Why I keep a log
1 min read#meta
I build things at a few different depths. A key-value store and a WebSocket engine in Rust. Firmware for ESP32 and STM32 boards. The Linux servers some of that eventually runs on.
The interesting part of each one rarely makes it into a README. A README says what a thing does. It doesn't say why the first design was wrong, which measurement changed my mind, or what the oscilloscope showed at two in the morning.
That is what this log is for.
What goes here
- Storage and durability. Write-ahead logs, crash recovery, and what "durable" costs in practice.
- Realtime systems. Presence, ordering, fan-out, and the gap between a demo and something multi-tenant.
- Firmware and signal. Latency budgets, DMA buffers, filters, and radio links that almost work.
- Operations. The incident, the cause, and the change that stops it happening again.
The rule
Every entry has to contain something I measured or something I broke. If it has neither, it isn't finished.
A storage engine keeps a write-ahead log so it can rebuild its state after a crash. This is mine.