Documentation
¶
Overview ¶
Command killswitch-lambda fires the killswitch from an AWS Budgets action, so the fast trip is driven by the spend signal itself.
Give it a generous timeout — 5 minutes is a reasonable floor. Discovery across many regions is not fast, and a Lambda killed mid-fire leaves a snapshot written with only some of it applied. That is recoverable, because the snapshot goes in before the first change, but it is not what anyone wants to find.
A cron schedule can only be as fast as its interval: the worst case is a full interval of spend before anything happens, and shortening the interval trades cost for latency without ever getting below it. A Budgets action fires when the threshold is crossed.
Everything about *what* gets stopped lives in internal/trip, which the CLI uses too. There is no second implementation here — the path that only runs at three in the morning during an incident is the one nobody would notice was wrong.
Source Files