23:41, 15 October 2026. The power goes out. Not unusual — it happens a few times a year in our area, usually for seconds. The home server where part of our own systems and client sites live sits on a UPS. The UPS holds for a bit, power doesn't come back fast enough, the machine shuts down.
Power returns at 23:58. The server boots itself — that's what the UPS and a BIOS setting are for. But the virtual machine that actually hosts the sites doesn't start. Because nobody had ever ticked "start automatically when the host boots." A small thing. The kind of small thing you never notice until this exact night.
At 00:10 power drops a second time. At 00:26 it's back for good. The host comes up on its own. The virtual machine — again, doesn't.
We run a watcher that checks whether the sites respond and pushes an alert to Telegram if something's wrong. It's worked fine for months. Except that watcher lived on the same machine that just went dark. The guard and the alarm were in the same house — when the house went dark, so did the alarm.
We found out by accident, around 01:38, when one of us opened a phone for an unrelated reason and noticed a page wasn't loading. From the first outage to the moment we actually knew, roughly 1 hour and 50 minutes passed. That's how long the sites sat dark with nobody — not us, not any automated system — aware of it.
| Time | What happened |
|---|---|
| 23:41 | First power cut |
| 23:58 | Power returns, host boots itself, the VM doesn't |
| 00:10 | Second power cut |
| 00:26 | Power stable for good, same problem repeats |
| 01:38 | We notice by accident, on a phone |
| ~01:50 | Sites manually restored |
Times are reconstructed from logs and our own memory of that night — we won't claim second-by-second accuracy, but the sequence and total length are accurate.
A guard that lives in the house it's guarding isn't really a guard. If your alerting depends on the same power, the same router, the same cloud region as the thing it's supposed to watch, it tells you nothing precisely when you need it to tell you something. This applies equally to a home server in a living room and to "production" sitting in a cloud virtual machine — if the monitoring sits in the same account, same region, same provider, it will go quiet exactly when you need it to speak.
The next morning we stopped regular work and built five things. Not because that night was a catastrophe — it was two quiet hours, no client lost a payment or a booking in that window — but because next time the timing might be different, a weekday afternoon with real orders moving through the system.
We won't sell this as magic. Two trade-offs remain, and we're saying them plainly:
For a brochure site with no bookings or orders, two quiet night hours cost nothing. For a system handling payments, shift bookings, or live orders, the same two hours cost real money and trust. We judge this honestly — not every project needs a mirror in another country. A small share of our clients genuinely do.
You don't need to build a system like ours to check your own exposure. Three questions for whoever runs your hosting or site maintenance, whatever that setup looks like:
If the answers are vague, or "we haven't thought about it" — you're probably not alone. Most small systems have no outside watchman, because nobody has had a night like ours to make them think about it.
What we built for ourselves is now available as an option for client systems that want it, through our control panel MAESTRO: a mirrored standby, an outside watchman, automatic failover. It's not default on every project — for most sites and small systems it would be an unnecessary expense. For a system where two quiet night hours cost real money, it's cheap insurance.
If your own system has gone dark some night and nobody called you — describe the situation in one sentence to the live agent at pentimento.pro and see what a fix could look like. Or just send us the question — it might become the next article.
PENTIMENTO