Incident Response
The hour production is broken, and the write-up that decides whether it happens again.
5 topics
On-call & alerting
Every page should be urgent, actionable, and about something a user can feel — everything else is a dashboard or a ticket.
core22 minTriage, severity & roles
Declare early, assign an incident commander separate from the person debugging, and communicate on a fixed cadence.
core20 minMitigate before you diagnose
Stopping user pain and understanding the cause are different jobs, and doing them in the wrong order is what turns a ten-minute blip into an afternoon.
core18 minDebugging production systems
Form a hypothesis, find the change, bisect the request path with logs, metrics and traces, and never trust a story you haven't confirmed with data.
core22 minBlameless postmortems
The point is to find the conditions that let a competent person make that mistake, and to leave with owned, prioritised actions rather than a promise to be careful.
core22 min