This is a personal notebook about running and maintaining Linux infrastructure: hardware that sits in a rack and forgets how to retire, configurations that drift, and the occasional 3 a.m. haircut when something gives out. Written slowly, so it will probably keep making sense in a year.
Most replication problems are not replication problems. They are assumptions with a timestamp attached. A field report from a migration that finished with zero re-uploads — and what the checklist missed on the first pass.
Every hand-rolled change on a production box is a commit that never happened. Why drift is cheaper to prevent than to explain, and how a stroll around the configs once a week beats an audit in the worst moment possible.