SCRATCH — radagast overnight status (2026-09-09)

Scratch note — disposable, overwrite freely. Full project detail/history: hosts/radagast/source-upgrade.

Outcome: buildworld114 FAILED, VM powered off on schedule

• Pipeline launched ~21:34 on 2026-09-08 to run the rest of the 11.4 hop (buildworld → 8-vCPU restart → buildkernel → snapshot → installkernel → reboot → installworld → reboot) unattended overnight, with a hard deadline: disks free by 06:00, host hibernated by 06:15 (John travelling to London, train 07:00).
• buildworld for 11.4 (/usr/src12, from archive.freebsd.org's 11.4-RELEASE source) FAILED with a real build error at 01:31:07 — NOT a deadline timeout. The pipeline correctly detected this (distinguishes real failures from aborts) and did not attempt any further stages.
• Exact error not yet known — it's in /root/buildworld114.log on radagast's own filesystem, which is only reachable by booting the VM. Deliberately NOT done overnight, since the disks needed to come free by 06:00 regardless of build outcome.
• At the 06:00 hard deadline the pipeline ran its safety poweroff sequence (shutdown -p now inside the guest, waited for the qemu process to actually exit) — confirmed VM fully powered off at 06:00:20. Disks (sda/sdb) are physically safe to remove.
• Host hibernate scheduled for 06:15 via the same script (systemctl hibernate), separate from the VM poweroff.

What's safe / what's not lost

Nothing on zroot is at risk: the failure was in buildworld, which only ever writes into the disposable zroot/usr/obj dataset. The system is exactly where it was after the successful 11.1 hop — fully matched, bootable 11.1-STABLE world+kernel, snapshots zroot@post-11.1-hop / zroot/usr@post-11.1-hop available if ever needed. The 8-vCPU restart, buildkernel, the pre-11.4-world snapshot, installkernel, and installworld were never reached.

Next steps (when radagast's disks are back)

• Boot the VM (start-freebsd-zroot-vm.sh, now requests 8 vCPUs), check /root/buildworld114.log on the guest for the actual compiler/linker error around 01:31:07.
• Given the libc++ header-corruption bug already found and fixed for the 11.1 hop, check first whether a similar or new header/toolchain inconsistency reappeared for the 11.4-era compiler bootstrap before assuming it's a new class of problem.
• Once diagnosed and fixed, zroot/usr/obj will likely need wiping and buildworld restarted for 11.4 — not necessarily from absolute zero if the failure was late enough that make's dependency tracking can resume, per John's own past experience doing exactly that.