Availability zone

On December 7, 2021, the Amazon Web Services us-east-1 region lost part of its management interfaces and internal services for about seven hours. An automated scaling activity had congested the devices linking AWS's internal network to its main network, and that internal network serves every zone in the region, none of which had lost a data center. An availability zone is a subdivision of a cloud region, made up of one or more data centers with their own power, cooling and network feeds. It is sold on the promise that a failure in one will not reach the others. That promise is the economic basis of every so-called high-availability architecture, since the customer pays twice for the same capacity so that the probability of simultaneous loss becomes negligible. The point underwriting needs to press is that the advertised independence covers physical plant, not control planes, those cross-cutting authentication, routing and orchestration services that are by design shared across an entire region. A control-plane incident therefore leaves the zones unusable together while none of them has actually gone down, and the insured discovers that the redundancy bought protection against fire but not against a software fault. The useful question is therefore which components those zones still share, never how many zones the insured has subscribed to.