Uptime that actually means something

“Uptime” is the most abused number in server hosting. If your monitor pings once a minute and counts anything that answers as “up”, a server that crashes and restarts every five minutes can still post a great percentage.

State, not a ping

Tickr runs a small state machine instead of a naive ping:

online → unreachable (3 failures) → down (5 more) → recovering → online

Debouncing means a single dropped query packet never flaps your status, and a genuine outage opens an incident with a start, end and cause.

Classifying the cause

When an incident closes, Tickr classifies it:

• crash: process exited non-zero, or a plugin’s clean-shutdown marker was missing.
• oom: RAM was pinned near the limit right before the drop (or the host node reported an OOM kill).
• restart: an expected, clean shutdown.
• network: the host was unreachable but the process never reported a crash.

A crash loop (3+ crashes in 15 minutes) fires its own alert, because that’s the failure that actually drives players away.

The point

A reliability number is only useful if it distinguishes “we restarted for an update” from “we fell over four times last night.” Classification is what turns an uptime badge into an operations tool.