# The UPS That Lied

I trusted the battery for six years without a thought,
and learned what it was worth the day the power actually dropped.

The unit had a display and the display said it was fine.
It showed a full green bar and an estimate of nine

full minutes of runtime, which was plenty for the shutdown,
enough to close the databases and bring the whole thing down.

The power went at 6 AM. The runtime was eleven seconds.
The bar had measured voltage, not the thing that anyone reckons.

A lead-acid cell holds voltage well until it holds no charge.
The number on the front was true and uselessly at large.

The only honest measurement is drawing the real load,
which means unplugging from the wall and watching what it holds.

I had never run that test. I had read the panel instead,
and a panel is a story that the manufacturer said.

Here is the part against me, and I think it is a fair one:
pulling mains on a live rack to prove a battery is a rare one.

It risks the very outage that the battery was to stop,
and doing it on Friday is how weekends get chopped.

Do it anyway, but do it when the stakes are low and small,
and write the runtime down somewhere that outlives you on the wall.

What bites next is not the battery. It is the thing behind it.
The rack came back at 6:04 and half of it stayed quiet,

because the servers were configured to stay off after a fail,
and the setting had been sensible in 2010 without fail.