Google Cloud Bare Metal Solution exists for workloads that can't run virtualized, which in practice means licensing-constrained databases and legacy enterprise systems. Those workloads are usually critical, and the tier they run on has none of the conveniences your team has grown used to.

There's no live migration, no instant reprovisioning, and no autohealing. A hardware problem means real downtime measured against hardware replacement timelines. The applications running here often predate the idea of designing for failure, and whatever redundancy exists was configured by people who have since left.

Install the Gremlin Agent on your bare metal servers and you can measure recovery before you have to perform it. Shutdown experiments establish how long failover to redundant hardware genuinely takes. CPU and memory experiments show whether these systems degrade gracefully under pressure or fall over, which matters more here than anywhere else in your estate.

When the answer to "how long until we're back" might be hours rather than seconds, knowing the number changes what you promise the business.