Notes on small systems
I keep a handful of services running for a few dozen people. These are notes to myself about the unglamorous parts: budgets, caches, dashboards, and the paperwork that keeps a bad afternoon from becoming a bad week.
Recent notes
-
Latency budgets are a budgeting problem
A p99 target that nobody spends is a wish, not a target. Splitting the number up front makes the trade-offs visible.
-
Cache invalidation is mostly a naming problem
Most “stale cache” incidents are really cases of two different things sharing one key.
-
Small services, boring dashboards
A dashboard should answer one question quickly. Four panels are usually three too many.
-
A postmortem template people actually fill in
Short sections, no blame, and a place to write down what we decided not to fix.
What this is
A static notebook. No accounts, no analytics, no comments. Pages are plain HTML served by nginx; the whole site is a few dozen kilobytes and builds with a shell script.