An N2D machine type with AMD EPYC Milan CPU platform
AMD SEV Confidential Computing technology
Be created after January 15, 2024
All other Confidential VM types don't support live migration, and must set their
onHostMaintenance policy
to TERMINATE when being created. This means the VM stops during
host maintenance events.
Host maintenance events for Confidential VM instances that don't support live migration
During host maintenance events, Confidential VM instances that don't support live migration and have default host maintenance policy settings are stopped. They are restarted after the maintenance is complete.
The following table shows the default values for host maintenance policy settings on Confidential VM instances that don't support live migration.
| Host maintenance policy | Confidential VM default values | Description |
|---|---|---|
onHostMaintenance |
TERMINATE |
This property must be set to |
automaticRestart |
true |
When set to true, your Confidential VM instance restarts
after the maintenance is complete. |
hostErrorTimeoutSeconds |
330 |
The number of seconds between 90 and 330
before the host attempts to restart an unresponsive VM. |
With planning, you can minimize the impact of host maintenance events on your Confidential VM instances.
Minimize the impact of host maintenance events
To minimize the impact of host maintenance events on Confidential VM instances that don't support live migration, you can do the following things:
Monitor for maintenance event notices
To receive advance notice of host events, monitor the
/computeMetadata/v1/instance/maintenance-event metadata value. To do so,
request an event notice by running the following command on your Confidential VM
instance:
curl http://metadata.google.internal/computeMetadata/v1/instance/maintenance-event -H "Metadata-Flavor: Google"
If the request to the metadata server returns NONE, then the VM isn't
scheduled to stop. If the metadata server returns
TERMINATE_ON_HOST_MAINTENANCE, then your VM is scheduled for stopping.
The following table shows the host maintenance event notification period for Confidential VM instances that don't support live migration, organized by Confidential Computing technology type. Use these periods to plan around host maintenance events accordingly.
| Confidential Computing technology | Machine type | Notification period |
|---|---|---|
| AMD SEV | C2D | None |
| C3D (Preview) | 7 days | |
| AMD SEV-SNP | N2D | 1 hour |
| Intel TDX (Preview) | c3-standard-* |
7 days |
Simulate a host maintenance event
To test how your workload behaves when a Confidential VM instance is restarted, you can simulate a host maintenance event. As you work through the simulated event, think about how you can change your workload to be resilient when a real event happens.
Change your workload
After simulating a host maintenance event, you can use the experience to modify your workload to respond to one. For example, you might make use of one of the following techniques:
Adjust your workload to monitor regularly for upcoming host maintenance events.
Before an event happens, temporarily move your work in progress to a Cloud Storage bucket. After the Confidential VM instance restarts, have the workload retrieve the data and resume work.
Write all data to a secondary Persistent Disk. After the Confidential VM instance restarts, reattach the Persistent Disk so your workload can resume work.
Provision your Confidential VM instances on sole-tenancy nodes
If it fits your needs, you can provision your Confidential VM instances on a sole-tenant node. This lets you determine a 4-hour window in which Google can perform maintenance on your VMs. It also lets you perform a manual live migration, which moves your VM to a different node or node group you control.
What's next
Learn about designing resilient systems.