The CONTROLLER_IMAGE default (arg 6) was a stale :0.43.0, so an argument-less
golden build baked an ancient controller (fresh Day-0 boxes started at 0.77).
Bumped the default to the current :0.85.1; always pass it explicitly per rebuild.
Golden rebuilt at 0.85.1 on felhom-pve (volid vzdump-lxc-9100-2026_06_27-11_42_51);
baked-image verify confirmed :0.85.1 in the build guest. No agent binary change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FSZmmSFVzGwEzhYmxbkgBK
New local-API POST /controller/swap (+ GET /controller/swap/status), withGuest-
scoped: the agent records the previous image, confirms the target is present,
rewrites /etc/felhom-controller-image, restarts felhom-controller-bootstrap.service,
verifies the new controller is healthy (docker inspect, <=90s), and ROLLS BACK to
the previous image if not. Single-flight per guest; strict image-ref validation;
crash-safety state file. GuestBinder.GuestExec is the pct-exec seam.
--selftest=controller-swap exercises it directly.
Tests: happy/rollback-on-unhealthy(+red-proof)/image-absent/no-healthcheck/
bad-image-400/single-flight-409.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TtXesNa2LGbMmE4DNL6SE7
Destroyed + re-provisioned 9201 through the real provision chain with v0.41.0: fresh pct config
showed onboot:1 with no manual set. Host reboot of felhom-pve -> 9201 auto-started (running, no
manual pct start); onboot:0 scratch guests stayed stopped; drives re-bound at stable, sys_drive
separate, controller + infra healthy. Demo controller redeployed to v0.82.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FpBYrZCt9sFDqLgbG5GRGD
The golden bakes --onboot 0 (template safety) and the back-half never overrode it, so every
provisioned customer guest was onboot:0 -> after a host reboot/power-cut the customer's whole
home-server stayed stopped until a manual pct start. Add a fatal 'pct set <vmid> -onboot 1' step
to BackHalf.Provision (right after the config-mount attach), mirroring the existing pct set ops.
No startup/boot-order: the v0.75 mountpoint-gate covers the drive-bind race at boot.
Golden build-golden.sh unchanged (templates must not auto-start). Unit-tested
(TestProvision_SetsOnbootOne + red-proof). RUNBOOK note added.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FpBYrZCt9sFDqLgbG5GRGD
Extends the OS/Docker-data split to a three-volume layout: rootfs + Docker-data
(mp0) + SSD user-data (mp1 @ /mnt/sys_drive, backup=1) = the controller's
system_data_path. Clears the controller's "not a separate drive" warning with
zero controller change (it already auto-discovers <sys_drive>/felhom-data and
warns via system.IsMountPoint; the mp reaches the container via the existing
-v /mnt:/mnt:rslave bind).
- build-golden.sh: --mp1 ...,mp=/mnt/sys_drive,backup=1 (env GOLDEN_SYSDATA_GB=8);
findmnt /mnt/sys_drive separate-mount guard + vzdump aborts if mp0 OR mp1 excluded.
- bringup.go: DefaultSysDataMount=mp1; BringUpSpec.{SysDataGrowGB,SysDataMount};
new "4c" online grow-only block mirroring the "4b" Docker-data grow.
- main.go: -sysdata-grow / -sysdata-mount flags wired into all three call sites.
- Tests: SysDataGrow (asserts ResizeLXC mp1 +42G) + SysDataGrowZeroNoResize.
- RUNBOOK extended to the three-volume layout (32 rootfs + 200 docker + 50 user-data).
Static CT volume, NOT an enrolled drive — never enrolls/ejects/decommissions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017PsnU2ASocYrvzqE82YDYW
Live PBS coord: new internal/pbs/live_reporter.go (LiveSnapshotReporter implements
hub.PBSReporter via the cheap Client.Snapshots() list with last-known-good fallback,
bounded by an 8s timeout, list-only — never triggers a verify). Closes the gap where
the recipe's pbs block was omitted whenever the verify-loop SnapshotStore was empty
(one-shot collect + the first ~6h after a daemon restart). SnapshotStore.Get added
(per-datastore LKG). Wired into the collector in both runDaemon and runSelftestHub;
the verify loop keeps Recording into the SAME shared store via one hoisted pbsTargets.
v1 host-half drive shape: dropped drives[].role (hub/operator-owned manifest concept,
not host-derivable) and drives[].restic_repo_coord (named a backup tier that doesn't
exist). Drive shape is now {durable_id, mount_path, intent, fs_type?, total_bytes}.
Hub reads drives as json.RawMessage → no hub struct change; goldens re-pinned
byte-identical (agent + hub copies).
Tests: live_reporter_test.go (T1 load-bearing coord-without-verify + T2..T6),
TestDRRecipeHostHalf_V1DriveShape; each companion demonstrated to fail pre-fix then
reverted. go build/vet/test green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
DR recipe slice (agent half), grounded in SPIKE-dr-recipe-2026-06-16. Additive
`dr_recipe` host-report section = the non-secret reconstruction scaffolding the
operator must rebuild before PBS bytes can land. Built by pure
BuildDRRecipeHostHalf from facts the report already collects (no new reads):
guests[] sizing, drives[] (user-data by durable_id/role/mount/intent),
pve_storage[] (storage.cfg), pbs coordinates.
BOUNDARY (Phase-1 lesson): every field is an identifier/intent/size/coordinate —
never a key/password/token/hash/ENC:. PBS key stays in escrow; restic password
stays in escrow; the recipe names only the coordinates the restore targets.
Tests: BuildDRRecipeHostHalf selection, NoPBS, NoSecrets (boundary mirror),
dr_recipe key-set in the cross-repo golden contract test. recipe_version=1,
ignore-unknown on read.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The skip-if-mounted optimization defeated the actual root cause: felhom-usb is
mounted now but its unit is `disabled`, so a host reboot would not auto-mount
it. ReassertEnrolledMounts now skips ONLY the durable steady state (mounted AND
enabled) via the pure shouldReassertMount; a mounted-but-disabled unit is
re-asserted so enable --now re-creates the wants-symlink. Enabled-state read by
privilege-free Lstat of the multi-user.target.wants symlink (unitEnabled) — no
systemctl is-enabled subprocess, no new sudoers entry.
Tests: TestShouldReassertMount (4 combos), TestUnitEnabled (wants-symlink).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
TASK A — close out the reboot story (agent half). Root cause (pinned live on
felhom-pve): an enrolled .mount unit left `disabled` by a prior detach never
auto-mounts at boot, and kernel re-enumeration can move a drive's node
(/dev/sdb->sdc). Fix re-asserts every enrolled mount by filesystem UUID at
startup + on the periodic tick.
- ResolveStorageDevice: resolve uuid:<fs-uuid> -> current /dev node via
/dev/disk/by-uuid (never a cached node); errors if absent.
- parseFelhomMountUnit: pure inverse of renderMountUnit (marker-gated).
- (*SudoHostOps).ReassertEnrolledMounts: re-run EnsureMount (enable --now) for
any enrolled unit not in /proc/mounts; idempotent, skips mounted/absent.
- main.go: runs before ReassertGuestBinds at startup + on the 20s tick.
- tests (Linux, seam=device resolution): letter-move tolerated (sdb->sdc) +
red-proof companion, absent/scheme rejection, render->parse round-trip.
TASK A2 verdict: enrolling a NEW drive does NOT need an LXC restart — the path
lands on the live AttachDrive (shared parent, named live slots), not RebootGuest.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
make-private+make-shared only when first creating the self-bind; re-running it
churns the peer group and orphans the guest's slave (propagation dies).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The self-bind inherited /'s shared peer group, so binds under /mnt/felhom-drives
propagated back and doubled. make-private (own group) before make-shared fixes it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
countHostMounts + normalize: no-op only when exactly one bind is guest-visible;
else strip all binds and lay one fresh. Converges a stacked double-bind to one
(the old umount-one+mount-one never did).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A GuestBinder mutex prevents a concurrent reconnect + periodic reconcile from
both passing isHostMountpoint and double-binding a stable path.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Detach now removes ALL stacked binds at a stable path, not just one layer, so an
eject fully detaches even with >1 bind (keeping fail-close intact).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Eject now DetachDrive's the bind under the parent but leaves the raw /mnt/<name>
mounted, so disconnect->reconnect re-binds on a non-removable drive. Tests updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The decommission unmounted the raw /mnt/<name>, orphaning a non-removable drive
so re-enroll bound an empty dir. Now DetachDrive only (bind under parent); raw
stays mounted so re-enroll re-binds. Test: raw NOT unmounted + DetachDrive called.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
GET /disks emits guest_boot_id = <host-btime>-<guest-init-starttime>: changes on
every guest/host boot, stable across controller-only restarts. The controller
persists it + deterministically recreates drive-backed apps on change (replaces
the timed state-sample). Non-hollow parser test + companion.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The shared-parent oneshot never ran before pve-guests (WantedBy=multi-user
wasn't pulled into the boot transaction), so on a host reboot the guest bound a
not-yet-shared parent -> private -> propagation broken. Now WantedBy=pve-guests
(pve-guests pulls it in + Before= orders it first); EnsureSharedParent reinstalls
the unit when content differs so the fix deploys.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A guest's parent bind is non-recursive, so a guest reboot leaves enrolled drives
bound on the HOST but invisible in the fresh guest ns (propagation only delivers
new events). AttachDrive(vmid) now checks GuestSeesMount (/proc/<pid>/mountinfo)
and force re-binds (umount+mount) to re-propagate; a 20s periodic reconcile
self-heals guest reboots without an agent restart; BoundUnderParent reflects guest
visibility (the controller gate's signal). Caught + fixed in the live migration.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replaces the per-drive 'pct set -mpN' bind with ONE permanent parent bind
/mnt/felhom-drives plus host-side felhom-data swaps underneath it (propagates
into the running guest live, no pct, no reboot; C1-immune; confined; fail-closed
when absent). EnsureSharedParent installs a boot unit ordered Before=pve-guests.
ReassertGuestBinds is now a pure host-side reconcile. /disks reports GuestPath +
BoundUnderParent for the controller repoint+gate. Non-hollow tests + companions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Pre-start PVE hookscript (internal/guesthook) creates host-root placeholders for
absent bind-mount sources so the guest always boots (fail-closed); decommission
now pct set --delete's the dead mp (GuestBinder.DetachBind) so a missing source
can't brick the next reboot (B3 C1 bug). Non-hollow tests + companions. Installed
+ registered per-guest by the provision back-half. Transitional ahead of the
intermediary-mount re-architecture which makes C1 structural.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
POST /disks/decommission mirrors eject (withGuest, user-data role gate) — no
operator signature, non-destructive (never formats): sets IntentDecommissioned,
prunes the GuestBindStore entry, unmounts. ReassertGuestBinds is now intent-aware
(skip non-enrolled) so a decommissioned-but-present drive never auto-rebinds on
agent restart — the load-bearing F9-reconnect fix. GuestBindStore.Remove added.
Operator-signed DecommissionExecutor + classify untouched. Non-hollow tests incl.
the intent-aware reassert companion (mutation-proven to fail on intent-blind code).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per the F9 storage episode — live validation of a user-facing feature
must exercise the real flow end-to-end, not shortcut via direct
API/agent/CLI calls. Low-level mechanism tests are exempt.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The format ran mkfs under the HTTP request context, so the controller's 15s client
timeout cancelled it → SIGKILL mid-write → corrupt disk. Now mkfs runs DETACHED off
s.baseCtx (a dropped request can't kill it) via a persisted formatJob record; the handler
still waits to return the synchronous result (backward-compatible with the v0.62.0
controller) but abandoning the wait on client-disconnect leaves the mkfs running to
completion. New GET /disks/format/status surfaces the job (additive). RecoverFormatJob
runs on agent startup: a record left 'running' (agent died mid-format) is re-resolved by
durable-id (anti-retarget — absent/swapped disk NOT re-formatted) and the mkfs re-run; a
blank/path-bound interrupted format is marked failed (retry), never auto-re-run.
Tests: detached run persists running→done + binds durable-id; status endpoint; recovery
re-runs an interrupted durable-id-bound format; skips blank; skips unresolvable durable-id.
Version 0.30.0 → 0.31.0.
The in-guest bind (pct set -mpN) is config state that a destroy+re-provision drops, and
nothing restored it — so a re-provisioned guest came up with its enrolled HDD unattached
(the live-drive F9 finding). New GuestBindStore persists, per guest, the durable-ids of
enrolled drives (recorded at guest-attach); ReassertGuestBinds runs on agent startup (the
host's bring-up/reconcile trigger) and re-adds any bind a guest is MISSING — but ONLY when
the durable-id still resolves to a present, mounted drive (a swapped/absent drive is never
auto-bound) and the guest lacks it (idempotent). The re-added bind activates on the guest's
next reboot, like the enroll flow. Wired in main.go (store opened beside drive-intents.json;
ReassertGuestBinds called before the local API serves).
Tests: restores a missing bind with no manual call (the operator's real-trigger proof);
skips absent/swapped durable-id; no-op when already bound; store survives reopen (restart).
F20-BUG2: the /disks list only carried DurableID in the uuid: scheme (for /disks/assign),
but the wipe gate resolves devices in the byid:/byuuid: scheme — so a customer confirming a
wipe with the advertised id was refused (binding_mismatch). Added a shared s.deviceDurableID
seam used by BOTH handleDisks (new DiskInfo.WipeDurableID) and the format gate, so the id the
customer copies from the list is exactly the id the gate accepts. DurableID (uuid:) is unchanged
(still feeds assign).
F9 (reporting half): added DiskInfo.GuestAttached — whether the drive's namespace is actually
bound into THIS guest's config (guestBoundPaths), distinct from mere host presence (State). This
is the signal whose absence made the HDD look available when it wasn't attached, and resolves the
F2 hdd_configured-vs-/disks disagreement.
Tests: wipe_durable_id is the gate scheme + distinct from uuid:; the list's wipe id matches the
gate's device-id binding (no mismatch); guest_attached true iff bound into the guest.
Expands the 'push to main; no feature branches' convention into the full rule:
report-only artifacts → felhom.eu/documentation/; risky/supervised (agent/golden/
provisioning/destructive) fixes implemented on main during the supervised session,
not prepared on a branch; unattended escape hatch = revert+report. Supersedes the old
'prepared on fix/... branch, pending review' pattern.
Closes the inline customer-confirmed wipe TOCTOU: handleDiskFormat now formats
the re-resolved device (durable id -> current device, re-derive+match, re-inspect)
via antiRetargetResolve/reresolveWipe, at parity with signedjobs.WipeExecutor.
Reviewed + approved. Agent-only host-daemon deploy (no golden rebake).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
handleDiskFormat's customer-confirmed branch formatted the mutable req.Device
path; the durable id only bound the confirmation, never the mkfs target. A /dev
reassignment between inspect and mkfs could wipe the wrong physical disk.
Now mirrors signedjobs.WipeExecutor: resolve confirmed durable id -> current
device, re-derive + require exact match, re-inspect (still data-bearing), then
format THAT device. Any refusal -> 409, no mkfs. New antiRetargetResolve helper
(injected deps, unit-tested: mismatch/gone/blank/empty all refuse). Injectable
reresolveWipe seam on Server (defaults to real storage funcs).
BRANCH ONLY — pending supervised review/deploy (see AGENT-001-FIX-NOTES.md).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After a guest DHCP IP move, the split-horizon resolver kept serving the old IP:
the drop-in (address=/domain/ip) updated but 'systemctl reload dnsmasq' (SIGHUP)
does NOT re-read /etc/dnsmasq.d config — only /etc/hosts + cache. Changed
reload() -> restartDnsmasq() so address= changes actually take effect.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
docker-ce auto-starts on install with the default config; only a restart picks
up the just-written daemon.json. Same start-vs-restart trap the spike flagged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Validation caught that Docker 29's default containerd-snapshotter keeps the
image store at /var/lib/containerd (on the rootfs), so mounting the data volume
at /var/lib/docker only moved named volumes — images (1.2G) stayed on the
rootfs, defeating the split. overlay2 stores images+overlay+volumes under
data-root = the data volume, which the split and the controller's statfs(/)
guard both require. Golden daemon.json now sets features.containerd-snapshotter
false + a driver guard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
build-golden.sh bakes a small OS rootfs + a dedicated /var/lib/docker volume
(mp0, backup=1) carrying the baked images, plus Docker log rotation. bringup.go
grows the golden's data volume to the per-customer target (DataVolGrowGB) and
emits backup=1 on data mounts (GuestMount.Backup) — closing the spike-B3 silent
DB-loss trap. CLI gains -rootfs-grow/-datavol-grow/-datavol-mount. New
RUNBOOK-provisioning-storage.md. Phase 2 = felhom-controller v0.58.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
admin