admin
081c1cd77f
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017PsnU2ASocYrvzqE82YDYW
76 lines
5.6 KiB
Markdown
76 lines
5.6 KiB
Markdown
# REPORT — controller v0.78.0: storage register uses the STABLE path (not the raw path)
|
|
|
|
## Baseline
|
|
- felhom-controller `main` @ `66d8462`, **v0.77.0 → v0.78.0**. Trunk-based, `main`.
|
|
- Fixes the diagnosis `felhom.eu/documentation/audits/DIAGNOSE-drive-bind-after-reprovision-2026-06-23.md`
|
|
(§4/§5). Agent untouched (it already binds drives at the stable path).
|
|
|
|
## Problem
|
|
`handleStorageRegister` (the "Regisztrálás" action for an already-mounted, unregistered drive) registered
|
|
the **raw** `/mnt/<name>` path verbatim, while `runStorageInit`/`runStorageAttach` register the **stable**
|
|
intermediary path `/mnt/felhom-drives/<name>`. In the intermediary model the drive is live in the guest
|
|
only at the stable path, so the controller watched an empty placeholder dir on the guest rootfs →
|
|
"Rendszermeghajtón" badge (~31 GB), `0 connected / N disconnected`, and a stuck "activation pending"
|
|
banner. The real drives were fine the whole time.
|
|
|
|
## Fix (`controller/internal/web/storage_handlers.go`, `handleStorageRegister`)
|
|
- `stable := stablePathForName(path.Base(req.Where))` → `registerStoragePath(stable, …)` (was
|
|
`registerStoragePath(req.Where, …)`), mirroring `runStorageInit`/`runStorageAttach`.
|
|
- `attachIntoGuest(req.Where)` unchanged — the agent still operates on the **raw** path.
|
|
- Success `[INFO]` log + the `writeDiskJSON` payload now report the stable registered path (+ raw).
|
|
- No change to `runStorageInit`/`runStorageAttach`/`registerStoragePath` or any other behavior.
|
|
|
|
## Tests + red-proof
|
|
- `go build ./... && go vet ./... && go test ./...` — **all green** before and after.
|
|
- New `TestHandleStorageRegister_RegistersStablePath` (`storage_handlers_test.go`): drives the HTTP
|
|
handler with `where=/mnt/felhom-flash`, asserts the registry holds `/mnt/felhom-drives/felhom-flash`
|
|
and the JSON reports stable `where` + raw. PASS.
|
|
- **Red-proof:** reverted the handler to `registerStoragePath(req.Where, …)` → the test FAILED
|
|
(`got "/mnt/felhom-flash"`) → re-applied the fix → green. ✅
|
|
- Existing `TestRunStorageInit_Success` / `TestRunStorageAttach_Success` (which assert stable registration
|
|
+ raw guest-attach) kept green, unmodified.
|
|
- `internal/web` test count: +1 (the new register test). Full suite green.
|
|
|
|
## Deploy (v0.78.0 → guest 9201)
|
|
- Built + pushed `gitea.dooplex.hu/admin/felhom-controller:0.78.0` on 192.168.0.180 (`build.sh 0.78.0
|
|
--push`, 142M). Deployed to 9201: `docker pull` → `printf … > /etc/felhom-controller-image` → restart
|
|
`felhom-controller-bootstrap.service`. `docker ps` → `:0.78.0 Up (healthy)` (was `:0.77.0`).
|
|
|
|
## 9201 registry remediation (non-destructive — drive never unmounted)
|
|
The registry held the two stale RAW entries `/mnt/felhom-flash` + `/mnt/felhom-usb`. Remediated through
|
|
the real server pipeline:
|
|
1. Re-registered both via the FIXED `POST /api/storage/register` (in-guest `docker exec … curl
|
|
127.0.0.1:8080`; auth+CSRF skipped, no password set) → response `where=/mnt/felhom-drives/felhom-flash`
|
|
(+`raw=/mnt/felhom-flash`) and `…/felhom-usb` — confirming the fix end-to-end.
|
|
2. Deregistered the two raw entries via `POST /settings/storage/remove` (deregister-only — `RemoveStoragePath`,
|
|
**NOT** eject; the host drive mounts were never touched). Registry now holds ONLY the two stable paths.
|
|
3. Removed the orphaned rootfs placeholder trees `/mnt/felhom-flash` + `/mnt/felhom-usb` in the guest
|
|
(verified 0 files — only an empty userdata skeleton from the buggy raw registration — and verified NOT
|
|
mounts before `rm -rf`). Guest `/mnt` now: `felhom-drives` + `sys_drive` only.
|
|
|
|
## Verification (§6-C) — server-side pipeline + rendered page (NOT browser)
|
|
- **Registry:** `settings.json` `storage_paths` = `/mnt/felhom-drives/felhom-flash` +
|
|
`/mnt/felhom-drives/felhom-usb` (stable only).
|
|
- **Selftest:** `[PASS] Storage paths: 2 connected, 0 disconnected` (was `[WARN] 0 connected, 2
|
|
disconnected`); `[PASS] System data path: /mnt/sys_drive`.
|
|
- **Per-path device resolution (what the UI renders):** `/mnt/felhom-drives/felhom-flash` →
|
|
`/dev/sdc1[/felhom-data] 117.1G`; `/mnt/felhom-drives/felhom-usb` → `/dev/sdb1[/felhom-data] 915.8G` —
|
|
both a DIFFERENT block device from `/` (`overlay`), so the system-drive check is false.
|
|
- **Rendered `/monitoring` HTML** (the bytes the browser renders): "aktiválásra vár" banner = **0
|
|
occurrences**; "Rendszermeghajtón" badge = **0 occurrences**; storage bars show `felhom-flash` /
|
|
`felhom-usb` at `117 GB` / `916 GB`.
|
|
- **Method:** verified via the controller's server-side selftest + the exact server-rendered page HTML +
|
|
the per-path device resolution the page calls — NOT claude-in-chrome. The only residual is the
|
|
browser's visual render of that verified HTML.
|
|
- **Regression:** the correct `runStorageInit`/`runStorageAttach` paths are unchanged (unit tests green);
|
|
not re-run live to avoid mutating a real drive.
|
|
|
|
## Observations (recorded, NOT acted on)
|
|
1. `pendingActivationDrives` (`storage_handlers.go`) keys `attached` by the agent's raw `MountPath` and
|
|
compares to the registered `sp.Path`. With stable registration the activation banner clears, but the
|
|
function no longer detects a genuinely not-yet-propagated drive. The intermediary model activates LIVE
|
|
(no reboot), so the "~30 mp újraindítás" banner copy is likely obsolete; activation detection should
|
|
key off `GuestPath`/`BoundUnderParent` like `planDriveGates`, or the banner be retired. Follow-up.
|
|
2. The free-form settings storage-add (`handlers.go`) registers a user-typed path verbatim; a raw
|
|
`/mnt/<name>` typed there would hit the same shape. Advanced manual flow, out of scope.
|