Unlike a mitigation cutover, which sends traffic through a scrubbing tier and hands the clean part back to the same origin, DDoS failover does not clean anything; it moves the service itself to a second path, site, region or provider when the first is overwhelmed. It is a resilience decision taken long before the attack, and whether the second place can actually carry the load is a question only a DDoS test answers. On paper, every failover works. Under a flood, the weak points are the trigger, the move itself, the capacity waiting on the other side and the way back.
How DDoS failover moves traffic
There are three common levers, and each moves something different.
DNS failover changes which address a name resolves to, usually driven by health checks. It moves the users who look the name up again. It does not move an attacker who already has the primary address, and it is as slow as the TTL and the resolvers that ignore it.
Routing failover withdraws or shifts a BGP announcement, or an anycast site stops advertising, so the network delivers traffic elsewhere. It moves everything aimed at the prefix, attack included.
Provider failover swaps a CDN, a scrubbing contract or an upstream ISP for another. It is often a manual decision with a phone call in the middle, and it depends on the second provider already knowing your certificates, origins and rules.
Active-active designs blur the line: both sites already carry traffic, and failover just means one takes it all. That removes the cold start, but not the capacity question.
Why it matters under attack
The secondary is usually smaller. It was sized for a disaster that takes out a data centre, not for the whole of a flood plus the normal peak. Fail over into it and the attack may simply follow.
Whether the attack follows is the real question. If the flood is aimed at the name, a DNS failover delivers it to the new site within one TTL. If it is aimed at an address, the primary keeps taking it while users drift away, which works only if nothing still depends on the primary. An exposed origin address, the problem origin IP exposure describes, defeats the whole move.
Health checks lie under load. A check that fails because the link is saturated, then passes because the attacker paused, flips the service back and forth. Each flip drops sessions and restarts a cold cache. A pulse-wave attack produces exactly that rhythm.
Shared dependencies fail together. Two sites behind the same DNS provider, the same identity service or the same upstream link are one failure with two addresses.
What a DDoS test proves about failover
A test runs the failover under attack-shaped load rather than by pulling a cable on a quiet afternoon, and times each stage on one clock: when the trigger fired, when the first real request reached the secondary, and when the last stopped reaching the primary. Those stages are part of the time to recovery the business actually sees.
Then it checks what a stopwatch cannot. Did the secondary hold the same flood, or only the users? Did the attack follow, and was the plan built for either answer? Did the switch settle, or did health checks flap it back and forth? And did failback, which teams rarely rehearse, return traffic without a second outage?
A clean result is a move that completes within the agreed time, settles once, and leaves real users served on the far side while the flood keeps running. Anything less is a plan that has not yet met a flood. Failover across a cloud and an on-premise site adds its own seams, the interconnect and the protection that differs on each leg, which hybrid-cloud DDoS testing works through.