Zero-downtime certificate rotation at the edge (part 2)
Over the last few months we have spent a lot of time on zero-downtime certificate rotation at the edge. This is what we learned.What actually happens during a floodThe first thing to…
Engineering
Anycast is not magic: how we route attacks across 40 PoPs
Here is a question we could not answer well a year ago: what really happens with anycast is not magic: how we route attacks across 40 PoPs? We can answer it now.Measuring…
Building a per-route baseline without storing every request
Over the last few months we have spent a lot of time on building a per-route baseline without storing every request. This is what we learned.Observability firstEvery mitigation decision is logged with…
Why our logs are append-only and how we query them (part 3)
We get asked about why our logs are append-only and how we query them more than almost anything else, so here is the long answer.Measuring successWe track three numbers for every incident:…
Keeping our control plane alive when the data plane is on fire
keeping our control plane alive when the data plane is on fire sounds like a narrow topic. It turns out to touch almost every part of how an edge network behaves under…
Why we wrote our proxy data plane in Rust (part 2)
This post is about why we wrote our proxy data plane in Rust. It started, as most of our posts do, with an incident that did not go the way we expected.Compliance…
Why we wrote our proxy data plane in Rust (part 3)
why we wrote our proxy data plane in Rust sounds like a narrow topic. It turns out to touch almost every part of how an edge network behaves under attack.Latency budgetOur budget…
How we deploy rule changes to every edge in under 30 seconds (part 4)
This post is about how we deploy rule changes to every edge in under 30 seconds. It started, as most of our posts do, with an incident that did not go the…
Tail latency is the only latency that matters (part 2)
We get asked about tail latency is the only latency that matters more than almost anything else, so here is the long answer.Latency budgetOur budget for the whole filtering pipeline is one…
How we test WAF rules against seven days of real traffic (part 4)
how we test WAF rules against seven days of real traffic sounds like a narrow topic. It turns out to touch almost every part of how an edge network behaves under attack.Latency…
Anycast is not magic: how we route attacks across 40 PoPs (part 2)
Here is a question we could not answer well a year ago: what really happens with anycast is not magic: how we route attacks across 40 PoPs? We can answer it now.Why…
What we learned from load-testing our own edge to failure (part 2)
what we learned from load-testing our own edge to failure sounds like a narrow topic. It turns out to touch almost every part of how an edge network behaves under attack.The economics…