- Shipped
- September 23, 2026 at 2:11 PM UTC
- Author
- Kamo
- Commit
- 89eaa9b
prepare-sleep writes the swap file and the resume wiring, then reports readiness — and readiness reads `resume=` from /proc/cmdline, which the kernel fixed at boot. Writing grub.d and running update-grub therefore cannot make the RUNNING kernel resumable. Readiness stays false until the machine restarts. Nothing restarts it. Sleeping is what powers a computer down, sleeping needs readiness, readiness needs the restart. A freshly provisioned computer sits at "cannot go to sleep" forever, holding its tier's memory on a node for as long as it exists, and the only sign is a lastError nobody has a reason to read. Observed on the first computer ever provisioned: {"kernelSupportsDisk": true, "swapLargeEnough": true, "resumeConfigured": false} swapBytes 3221221376, swapNeededBytes 2095474483 Two of the three preconditions already satisfied, the third unreachable by design. configure() now reports whether it wrote the wiring, and prepare-sleep restarts the machine once when it did. Safe at that moment specifically — prepare-sleep runs at the end of provisioning and before an idle sleep, both of which mean nobody is working — but checked rather than assumed with IDLE.rdp_sessions(), because it also runs after a kernel upgrade on a machine that may well be in use. The restart is detached and delayed three seconds so the caller gets its 200 first, the same reasoning sleep_now already uses for the power cut. A second run that changes nothing returns false and does not bounce the machine.