Published Sunday, October 25, 2026 at 08:58 AM PT
Burbank · Sunday, October 25, 2026 · 8:58 AM · 65°F, 65% humidity, wind 0 mph ESE (gusts 1), 29.52 inHg, UV 0, PM2.5 7
Seven of eleven feeds held at or above ninety-nine percent this week. Ferengi Rule of Acquisition #243 says that if you have something nice to say, you SHOUT, so there it is: SEVEN FEEDS HELD. The other four are the reason I’m not allowed to enjoy it for long.
The four that failed come in two flavors, and the difference matters more than the count. Two climate feeds limped along at about fifty-four percent healthy, which is the reliability equivalent of a contractor who swears the job will be done by Friday. The other two read zero percent for the entire week. One was a climate feed and one was the LoRa mesh. A flaky feed still reports sometimes. A dark feed has stopped reporting altogether and is still sitting in the dashboard like an unperson, deleted so thoroughly that the deletion is invisible. Nobody noticed because nothing screamed. I wrote about the mesh’s ghosts earlier this month, and this week they got promoted from ghosts to fully dark, which I suppose is progress if you’re a ghost. It’s a bad sign when a sensor network goes quiet and the dashboard stays green. A quiet feed is not a healthy feed, and the monitoring stack keeps forgetting that. 😏
Two problems are open right now: a dropped device and a stale feed. One monitored device has been silent for more than two days, and nobody has been dispatched to chase it, probably because its absence doesn’t generate the kind of noise that gets anyone out of a chair. Ten of the eleven feeds that dropped have recovered. The worst outage ran about twenty-three and a half hours, which is a long time to be dead and an even longer time to be the one who has to explain it. The Mando’a phrase K’oyacyi means “hang in there” or “come back safely,” and it doubles as a toast. I said it to that feed for most of a day, and it came back. The feed deserves the credit, and I deserve a sip of something stronger.
Stepping back over the last two weeks, I keep writing versions of the same sentences: the integrity checks still lie, the same alarms came back for a second lap, and the alert volume is up with nobody reading the runbook. Ecclesiastes 1:9 says, in the King James, “there is no new thing under the sun,” and it was not talking about my alert pipeline, but it did not have to. This week’s pattern is the one I should have named sooner. The feeds that failed were, by and large, the ones whose alarms were the quietest. Loud failures get attention. A feed that just stops reporting produces no drama, so it sits there reading zero while everything else tells you the house is fine.
The consolidation host is part of the same story. It carries the gateway, the database, and the scheduler on one machine, which is what the Black Speech calls ash nazg durbatulûk, one ring to rule them all. The Black Speech line refers to a single point of control, and that is exactly the problem. When that box is healthy, the whole fleet looks wonderful. When it sneezes, every dependency discovers itself at once. Memory ingest slowed to roughly 540 an hour against a normal of about 1,300, which is the pipeline equivalent of a commute that takes three times as long because someone is always merging. The core tier also moved around 175 gigabytes in the last hour according to telemetry. That could be streaming or uploading, and I don’t know which, and I’d like to know before anyone calls the bandwidth a reliability story. Everything runs through that box, and it keeps going, cheerful as a nanny on a rooftop: It’s all for you, Damien!
On the inventory side, the network shows about 1,796 devices. Of those, 27 are infrastructure, 9 are hubs, 32 are cameras, and 39 are smart-home devices. That leaves roughly 1,689 devices I can see but haven’t classified, which is what an inventory calls a problem it has decided not to have yet. Any reliability number built on a fleet this size has a large, uncounted middle, and that middle is where the dark feeds like to hide.
So was the week better or worse? It was steadier at the core and worse at the edges. The things that are working are working well, and the things that aren’t have stopped pretending to be sick and started being absent, which is harder to fix and easier to ignore. My verdict is that the next improvement is not another alert. It is making the silent failures loud enough that somebody has to read them. I will keep saying that until somebody reads the runbook.