muhaha

Running a small VPS without losing sleep

2026-09-18 · 5 min read

Most of the operational advice aimed at small deployments assumes you have time to build dashboards. You do not. What you need is a short list of things that will definitely break and a way to notice before a user tells you.

Certificates renew, or they do not

The single most common self-inflicted outage I have caused is an expired certificate. Modern ACME clients renew automatically, so the failure mode is almost never the renewal itself. It is a certificate that got copied somewhere at setup time and then quietly stopped matching the one being renewed.

If two services need the same certificate, point both at the live file. If you must copy, make the copy part of a periodic job that compares checksums and reloads when they differ.

Log rotation is not optional

A journal that grows until the disk is full takes down everything at once, usually at three in the morning. Cap it:

# /etc/systemd/journald.conf
SystemMaxUse=500M
SystemMaxFileSize=50M

Two alerts are enough

Disk above 85%, and the service not responding. Everything else can wait until morning. Resist the urge to monitor CPU; on a small box it spikes constantly and the alerts teach you to ignore alerts.

Backups you have actually restored

A backup that has never been restored is a hypothesis. Once a quarter, restore one onto a throwaway instance and make sure it boots. It takes twenty minutes and it is the only way to know.