Redundancy
Having more instances of a component than the load requires, so that failures can be absorbed without loss of service.
The notation is worth knowing because it is the difference between surviving a failure and surviving a failure during maintenance. N is the capacity needed. N+1 tolerates one failure. N+2 tolerates one failure while one node is already out for patching — which is the realistic combination, since incidents do not wait for change windows. 2N is full duplication.
Two things undermine it. Correlated failure: instances that share a host, a zone, a control plane, a configuration push or a library version are not independent, and the availability arithmetic silently stops holding. Untested failover: standby capacity that has never taken traffic frequently cannot — stale configuration, expired certificates, cold caches.
And redundancy must include headroom for the failure itself. Two instances each running at 70% cannot absorb the loss of one.