Correlated failure
termcloudIT support
Failures that happen together because the things that failed share something - a control plane, a configuration pipeline, a provider, a physical route - rather than failing independently as a redundancy calculation assumes.
Redundancy arithmetic assumes independence, and almost nothing is independent. Two providers reselling one platform, a backup in another region of the same cloud, a secondary DNS service peering at the same exchange: all present as diversity and behave as one component. The practical test is not how many vendors appear on the invoice but what happens if any single shared element is removed - and the answer usually has to be traced through the dependency graph rather than recalled from the architecture diagram.
Also known as: shared failure domain, common mode failure, retry storm