Compute Engine Cost Control: Schedules, Spot, Resize

What stops billing when a VM stops

Google’s stop and start guide draws the line clearly. After you stop an instance, vCPU and memory are no longer charged, but you keep incurring charges for resources that remain attached to it: persistent disks and static IP addresses. Two cautions from the same page:

• Local SSD data. When you stop a VM with Local SSD disks you can choose to discard or preserve that data, but stopping from within the guest OS discards it. A schedule on a VM with Local SSD needs that decision made up front.
• Stop is not suspend. If you need in-memory state back, Google’s answer is suspend, not stop.

How ZopNight handles Compute Engine

GCE is on ZopNight’s start/stop list. Schedules are cron expressions in your timezone, with a weekly 24-hour grid preview and an estimate of active versus inactive hours before you save. Groups let a set of VMs follow one calendar with sequenced, delayed starts. ZopNight computes savings as the 24/7 cost minus the actual cost of the hours the VM ran. Execution requires a read + write connection.

The GCP catalog names these Compute Engine rules:

• Idle GCP VM Instance and GCP VM Non-Production Scheduling propose a recurring off-hours schedule priced from the VM’s measured idle window. Schedule recommendations only use the latest 21 days of metrics and need at least 18 days of history.
• GCP VM Oversized - Downsize to Smaller Predefined Machine Type can be applied from the recommendation through the Remediate workflow, which stops, resizes and restarts the VM, with an admin-approval gate for non-admin initiators.
• GCP Spot VM Opportunity and GCP Prod VM Without Committed Use Discount cover pricing model rather than size.
• Stopped Standalone GCE VM prices the disks and reserved IPs a long-stopped VM still pays for, which is exactly the residue described above.
• Unattached GCP Persistent Disk can earn automatic cleanup when the disk has provably been detached for at least 7 days.

A one-shot stop of a running VM is never run automatically: the safety gate demotes it to advisory because no API can see what the guest is doing, and points you to a recurring schedule instead.

Fact sheets: Compute Engine, persistent disks. Schedule notes: scheduling Compute Engine. Background: spot instances.