Infrastructure notes
Short write-ups on the boring parts of running small services — backups that actually restore, certificates that renew themselves, and logs nobody reads until they matter.
Recent
Notes are published irregularly. Most of them start as an incident postmortem and get cleaned up later, which is why the dates are uneven.
Certificate rotation
Why the renewal hook mattered more than the renewal itself.
Log retention
Keeping 30 days of logs on a disk sized for 7.
Restore drills
A backup you have never restored is a hypothesis.
Time drift
Two servers, one NTP source, three different clocks.
Conventions
- Configuration lives in version control; secrets do not.
- Every scheduled job writes a line on success, not only on failure.
- Anything that runs from
crongets a timeout. - Health checks test the dependency, not the process.
About
A personal notebook, kept public mostly so the author can find things again from a different machine. No tracking, no analytics, no comment section.