skip to content

During a real layer-7 flood, Cloudflare's HTTP DDoS Attack Protection is mitigating yet the origin still falls over; how do you tighten the response?

level: seniorimportance: should knowfreq 26%

answer

  1. mitigated is not the same as stopped
  2. low confidence means a softer action
  3. the leftover event override
  4. Block at rule scope

basics

~20 s

Find the rule and the softer action it applied, override that rule to Block, undo any earlier override that lowered sensitivity, and if the flood stays under the rules' thresholds add Under Attack mode, rate limits and custom rules.

solid answer

~50 s

This is an **incomplete mitigation**: a rule fired, but with an action softer than the attack needs. Lower-confidence rules apply Managed Challenge or Force Connection Close, and `generic` flood rules rate-limit or challenge by default. Filter security events by `Service equals HTTP DDoS`, note the rule and its action, and override that rule's action to Block; it applies within one or two minutes. Then audit existing overrides: an on-sale exception that set rules to Low or *Essentially Off* still applies to the attack, and an override only matches traffic above its own sensitivity threshold. Restore the defaults Cloudflare recommends, High sensitivity and default actions. If part of the flood matches no rule (a false negative), add Under Attack mode on the targeted paths, rate limiting rules and WAF custom rules, and contact Support with timestamps if the origin still suffers.

go deeper

for a junior

Know that mitigated is not the same as stopped: a DDoS rule can challenge or close connections instead of blocking, and an override can change its action to Block.

for a middle

Explain how rule confidence picks the action, why Force Connection Close only makes clients reconnect, and that High is already the most sensitive level.

for a senior

Separate incomplete mitigation from false negatives, audit leftover overrides and their sensitivity thresholds, restore defaults, and layer Under Attack mode, rate limits and custom rules where rules do not match.

for a principal

Make temporary DDoS exceptions expire by policy: each override has an owner and a revert date, so a past event's tuning cannot quietly weaken the zone during the next attack.

## Two different failures When an origin keeps failing during a layer-7 flood that Cloudflare is visibly mitigating, one of two things is usually happening, and the DDoS docs name both: | Failure | What it means | Typical cause | |---|---|---| | **Incomplete mitigation** | A DDoS rule fired, but its action lets too much through | A low-confidence rule chose Managed Challenge or Force Connection Close; a `generic` rule rate-limited instead of blocking | | **False negative** | Part of the attack is not identified at all | Traffic stays under every rule's threshold, or matches no rule | The fixes differ, so read the events before acting. ## Diagnose from the events In the zone's security analytics, filter by `Service equals HTTP DDoS` and read: - **Which rules fired**, under **Top events by source** and **HTTP DDoS rules**. - **Which action each applied.** The system picks the action by confidence: high-confidence rules block, low-confidence rules challenge or force connections closed. Force Connection Close does not block a request; it makes the client reconnect. A determined flood can reconnect and continue. - **Whether an override is in play.** The HTTP DDoS Attack Alert names the matched rule and any override in force. Alerts are not repeated within the same one-hour window, so a quiet inbox mid-attack does not mean the attack has ended; read the events. - **What reaches the origin.** Compare mitigated volume with what the origin logs. If the origin also sees traffic that never appears in the events, attackers may be bypassing Cloudflare, which no DDoS setting can fix; closing that path is origin-lockdown work. ## Fix an incomplete mitigation 1. Create an override, or edit the existing one if the zone has used its allowance. Browse rules, paste the rule name, and set its **Action** to **Block** at rule scope. 2. Save. The change takes effect within one or two minutes and should show in the analytics straight away. 3. Do not try to raise sensitivity above the default: *High* (API value `default`) is already the most sensitive level. The internal actions Connection Close, Force Connection Close and DDoS Dynamic cannot be set in an override. ## Audit what earlier tuning left behind This is where a ticketing site's history bites. An override written for last month's on-sale may still be in force: - A ruleset or tag set to Low or *Essentially Off* raises thresholds for every attack, not just for buyers. - An override only matches when traffic crosses its own sensitivity threshold. With Advanced DDoS Protection and several ordered overrides, a smaller attack under a Low override's threshold falls through to the next one, which may be a Log override. - If the zone has its own overrides, account-level overrides are not evaluated for it, so a protective account-wide rule may be silently out of play. Cloudflare's proactive-defence guidance is to keep the DDoS managed rulesets at default settings (High sensitivity, default mitigation actions), so restoring those defaults is often the most effective single change. ## Close a false negative When part of the flood matches no rule, the documented route for a WAF/CDN zone is to combine controls that the DDoS rulesets do not provide: - **Under Attack mode** on the targeted browser-facing paths, through a Configuration Rule, as a last resort; - **rate limiting rules** for per-client budgets on the endpoints being hit; - **WAF custom rules** for positive and negative security policy based on known usage; - **caching** of everything that can be cached, so the edge answers repeated requests instead of the origin. Each of these has its own configuration material; the point here is that they complement the DDoS rules rather than replace them. ## Escalate with evidence If the origin still cannot cope, contact Cloudflare Support with: 1. the attack's time window as UTC timestamps; 2. the zone and the targeted host or path; 3. the attack frequency; 4. steps to reproduce, with actual against expected results; 5. relevant origin logs, error messages or screenshots. Afterwards, keep the Block override if the rule proved reliable for your traffic, and review it when Cloudflare changes the rule's default action in its changelog.

  • Why does Force Connection Close sometimes fail to protect an origin during an HTTP flood?
    It does not block the request; it closes the connection so the client must reconnect, with Connection: close on HTTP/1 or a GOAWAY frame on HTTP/2. A legitimate client reconnects and carries on, and so does a determined attacker. When the analytics show this action on a rule that keeps overloading the origin, override that rule's action to Block.
  • A zone has an account-level HTTP DDoS override that blocks a known bad IP list, then adds its own override for an on-sale; what happens?
    The account-level override stops being evaluated for that zone, because only zone-level overrides apply once the zone has its own. The bad-list configuration has to be repeated in the zone's override, or the zone silently loses it during the next attack.

  • Raising sensitivity above High to make a rule fire harder.
  • Setting an override's action to Force Connection Close to stop a flood.
  • Leaving last month's Essentially Off on-sale override in place during an attack.
  • Treating every overloaded origin as a DDoS tuning problem when attackers may bypass the proxy.
  • Expecting the DDoS rules alone to cover floods that stay under every threshold.