I Deleted Ten Months of Metrics Nobody Looked At
Grafana dashboards last edited in October. Admin last logged in in October. The database grew for ten months anyway, on the smallest disk in the fleet. Teardown as maintenance.
What broke, what I built, and what I’m still learning.
Newest first. The useful failures tend to age well.
Grafana dashboards last edited in October. Admin last logged in in October. The database grew for ten months anyway, on the smallest disk in the fleet. Teardown as maintenance.
I thought my smart home had 800 devices. An audit found 1,751 — and 29 of my 42 scripts had never run once. The fix wasn't more automation. It was less.
Bubbles runs on Hermes, lives on a Mac Mini, and has an iMessage number through Photon. The useful part isn't the chat—it's the skills that connect it to the tools I already use.
A container update killed my Cloudflare tunnel by pointing it at a Docker network that no longer existed. Nothing logged, nothing alerted. Four public services, dark for three days.
My stuff lives in two countries, three time zones, one closet, and a cloud droplet. Tailscale is the reason it behaves like one house — and the one piece of the stack I'd have to replace before anything else could work.
The family AI went silent and everything pointed at an API outage. The actual cause: a debug command I'd forgotten about, hung for 16 days, eating all 16,384 network ports.
A dozen self-hosted apps, each remembered as a port number. Three pieces closed it out: a wildcard domain resolving to a private address, genuine wildcard certificates issued to a server no certificate authority can visit, and a deploy that reconciles itself.
The AT&T gateway passed its speed test, but my own router could not get online. The fiber provider had authorized one network-port address while the cable was plugged into another.
A power outage took 840 smart-home entities offline with no error anyone saw. The cause was two services wanting one port and an appliance OS that re-enables its own daemon on every boot — so the fix isn't a config change, it's a boot-time reconciler.
Home-and-away automations were firing for people who weren't there. What worked was a second, independent presence signal built from something I already owned — the WiFi controller — plus an aggregation rule with one sharp edge worth understanding before you use it.
Most people name their NAS nas-01. Mine is Randy, from Trailer Park Boys, because the rule is simple: the character's job on the show is the server's job in the house.