`disk-forecaster/INSTALL.md`

23 lines
# Installing the Disk Runway Forecaster on Kamatera

The script runs locally for testing but its home is Kamatera (45.61.58.125),
where the TK-10236 ENOSPC crash happened. Installing a cron on the prod box is
a remote scheduled-job install — **staged for Steve's go**, not auto-run.

## One-time install (run in a live session or paste to Steve)
```bash
scp forecast-disk.sh root@45.61.58.125:/root/ops/forecast-disk.sh
ssh root@45.61.58.125 'chmod +x /root/ops/forecast-disk.sh && \
  ( crontab -l 2>/dev/null; echo "0 5 * * 1 ROOT=/root TOP=20 THRESHOLD_DAYS=14 GEORGE_BASIC_AUTH_PASS=<pass> /root/ops/forecast-disk.sh >> /root/ops/forecast-disk.log 2>&1" ) | crontab -'
```
Weekly (Mon 05:00). First 2 runs only build the baseline log; forecasts begin
once ≥2 time-separated samples exist.

## Reversible
`ssh root@45.61.58.125 'crontab -l | grep -v forecast-disk.sh | crontab -'`

## Note on George pass
The alert path needs `GEORGE_BASIC_AUTH_PASS` (inner George auth). Per memory
`george-gmail-auth-topology` it lives only in the pm2 process env — wire it in
during the live install, don't hardcode it here.