The $6 VPS that outlived three client clouds

I rented a virtual machine in Newark in March 2019 for six dollars a month. It has one shared core, a gigabyte of memory, and 25 GB of disk. It currently serves this site, four documentation sites, a status page, and the webhook endpoint that tells me when driftwatch finds something.

In the same period I have watched three client platforms get migrated, rebranded, repriced, or shut down. The little machine in Newark has been rebooted eleven times, nine of them by me, and has never once sent me an email about a pricing tier.

What is actually on it

caddy      reverse proxy, automatic TLS
systemd    four units, one timer
sqlite     two databases, litestream to object storage
crosstown  deploys, health checks, rollback

That is the whole stack. No container orchestrator, no service mesh, no YAML I have to re-learn every eight months. Deploys are a crosstown command that uploads a binary, starts it on a second port, waits for two consecutive health checks, and flips the proxy. If either check fails, nothing flips and I get a message.

Small infrastructure isn’t a compromise. It’s the version you can hold entirely in your head at 2am.

The honest caveats

1. One machine means one region and one blast radius. Everything important is backed up off the box, hourly, and I have restored from those backups on purpose twice.
2. It would fall over under a real traffic spike. It has been hugged to death exactly once, by an orange website, for about forty minutes in 2022.
3. I am the on-call rotation. That is fine for my own things and unacceptable for a client’s payroll system, which is why those live somewhere with a support contract.

Uptime since the last reboot is 214 days. Total cost of ownership over six years is about four hundred and thirty dollars, or roughly one afternoon of consulting. I am aware this is not a benchmark. It is a love letter.

— O.P., FORT GREENE, VIA NEWARK