Definition: Scale to zero
Scale to zero means idle compute stops billing: functions charge 0.01 credits per GiB-second only while executing, and workers with min_instances: 0 stop after their idle_timeout_ms. An idle project spends on storage (0.04 credits per GB-hour), not on standing replicas.
How each unit idles
Functions have no standing instance to idle: the meter is execution itself. Workers are processes, so their declaration carries the idle policy — idle_timeout_ms before an instance stops, min_instances for how many stay warm (0 is scale to zero), max_concurrent for how much each instance pulls. The policy is per worker, so the latency-critical one can stay warm while the nightly report worker sleeps twenty-three hours a day.
What scale to zero does to the bill’s shape
Projects stop having a floor. An internal tool used on weekday mornings, a preview environment clicked twice, an MVP awaiting its first customers — each trends toward its storage cost between uses, which is why the credit balance reads like a usage meter instead of rent. The trade is the cold start on the next request, bounded by keeping min_instances at 1 where humans wait impatiently.
Frequently asked questions
A stopped worker or an uncalled function starts on demand; where latency matters, declare min_instances: 1 and pay for one warm instance deliberately. The point is that warmth is a priced choice per workload, not a platform-wide default.
Its storage: tables, buckets, and indexes at 0.04 credits per GB-hour. For a small project that is single-digit credits a month — the reason preview and side projects can exist without a finance conversation.
Yes. A workflow step or schedule invoking a stopped worker starts it; scale to zero governs idleness, not availability. The run’s history shows the start like any other step.
Related capabilities, architectures, and comparisons
Ship your stack on the Software Lifecycle Cloud.
Closed preview on dedicated European infrastructure. Bring a repository or start from a blueprint.