Sole-tenant nodes on Google Compute Engine are usually adopted for licensing or compliance reasons rather than architectural ones. Finance or legal drives the decision, and the operational consequence arrives as a side effect nobody explicitly chose: a group of VMs now shares one physical failure domain.

That matters when those VMs include redundant instances of the same service. Anti-affinity rules in your orchestration layer mean nothing if the underlying nodes place both replicas on the same hardware. The redundancy looks correct at every level anyone routinely inspects.

Install the Gremlin Agent on the VMs sharing your node and Shutdown experiments reveal whether your placement genuinely distributes risk. CPU and memory experiments apply pressure to one VM while you watch its neighbors, which tells you how much isolation the tenancy model actually delivers under load.

The question is whether your logical topology matches the physical one, and it goes unexamined precisely because the decision to use sole-tenancy was driven by concerns other than availability.