Regional Evacuation

The distinction from disaster recovery is that evacuation is practised regularly and invoked readily, which changes its reliability entirely. A procedure exercised monthly works; one documented and never run is a hypothesis.

Netflix's published practice of regularly evacuating a region is the well-known example, and the argument generalises: if you can move traffic out of a region on a normal Tuesday, you can do it during an incident, and you no longer need to diagnose what is wrong in that region before acting.

What it requires. Capacity headroom in the remaining regions to absorb the evacuated traffic — which must be verified, not assumed, and is the constraint that most often fails. Traffic shifting that is fast and gradual, so the move can be aborted. Data that is already replicated, since an evacuation cannot wait for a copy. And no hidden regional dependencies — a service in the surviving region that calls something only present in the evacuated one.

The organisational benefit is as large as the technical one: it turns regional failure from an extraordinary event requiring senior decision-making into a routine mitigation an on-call engineer can invoke. That is the same principle as draining a cell — a generic mitigation applied without diagnosis.