Cluster, Resource, and Availability Management
High availability
vSphere HA restarts VMs on surviving hosts after a failure. Admission control reserves capacity using one of several policies: cluster resource percentage, slot policy, or dedicated failover hosts. Heartbeat datastores detect host isolation; isolation response decides whether to power off or leave VMs running; APD and PDL responses handle storage loss. Proactive HA reacts to hardware health reports from supported vendors.
DRS and balance
DRS migrates VMs to balance CPU and memory load and to satisfy rules. Automation levels run from manual to fully automated, with migration thresholds. VM/VM and VM/Host affinity rules may be "should" (soft, violated if needed) or "must" (hard, may block power-on). EVC masks CPU features so a cluster presents a common baseline, enabling migration across generations.
Resource pools
Shares are relative among siblings, reservations guarantee resources, and limits cap them. An expandable reservation lets a child pool borrow from its parent; a non-expandable reservation does not. Resource pools do not enforce per-VM fairness inside a pool the way shares at the VM level do.
Traps
Percentage-based admission control is not the same as slot policy, and slot size is driven by the largest reservation. "Must" VM/Host rules can prevent a VM from powering on after host maintenance. Changing an EVC baseline may require powered-off VMs. Shares only matter under contention.
Rule of thumb
Size admission control so that N+1 host failure leaves enough reserved capacity for the full VM set, and validate any "must" rule against maintenance-mode operations before enabling it.