The full story · Nov 2026 to now · solo

Demo platform

Hosting for my demos that stays off until a visitor presses Launch, then stops itself.

0servers running when nobody is visiting

104 swarm start for an app

190 sto recover from losing a worker abruptly

15 minvisitor session, inside daily and monthly limits

The hard parts

Problem: Always-on servers for demos cost money every hour, used or not.
What I did: Each app has its own group of machines that scales from zero. A Kubernetes operator I wrote in Go starts a session, applies the pinned release, waits until it is ready, and drains the machine on stop.

Problem: Demo servers on the public internet attract attacks.
What I did: There are no inbound ports. Traffic arrives through Cloudflare Tunnels, launches pass a bot check, and I connect without SSH.

Problem: The setup went from one server to several while people were using it.
What I did: A phased migration with a live cutover. When one change stopped halfway, I stopped at a known state and audited every recovery plan before finishing, with zero data volumes recreated.

Terraform owns the AWS and Cloudflare resources
A launcher that enforces session length and daily and monthly hour limits
A launch page per app, with a bot check
Pull-based releases with tested rollback and empty-disk rebuild
Controlled from an admin console with passkey sign-in and short-lived AWS credentials

GitAWSKubernetesOIDC federationLinuxGoDynamoDBPasskeys (WebAuthn)TerraformKubernetes operatorsCloudflareAWS Lambda