beefcake: hold impermanence flip OFF pending dbus-loop fix #725

Closed

lytedev wants to merge 1 commit from hold-impermanence-flip into main

lytedev commented

2026-07-08 15:51:24 -05:00

Owner

URGENT — merge to make main safe to deploy to beefcake again.

The 2026-07-08 impermanence activation booted and CORRECTLY cleared initrd (the @blank rollback + /persist mount both worked — the risky mechanism is sound). But the full beefcake service graph on the ephemeral root sent dbus-broker into a crash-loop, which blocked NetworkManager → no network → boot never completed (~20min dark). Rolled back to gen 606 via the systemd-boot menu (old persistent root); beefcake is healthy, all core services up, boot default reset to 606.

This sets lyte.impermanence.enable = false so main can't redeploy the broken generation (the build-tier gate passes it — it's a runtime/boot failure, not a build failure). Everything else stays: the module, runbook, prototypes, and the staged datasets (@blank + rpool/persist, ~68G) for the retry.

Root cause + fix tracked separately: reproduce the dbus loop in an impermanence VM carrying beefcake's real dbus/NM surface (the minimal semantics prototype missed it — exactly the 'full-closure boot' gap the design doc flagged as untestable at dragon-scale), fix, prove, then re-attempt the flip.

🤖 Generated with Claude Code

https://claude.ai/code/session_01G7HHrn1sxz4ozvREyHmi4i

URGENT — merge to make main safe to deploy to beefcake again. The 2026-07-08 impermanence activation booted and CORRECTLY cleared initrd (the @blank rollback + /persist mount both worked — the risky mechanism is sound). But the full beefcake service graph on the ephemeral root sent dbus-broker into a crash-loop, which blocked NetworkManager → no network → boot never completed (~20min dark). Rolled back to gen 606 via the systemd-boot menu (old persistent root); beefcake is healthy, all core services up, boot default reset to 606. This sets lyte.impermanence.enable = false so main can't redeploy the broken generation (the build-tier gate passes it — it's a runtime/boot failure, not a build failure). Everything else stays: the module, runbook, prototypes, and the staged datasets (@blank + rpool/persist, ~68G) for the retry. Root cause + fix tracked separately: reproduce the dbus loop in an impermanence VM carrying beefcake's real dbus/NM surface (the minimal semantics prototype missed it — exactly the 'full-closure boot' gap the design doc flagged as untestable at dragon-scale), fix, prove, then re-attempt the flip. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01G7HHrn1sxz4ozvREyHmi4i

lytedev added 1 commit

2026-07-08 15:51:24 -05:00

revert(beefcake): hold impermanence flip OFF pending dbus-loop fix

/ check-format (push) Successful in 7s

Details

/ build (push) Successful in 38m17s

Details

24c9170d15

The 2026-07-08 activation booted and correctly cleared initrd (blank
rollback + /persist mount worked), but the full beefcake service graph on
the ephemeral root crash-looped dbus-broker, which blocked NetworkManager
-> no network -> boot never completed. Rolled back to gen 606 via the boot
menu (old persistent root); beefcake healthy, boot default reset to 606.

Set lyte.impermanence.enable = false so main can't redeploy the broken
generation. Module, runbook, prototypes, and the staged datasets (@blank,
rpool/persist) all stay for the retry once the dbus loop is reproduced and
fixed in a VM (the minimal semantics prototype didn't exercise dbus/NM —
the full-closure-boot gap flagged in the design doc).

lytedev referenced this pull request from a commit

2026-07-08 16:32:55 -05:00

fix(beefcake/impermanence): seed machine-id in initrd — fixes the dbus crash-loop

lytedev referenced this pull request

2026-07-08 16:33:11 -05:00

beefcake/impermanence: seed machine-id in initrd (fixes the dbus crash-loop) #726

lytedev closed this pull request

2026-07-08 16:34:59 -05:00