How we push a kill switch to 312 edge nodes in 38 ms
A kill switch that takes a minute to arrive is a status page with extra steps. How our control plane fans a single write out to 312 edge nodes, and where the 38 ms actually goes.
Blog
How we build the thing that decides what your users see — and how teams use it.
A kill switch that takes a minute to arrive is a status page with extra steps. How our control plane fans a single write out to 312 edge nodes, and where the 38 ms actually goes.
Rollouts that advance on a schedule while every guardrail holds, and roll themselves back the moment one does not. Plus stale-flag reports and a faster CLI.
Lumenfold deploys 60 times a day to 4 million readers. Their rollback used to be a 40-minute redeploy. Now it is a toggle, and their on-call rota is boring again.
A 25% rollout is not "one request in four". It is a stable answer to "is this user in?", identical on every server, that never flips a user back out. That is a hashing problem.
A flag is borrowed flexibility: cheap to take out, expensive to carry. Three rules that keep the interest down, and the one report that tells you when to pay up.
OpenFeature gives your code one flag API whatever vendor sits behind it. Here is a complete provider for Driftwire, in about 80 lines of TypeScript.
A flag without an owner is an outage with a timer on it. Owners are now required, inherited from CODEOWNERS, and stale flags file their own removal tickets.
Ten boolean flags make 1,024 combinations, and you will not test them all. You do not need to. Test the defaults, the targets, and the pairs that touch the same code.
Parcelwick had 2,000 flags in a homegrown YAML service and nobody willing to touch it. They moved every one in a weekend, without a single user noticing.
Long-lived branches feel safe and merge badly. Trunk-based development with flags gets you small merges and safe releases — if you separate deploying from releasing.