Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
HTTP/2 Rapid Reset (CVE-2023-44487) abuses stream cancellation to force servers into unlimited request processing, producing the largest application-layer DDoS attacks ever recorded.

What Is the HTTP/2 Rapid Reset Attack?
HTTP/2 Rapid Reset, catalogued as CVE-2023-44487, is a denial-of-service technique that turns a protocol efficiency feature into an amplifier. A client opens a request stream and cancels it in the same breath, over and over, so that the server performs the expensive part of the work while never reaching the concurrency limit that is supposed to hold it back. Google, Cloudflare, and Amazon Web Services disclosed it jointly on 10 October 2023 after absorbing the three largest application-layer attacks measured up to that point.
HTTP/1.1 ties one request to one connection at a time. HTTP/2 multiplexes many independent streams over a single TCP connection, and each endpoint advertises a SETTINGS_MAX_CONCURRENT_STREAMS value—commonly 100 or 128—to cap how many may be open at once. The protocol also lets either side abandon a stream immediately by sending a RST_STREAM frame, which is genuinely useful when a user navigates away from a slow page.
The flaw is in the accounting. When a client resets a stream, the server frees the slot straight away, but the request has already been dispatched to the application. An attacker sends a HEADERS frame followed instantly by RST_STREAM, then repeats. From the server’s point of view no more than one stream is ever open; from the backend’s point of view, thousands of requests per second are arriving and being processed. The round-trip latency that normally limits a client’s request rate disappears entirely, because the attacker never waits for a response.
The numbers from August and September 2023 were an order of magnitude beyond anything published before. Google mitigated a peak of 398 million requests per second, Cloudflare 201 million, and AWS 155 million. For comparison, Cloudflare’s previous record was 71 million requests per second in February 2023. Google noted that two minutes of the attack delivered more requests than the total number of article views Wikipedia received during the whole of September 2023.
Just as striking was the size of the botnet. Roughly 20,000 machines generated the Cloudflare attack. Because each node could issue requests without waiting, a modest number of hosts produced volumes that previously required hundreds of thousands.
This was a design problem, not a single vendor’s bug, so almost every HTTP/2 implementation needed changes: nginx, Apache Tomcat, Envoy, HAProxy, Netty, Jetty, Microsoft IIS, the Go standard library, Node.js, and the Kubernetes ingress controllers built on them. A long run of related CVEs followed, including CVE-2023-45288 in Go’s HTTP/2 handling and the HTTP/2 CONTINUATION flood described in April 2024, all variations on the theme of cheap client work producing expensive server work.
Rate limiting by request volume alone is a poor fit, because legitimate HTTP/2 clients also cancel streams. Effective controls include:
RST_STREAM frames and terminate the connection with a GOAWAY frame carrying ENHANCE_YOUR_CALM once the ratio of cancelled to completed streams crosses a sensible threshold.keepalive_requests setting.The NCSC and CISA both issued guidance at disclosure urging organisations to apply vendor patches rather than disable HTTP/2, which would cost real performance for a problem that is fixable in software.
Volumetric techniques such as reflection and amplification aim to exhaust bandwidth with spoofed traffic. Rapid Reset consumes CPU, worker threads, and database connections using a legitimate, fully established TLS session from a real IP address. That makes it harder to filter upstream and easier to mistake for a traffic spike, and it is the reason application-layer DDoS defence needs protocol awareness rather than packet counting alone.
If you want to know how your origin behaves when concurrency limits stop protecting it, our DoS/DDoS Testing Services measure the breaking point under controlled conditions.
Only as a short-term measure if no patch exists for your stack. Disabling HTTP/2 removes multiplexing, header compression, and server push, which measurably slows page loads for users on high-latency connections. Every major server and proxy has shipped a fix that limits stream cancellation rates, so patching and tuning the reset threshold is the better answer in almost every environment.
Not by itself. A traditional WAF inspects request content, and these requests are individually well-formed and often never complete. Mitigation has to happen at the HTTP/2 connection layer, where the proxy can see the ratio of cancelled to completed streams. Many WAF and CDN providers added that logic in October 2023, so the protection comes from the platform rather than from rule sets.
HTTP/3 over QUIC uses a different transport and its own stream management, so the exact frame sequence does not apply. The underlying pattern—cheap client-side cancellation forcing expensive server-side work—does carry over, and researchers have demonstrated comparable request-cancellation pressure against QUIC implementations. Any multiplexed protocol needs explicit accounting for abandoned work.
Look at the ratio of HTTP/2 streams reset to streams completed, and at the time between a HEADERS frame and its RST_STREAM. Legitimate cancellations are a small fraction of traffic and usually arrive after a measurable delay. In an attack, nearly every stream is cancelled within milliseconds, backend CPU rises sharply while bytes transferred stay flat, and the source addresses are a stable set rather than a growing one.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.