- Dikirim
- 23 September 2026 pukul 23.28 UTC
- Penulis
- Kamo
- Commit
- a607fea
Thirty seconds after the first real resume, from the machine's own log: kamo-hc-agent: idle 13202s >= 1800s — going to sleep PM: hibernation: hibernation entry last_active lives in this process's memory, and hibernation restores the process exactly as it was — so after a resume it holds a timestamp from before the sleep. The first tick compares now against hours ago, sails past the idle limit, and puts the machine back to sleep while the member is still waiting for it to open. Every wake, on every computer, for as long as it had been asleep. The wall clock is what gives a sleep away: it keeps running while the machine is off, so a gap between ticks far larger than the poll interval means we were unconscious rather than slow. Five minutes is the threshold — ten polls, far past any plausible slowness, and short enough that the briefest sleep still counts as one. And restarting the clock is not merely a repair, it is the correct reading: a resume happens because somebody asked for this computer, which is the opposite of idle. This could only appear once hibernation worked and something actually resumed, which is also why the number in the log is so large: 13202 seconds is the idleness it accumulated across the sleep itself. Two tests: a four-hour sleep must not sleep again on the tick it wakes on, and an ordinary 30-second poll must still leave the clock alone, so a slow tick can never excuse an idle computer.