You are preparing for a scheduled maintenance window to upgrade AOS on a four-node Nutanix cluster hosting critical production VMs on AHV. The window is tight, and downtime is strictly prohibited. You initiate the upgrade via Prism Element. How does the upgrade process handle the running VMs to meet the zero-downtime requirement?

An AOS upgrade is a rolling upgrade of the Controller VMs, not of the hypervisor hosts. Prism stages the new AOS package on every CVM, then upgrades and restarts the CVMs one at a time, waiting for each to rejoin the cluster and for services to stabilize before moving to the next. The user VMs keep running on their AHV hosts throughout. While a host's local CVM is restarting, AutoPathing redirects that host's storage I/O to a healthy CVM elsewhere in the cluster, and once the local CVM is back, I/O returns to it so data locality is restored. Because only one CVM is out at a time and data is protected by RF across nodes, storage stays online and writable. VM live migration is what happens during AHV hypervisor upgrades or host maintenance, where the host itself must reboot; it is not needed for an AOS-only upgrade. Pausing all I/O and rebooting every node together would cause an outage, which is the opposite of the rolling design. The upgrade does not stop for manual approval after the first node; it proceeds automatically once pre-checks pass. Moving VMs to another cluster and powering them off is unnecessary downtime. Exam caveat: AOS upgrade = CVMs roll and I/O is redirected; AHV upgrade = hosts enter maintenance and VMs live-migrate. Operational check: run the upgrade pre-checks and NCC first, and confirm the cluster can tolerate one node down in Data Resiliency status.