Running services, network usage, memory usage, bandwidth, disk I/O, successful logins, whether the thing is even alive, etc…
So far my only method has been “hope and pray”.
You guys are monitoring your servers?
“can’t access this thing, server must be down”
I’ll fix it when I get home
I have a robust monitoring system for my Jellyfin server, been running for the last few years. Checks in periodically, at least once per 24 hours and notifies me if it’s down. Doesn’t use any electricity, but does consume a good amount of Cheerios and mac & cheese.
Lies! It uses electricity!
Indirectly. But as another commenter pointed out, it balances out with the tax rebates.
Does this monitoring system also happen to be a dependent on your taxes?
They’re in my attached garage. In the living room when it’s quiet I can hear them humming. Does that count?
Uptime Kuma! For HDDs I have them report their “ping” as percentage full
Can you teach me this wizardry?
Sure! In Uptime Kuma, add a monitor with type
Push. It gives you a URL with a unique token, and the monitor goes into “down” state if nothing calls that URL within the heartbeat interval. Then have a cron (or a systemd timer or whatever) run this on the machine with the drive every 5 or 10 minutes:use=$(df --output=pcent /mnt/yourdrive | tail -1 | tr -dc ‘0-9’) curl -fsS -G “http://your-kuma:3001/api/push/TOKEN --data-urlencode “status=up” --data-urlencode “msg=${use}% used” --data-urlencode “ping=${use}”Pingis of course meant to be response time, but Kuma will graph any number you toss in, so you get a “filled” chart for the drive. I also sendstatus=downonce usage passes a threshold. And if the box itself should fall over for some reason, the missing check-ins trigger the alert too.(I typed this quick on my phone, so please double-check and probably don’t blindly copy-paste)
SNMP.
The old engine was nagios. The new engine will be telegraf -> mqtt -> brokerfest -> timescale -> Prometheus.
I guess. Not sure yet. The brokerfest is a series of broker pubsub between sites to both split out data from the stream for off-site CC, or pull its own subscriptions in. Data could go host <- telegraf -> broker -> remote broker -> remote timescale -> remote Prometheus






