Comparison
Cascading failurevsSingle point of failure
Cascading failure
one service got slow, its callers filled up with waiting requests, and the failure walked all the way up to the front door.
Failure propagating between components as each one's resources are consumed waiting on the last. It is why a non-critical dependency can cause a full outage: the criticality is in the coupling, not the feature. Timeouts, bulkheads, circuit breakers and shedding all exist to break the propagation path.
Full entry →Single point of failure
you trace the dependency graph and find one component whose failure takes everything else down with it.
A component with no redundancy whose loss causes total failure. They are easy to spot in the data plane and easy to miss in the control plane: the config service, the DNS zone, the single deploy pipeline, the one person with production access. Redundancy that shares a failure mode is not redundancy.
Full entry →