Four days with no internet: which self-hosted services kept working and which only looked local
Storm took the line out. Power and LAN were fine, so it became an accidental audit.
Kept working, no attention
- Media, books, photos. Anything reading local storage.
- Notes and documents syncing over the LAN.
- Recipes, wikis, anything that is really a database plus a web page.
Broke, and half surprised me
- Name resolution, first, before any service. Containers could not find each other because they resolved through an upstream that was gone.
- Anything checking a licence or account on start. Two things ran fine for two days and refused to come back after a restart.
- Dashboards pulling icons and fonts from a CDN. Worked, looked broken.
- Every notification path went outward, so the monitoring that would have told me what failed was failing silently.
The test worth doing this week
Unplug the line, then restart everything. Uptime hides the dependency; a restart reveals it. I would have declared the whole stack fine on day two and been wrong.
Fixes after: local records in the resolver, short forwarding timeout, pinned images, one notification path that stays in the house.
@dns_dies_first · 3w ago
Why resolution goes first: your filtering resolver forwards what it does not know, and when the line dies that path either errors fast or hangs. A hang is much worse, because every service waiting on a lookup waits with it and the cupboard looks dead when it is fine.
Give the resolver explicit local records, set a short forward timeout, then test by unplugging. Five minutes and the answer is not guessable.
Reply
Report