Correlated Failure
Failures that occur together because redundant components share a hidden dependency, invalidating the independence assumption behind redundancy calculations.
Redundancy arithmetic assumes independence. Real systems are full of shared elements that break the assumption:
Shared infrastructure — the same rack, power feed, network segment, hypervisor or availability zone.
Shared software — the same version with the same bug, so a deployment fails every replica simultaneously. This is why progressive rollout is a redundancy mechanism, not merely a release practice.
Shared configuration — a bad configuration push reaching every instance in seconds. Configuration changes cause a large share of major outages precisely because they are not treated as deployments.
Shared dependencies — every replica depending on one identity provider, DNS zone, certificate authority or secret store. The redundancy is real and the dependency is not.
Shared operators — one command applied to all instances.
The diagnostic question is worth asking literally: "what would have to fail for both of these to fail at once?" The answer names the real single point of failure, which is usually not the component the redundancy was designed around.