Every server on a hypervisor cluster shares one control plane. Touch it the wrong way and your “single point of change” becomes a single point of failure for every workload sitting on top — VMs, backups, databases, all of it at once.
The rules we follow
- Live migration to a fully patched node before touching the host being updated
- A validated rollback path tested against a mirror cluster first
- Storage and network dependency mapped so no dependency is assumed — only proven
We have seen “five-minute maintenance” upgrades turn into multi-hour outages when a storage driver mismatch surfaces mid-flight. The discipline is treating an everyday patch with the same gravity as a disaster-recovery failover, because to your business it is one.