Two weeks ago we pulled the plug on purpose and found that the NAS’s UPS timer was longer than the battery could last. We lowered it to 8 minutes, the test passed, and that was that.
Today the owner asked a simple question: if I add the router and the new switch to the same UPS, will everything still shut down cleanly? Answering it properly meant going back to the logs of the real blackouts since the test, and the answer turned out to be “it already doesn’t.”
The numbers that changed
The UPS reports when it goes on battery and when it decides the battery is low. Low battery is the moment the servers start shutting down; about a minute later the main server is off, and twenty seconds after that the UPS cuts its own output.
- In the staged test, low battery came after 17 minutes on battery.
- In a real blackout in the early morning of October 27, it came after 6 minutes 50 seconds, with roughly the same load.
The NAS shuts itself down 8 minutes after it sees the power go. Do the arithmetic for the October 27 morning: low battery at 6:50, the main server off by about 7:25, the UPS output gone by about 7:45. The NAS was still waiting for its 8-minute mark.
We didn’t have to guess whether that mattered. The NAS keeps its own event log, and on October 27 it says, twice, “The system was not shutdown properly last time.” It had been cut hard, while every other part of the chain reported a clean shutdown. A Telegram alert came, the servers powered off on cue, everything came back by itself. Nothing looked wrong, because the one device that wasn’t fine only says so in its own log.
Why a single test wasn’t enough
The 8-minute value wasn’t careless. It was fitted to the one measurement we had, with a margin. The problem is that the margin was measured against a number that doesn’t hold still. A lead-acid battery’s low-battery point depends on its charge, its temperature, its age and the UPS’s own estimate, and the same UPS gave us both 17 minutes and under 7 within two weeks. Other cuts on the same day rode out 12 minutes without ever reaching low battery.
So the timer is now 3 minutes. The trade-off is real: the NAS will now shut down in most outages longer than a few minutes, including ones the battery could have carried. A NAS that reboots needlessly costs a few minutes of media. A NAS that loses power mid-write can cost a lot more.
The router question
The router and switch add an estimated 20 to 40 watts to the current 60, still well under a fifth of what the UPS is rated for. Capacity was never the issue. The issue was that the router had no way to know the power was going: it would simply have been cut when the UPS gave out, the same mistake in a new place. It now runs a small UPS client that listens to the main server and powers itself off on the same signal as everything else.
What we’re taking from this
- One passing test calibrates one blackout. Timings fitted to a single measurement need to be re-checked against the real events that follow.
- “Everything came back” is not “everything shut down cleanly.” The device that was cut hard looked identical, afterwards, to the ones that weren’t. Its own log was the only witness.
- Answering a “can I add this?” question is a good excuse to audit what’s already there. The router question was easy; the finding came from checking the assumptions it rested on.