Overheating

The only incident in the game whose cause is physical. Cooling on one host has failed, and no amount of software will persuade the metal to be cooler.

What it does

Temperature on one machine climbs past the WARN threshold at 85 °C. At 95 °C it hits thermal shutdown: the machine goes DOWN and stays there for a full sixty seconds — the longest forced outage of any incident.

Temperature is clamped to a 40–120 °C band, so this has a real ceiling rather than running away indefinitely. It does not need to run away; 95 °C is enough.

The log signature

The [hw] tag is unique to this incident, and the fan line is the tell — hardware reporting that it is already doing everything it can:

05:18:30 [hw]     web-03 temp 88°C — fan RPM max
05:22:14 [hw]     web-03 thermal shutdown at 95°C

If you leave it alone

Sixty seconds of guaranteed downtime on one machine, plus the SLA debt if it stays open. On a small early fleet, losing one of three web machines for a minute is a serious dent in the shift average.

Heat is often a symptom

This is the incident most likely to be a consequence. Sustained load raises temperature, so a machine running hot during a flood may be reacting to the flood rather than to a broken fan.

The way to tell: a genuine cooling fault reports under [hw] with the fan line and appears on one host without a matching load story. Heat that follows saturation arrives after the CPU is already pinned.

Two kinds of answer

There is a reversible way to take the load off a hot machine and a permanent paid repair. The reversible one lowers the temperature without fixing anything — undo it too early and the climb resumes.

When it shows up