothing rather than a $0 card, and it never substitutes the compute rate of a machine that is not running. A VM that is running but barely used belongs to Idle GCP VM Instance instead.

Adding up disks and addresses

saving = sum of priced attached persistent disks + sum of priced reserved static IPs
cost after fix = 0

The breakdown shows persistent disk and static IP amounts separately, and they add up to the total.

Retiring a stopped VM cleanly

1. Check that it is not a batch worker stopped on purpose, and that nothing starts it on a schedule.
2. Snapshot any disk whose data may be needed again.
3. Delete the VM with gcloud compute instances delete VM_NAME --zone=ZONE. Disks with autoDelete set to false survive; delete them with gcloud compute disks delete.
4. Release reserved addresses with gcloud compute addresses delete ADDRESS_NAME --region=REGION.
5. If the VM should be running after all, start it with gcloud compute instances start VM_NAME --zone=ZONE.