A DDoS runbook is the written plan that tells the people on duty what to do in the first hour of a flood: who declares it, who decides, who calls the protection provider, and what gets written down. Whether those steps still work with a real attack running is what a DDoS test measures. A runbook is not a policy or an architecture document. It is short, it names roles rather than people, and it assumes the reader was asleep ten minutes ago.
What a DDoS runbook contains
A declaration rule. Most outages are not attacks, and most attacks first look like an outage. The runbook states which signals turn "the site is slow" into "this is a DDoS": a traffic rate against the baseline, a share of requests from new sources, the provider's own alert. Without it, the first twenty minutes go on a capacity investigation.
Decision owners, with deputies. Each decision the hour forces has one role attached and a named fallback for nights and holidays. The usual ones are diverting traffic to scrubbing, moving a rule from count to block, blocking a region, and saying something in public.
Pre-authorised actions. Some steps should not wait for anyone. A good runbook lists the controls the on-call engineer may switch on alone, up to an agreed limit, and the ones that need the decision owner. That line is drawn in calm weather, not at 3 a.m.
The provider call. The support number, account identifier, the authentication the provider will ask for, and what to tell them: target addresses, start time, what you already see. This step commonly fails, because the details were last checked when the contract was signed.
A log template. Every change gets a time, the person who made it, and what it was meant to do. The log feeds the timeline, the post-incident review and any regulator who asks.
Why it matters under attack
The slowest minute is usually a decision. Detection and routing have improved; a person deciding whether they are allowed to divert has not. The gap between "we know" and "we acted" is often the largest share of time to mitigation, and a runbook, with its pre-authorised actions, is the control written for exactly that gap.
The runbook must survive the attack. If it lives on a wiki behind single sign-on, behind the same network that is being flooded, nobody can open it when it matters. The same goes for a provider phone number kept only in a ticketing system on that network. A copy that works offline is part of the design.
Each action has a side effect. Diverting the prefix starts a mitigation cutover with its own failure modes. Blocking a country removes real customers. A runbook that names the action without its known side effect hands the on-call engineer a second incident.
It goes stale quietly. People move teams, rules get renamed, the provider changes its process. Nothing breaks until the day it is used.
What a DDoS test proves about a runbook
A live test is the only planned time the runbook meets real traffic with nobody pretending. Run it from the document, not from memory, and time each step on one clock: when the declaration rule fired, when the decision owner answered, when the provider picked up, when the action took effect.
Then check what the clock cannot show. Did the provider number reach someone who could act? Did the on-call account actually hold permission for the pre-authorised controls, or only read access? Did anyone leave the runbook to improvise, and why? Every one of those is a line to rewrite, and it belongs in the DDoS test report next to the technical findings.
A test checks the steps work. Whether the people would take the right decisions under pressure is a separate rehearsal, and a DDoS tabletop exercise is where a team finds out before the next flood does.