Journal
What we learn from other people’s outages
Worked examples, product decisions and the occasional post-mortem of our own. Published when there is something worth reading.
- Bisecting a race condition that only happened on Tuesdays A worked example. Two recorded hours, one good and one bad, and the four commands that found the event where they parted.
- The hour nobody wrote down Incident reviews run on memory, and memory is rebuilt at the moment someone asks for it. That is the reason this product exists.
- What a replay is not A replay is a fold over a log. It is not a screen recording, not a load test, and not a promise about anything we did not record.