fault tolerance
termnetworkingcloud
Continuing to provide service while something is broken - which requires deciding in advance which failures are survivable and paying for that in redundancy.
The word is used loosely and the useful version is specific: you tolerate the failures you designed for, and no others. That means naming them - one power feed, one disk, one site, one region - and accepting that anything unnamed will take the service down. The recurring mistake is redundancy that shares a dependency: two devices on the same power strip, two links in the same duct, two regions with one identity provider. The second copy only helps if it fails independently, and independence is the property people forget to buy.
Also known as: resilience, high availability