Containers: always-on or minimum-instances option

[karanbalaji]

We run an always-on Rust worker that subscribes to a database over a long-lived outbound connection and runs sandboxed jobs. The Containers lifecycle sleeps after 10 minutes by default (sleepAfter), and billing is built around scale-to-zero. We would like a documented way to keep one instance running, with predictable cost, for workloads that watch a queue instead of serving requests.

A small always-on machine on Fly costs us about $2 to $4/month. We would like to know how a continuously running Container would be priced.

Related: https://github.com/cloudflare/workers-sdk/issues/16075

[sueun-dev]

For a Container subclass (default scheduling), the docs describe overriding the expiry hook:

import { Container } from "@cloudflare/containers";
export class AlwaysOn extends Container {
  override async onActivityExpired() {}
}

Without a stop()/destroy() call, the SDK renews the timeout instead of performing its idle stop. I wouldn't treat this as a minimum-instance guarantee: Cloudflare's FAQ says host events can stop active containers and gives no fixed-period uptime guarantee. I'd keep restart/reconnect handling for the subscriber.

Current pricing bills provisioned memory/disk while running, plus active CPU usage. A lite instance running for 720 hours in one billing month uses 180 GiB-hours of memory and 1,440 GB-hours of disk. With the monthly 25 GiB-hour and 200 GB-hour allowances wholly unused, those two charges calculate to about $1.71. The $5 Workers Paid plan is required. CPU, egress, Workers/Durable Objects and logs can add usage charges beyond their included allowances. This is an illustrative calculation, not a measured bill.