A disk does not fill overnight. Backups accumulate, logs rotate badly, a cache grows. Measured hourly, most disks fill at a steady rate for weeks. That rate is the whole trick: fit a line to the last seven days and you know the day it runs out.
1. Watch the worst volume, not the average
Every heartbeat carries each volume's used percentage and free space. The worst one is what breaks things. On a Proxmox host the pool matters, not the datasets, because every dataset reports the pool's free space.
2. Fit a line to the last week
Hourly rollups, least squares over 168 points. If the slope is under 0.05% a day, ignore it. Otherwise days to full is what is left divided by the slope. BE Hub shows it as "9d" on the fleet row and lists the soonest on the dashboard.
3. Alert at 90%, critical at 97%, and get a diagnosis with it
A threshold alert is a number. An alert with BE AI's assessment attached says what is eating the space, whether it is growing, and what to review first.
4. Fix it with an approval, not a login
List first, delete second, verify third. Each step is a proposal with the exact command; each approval carries a name; the result comes back in plain language.
5. Add the check that catches the sudden case
"Warn when D: has less than 50 GB free" becomes a free-space check that runs every minute on that machine, so a runaway log is caught even when the trend was flat.