Field notes
Every note here comes from a real production failure — what happened, why it was hard to see, and the check that now makes it impossible to repeat. Notes are maintained rather than archived, so each one carries the date it was last reviewed.
2 notes
A CDN answered DNS with two addresses, one of which was blackholed by Egyptian networks. Browsers picked at random, so roughly every other visit stalled about twenty-one seconds before silently falling back. Every monitor, every curl, and every origin check looked perfectly healthy throughout.
A container that mounts one specific directory pins that directory at startup. Renaming it during a deploy leaves the container serving the old files forever, with no error and no downtime. Because a catch-all route answers 200 for URLs that do not exist, status-code checks confirm a deploy that never landed.
1 note
Setting overflow-x to hidden on html or body turns the document into a scroll container. Native lazy loading then measures against the wrong root, and in Safari the images below the fold never load at all. Chrome usually hides the problem, so it looks like an Apple bug. It is not.
Why these exist. Every one of these cost real time to diagnose, and each ended in a permanent change to how I build — usually an automated check rather than a remembered caution. Publishing them is the same instinct: a lesson that lives in one person's head is a lesson waiting to be relearned.