1.7 KiB
Disposition — backups-IA finding 2 (MEDIUM): legacy-boot SystemDisks → dead drive wizard
Origin: the v0.124.0 backups-IA validation (2026-07-13; recorded in felhom-controller
CHANGELOG v0.124.0 + the drill live run): on a legacy-boot PVE (LVM root, no mounted ESP) the
agent's SystemDisks resolved no raw system disk → sysKnown=false → every disk classified
system → the drive wizard could never offer a candidate; the hot-added 5 GB scsi disk stayed
invisible on the drill box.
Disposition: RESOLVED 2026-07-13 — felhom-agent v0.87.0 (repo felhom-agent, commit
3c174bc; operator ruling approved 2026-07-13). The root filesystem's backing device is walked
recursively through /sys/block/<dev>/slaves down to physical disks (dm AND md; topology,
never VG names); those + any mounted-ESP holder are system. Per-branch conservatism: any
unresolvable slave fails the WHOLE walk → the unchanged all-system fail-safe — which is back to
being the walk-FAILURE error case, not the legacy-boot common case. The catastrophic direction
(a root-backing disk offered as a candidate) is pinned by the signature test
TestSystemDisks_WalkTopologies (never weaken) + red-proofs A/B/D.
Live evidence (2026-07-13): drill host (legacy shape) on 0.87.0 — the exact IA replay:
hot-added 5 GB scsi disk → GET /disks/candidates now returns it as an initialize candidate
(/dev/sdb, QEMU HARDDISK, not data-bearing); the root disk sda is NOT offered. felhom-pve
(EFI+LVM shape) regression: /disks roles + candidates byte-identical pre/post upgrade.
Remaining supervised leg: wizard enroll+format through the customer UI (operator session,
GO before format) + Day-0 manifest vouch of agent 0.87.0.