Instance unavailable after being terminated by Compute Engine

[Giuseppe Lanzi]

Hi, I've got one of my instances terminated and restarted by Compute Engine and then it became unresponsive.

I had to create another instance.

Usually the migration procedure is marked as "migration" not as "termination + start".

May I ask the difference? When does System decide that an instance has to be terminated?

Is there a way to prevent this? My customer got his service down for sometime.

Best

[Carlos (Cloud Platform Support)]

Hi Giuseppe,

Migrations are transparent events that appear in the operations logs as “compute.instances.migrateOnHostMaintenance”. You can find details on how live migration works on this blog. The instance maintenance behavior can be configured in the availability policies.

A different event happens when there is a host error. Those appear in the operation logs as “compute.instances.hostError”. In that case the VM will be restarted in a different host. This is not configurable, since if the host is down all the guests will also be. If the instance was having issues after restarting, it is possible that a problem at the application layer could have happened. A good place to have a look is the serial console as well as the application logs. You can find information on the different instance status here and on this one how to design robust systems.