docs: REPORT.md — badge-clip fix + M3 live re-verify + refresh-loop fix
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -1,39 +1,96 @@
|
||||
# REPORT — storage lifecycle completion + acceptance sweep (controller half) → v0.68.1 (2026-06-15)
|
||||
# REPORT — two UI cleanups + M3 live re-verification (2026-06-15)
|
||||
|
||||
## Type
|
||||
Finished the storage lifecycle on the intermediary-mount model + drove a full live acceptance sweep
|
||||
(incl. a real host reboot). Built + Linux-gated + deployed to guest 9201 per commit. Pairs with agent
|
||||
v0.36.7. Full audit: `felhom.eu/documentation/audits/storage-lifecycle-acceptance-2026-06-15.md`.
|
||||
**Deployed:** controller **v0.68.3** on guest 9201 / felhom-pve (bootstrap-managed; healthy).
|
||||
**Commits:** `37ed757` (v0.68.2 CSS) · `a821a9d` (v0.68.3 refresh fix). Trunk-based, pushed to `main`.
|
||||
|
||||
## What changed (controller)
|
||||
- **v0.68.0** — H2/H3/M1/M3 + deterministic boot-id:
|
||||
- **Boot-id determinism** (`processGuestBootChange` + `settings.LastGuestBootID`): the agent reports
|
||||
`guest_boot_id`; on change the controller recreates drive-backed apps deterministically (replaces the
|
||||
fragile container-uptime sample, which missed apps stopped at the sample instant).
|
||||
- **H2** decommission UI button (migrate / decommission-anyway type-to-confirm); **H3** one-click
|
||||
re-enroll of a decommissioned drive (`handleStorageReconnect` decommissioned branch); **M1**
|
||||
`defaultPromotionTarget` (auto-promote on decommission-of-default, BLOCK if it's the only drive);
|
||||
**M3** migrate merge-walk re-asserts 2775 setgid on userdata dirs (`isUserdataDir`).
|
||||
- Fixed the H1 `{path}`/`{where}` JS body mismatch.
|
||||
- **v0.68.1** — boot-id recreate ALL deployed drive-backed present apps **state-independent** (v0.68.0
|
||||
filtered on container state and **missed 5 apps** docker hadn't auto-restarted yet — caught in the live
|
||||
host-reboot test).
|
||||
---
|
||||
|
||||
Non-hollow tests + companions: `TestShouldRecreateOnBoot` (old sample missed a stale/stopped app),
|
||||
`TestDefaultPromotionTarget`, `TestIsUserdataDir`.
|
||||
## Part 1 — stack-card state-badge clipping fix (v0.68.2, CSS-only)
|
||||
|
||||
## Acceptance sweep (live) — per-item PASS in the audit doc
|
||||
HOST REBOOT (✅ ordering + auto-start + convergence, no manual intervention), guest reboot ×3, drive-yank
|
||||
fail-close (no leak), capability-proof fail-close, confinement, controller redeploy (rslave explicit),
|
||||
two-drive isolation, **H2 decommission**, **H3 re-enroll** (after the agent v0.36.1 raw-stays-mounted
|
||||
fix), **M1 default reassignment**, rapid eject/reconnect (single bind after the agent double-bind
|
||||
root-cause fix). E14/E15 migrate-all not re-run live (M3 unit-tested + 2775 convention verified live).
|
||||
**Symptom:** on the /stacks page, an **unhealthy** app's state badge clipped — "Nem egészséges" rendered
|
||||
as "Ner…". Healthy / not-deployed cards were fine.
|
||||
|
||||
## End state
|
||||
Both external drives registered at `/mnt/felhom-drives/<name>`, felhom-flash default, 25 containers
|
||||
healthy. controller v0.68.1 deployed (golden/bootstrap image tag mechanism).
|
||||
**Root cause (confirmed):** `.stack-detail-header` is `flex` / `space-between` holding `.stack-title-row`
|
||||
(logo + title + subdomain link + the `route-unpublished` warning) and `.stack-state-badge`. Only on an
|
||||
unhealthy app does the long "⚠ URL nem elérhető – útvonal nincs publikálva" warning render, inflating the
|
||||
title-row. The title-row had no `min-width:0` (so it wouldn't shrink below its content) and the
|
||||
`white-space:nowrap` badge had no `flex-shrink:0` (so the flexbox compressed the *badge* instead → clip).
|
||||
|
||||
## Known residuals
|
||||
- Boot-id first-sight bounce (fresh controller recreates apps once on first start — one-time per
|
||||
data-volume lifetime).
|
||||
- E14/E15 full migrate-all round-trip not re-run live (data-merge risk; M3 unit-tested).
|
||||
**Fix** (`internal/web/templates/style.css`):
|
||||
```css
|
||||
.stack-title-row { flex: 1; min-width: 0; } /* shrinks + wraps its own content */
|
||||
.stack-state-badge { flex-shrink: 0; } /* never compressed */
|
||||
```
|
||||
|
||||
**Browser-verified on /stacks (demo, guest 9201):**
|
||||
- BEFORE (v0.68.1): komga badge clipped to "Ner…" at the card's right edge (screenshot captured live).
|
||||
- AFTER (v0.68.2): komga badge reads the full **"Nem egészséges"**; the "⚠ URL nem elérhető…" warning
|
||||
**wraps within the title column** (screenshot captured live).
|
||||
- Regression: Mealie ("Fut", healthy) and all "Nincs telepítve" cards visually unchanged.
|
||||
|
||||
**Defensive note (not applied):** the spec offered adding `flex-shrink:0` to
|
||||
`.badge-missing-storage`/`.badge-orphaned` *if* a card showing BOTH unhealthy AND a missing-storage/
|
||||
orphaned badge still clipped. No such card exists live right now, and the primary fix's mechanism
|
||||
(title-row `min-width:0` absorbing all shrink) already shields sibling badges, so it was left out — a
|
||||
one-liner if that combo ever appears.
|
||||
|
||||
---
|
||||
|
||||
## Part 2 — M3 live re-verification (no code change — the migrate path is correct)
|
||||
|
||||
**Goal:** prove the migrate merge-walk re-asserts the `2775`-setgid/gid-1000 convention on a
|
||||
**pre-existing STALE dir** (the B3 calibre case), which the acceptance audit had only covered by unit
|
||||
test + static convention.
|
||||
|
||||
**Method:** two full **migrate-all** runs driven through the dashboard UI (the real
|
||||
`/api/storage/migrate` endpoint, all 8 flash apps: audiobookshelf, calibre-web, immich, jellyfin, komga,
|
||||
paperless-ngx, radarr, romm): **flash→usb**, then **usb→flash** (restoring apps to the default drive).
|
||||
Before each, the migrate **target** was pre-seeded host-side with deliberately-stale `0755`,
|
||||
**non-setgid** userdata dirs; `stat` recorded.
|
||||
|
||||
**Result — PASS (clean, non-hollow):**
|
||||
- Pre-seeded `userdata/documents` at `755 gid1000` — a dir **no app mounts** — became
|
||||
**`2775 gid1000`** after migrate (the `EnsureUserdataDir` re-assert in `walkMerge`, migrate.go:854-859).
|
||||
Same for the seeded `userdata` root and `userdata/import`. Whole target userdata tree = `2775`.
|
||||
- Integrity: static seeded files (`b3-document.txt`, `b3-movie.txt`, `b3-photo.txt`, `demo.jpg`) had
|
||||
**identical sha256** before/after.
|
||||
- All 8 apps redeployed **healthy** on the target (komga "unhealthy" = its pre-existing
|
||||
`/api/v1/actuator/health` 401 — serving, unrelated to migration).
|
||||
|
||||
**Secondary finding — the `import/calibre` 755 is an APP clobber, not a migrate bug.** In both runs
|
||||
`userdata/import/calibre` ended at `755`. Isolated live: **calibre-web (CWA)** bind-mounts
|
||||
`${USERDATA_PATH}/import/calibre` as `/cwa-book-ingest` and **chmods it to 755 (strips setgid) on every
|
||||
startup** — proven by setting it to `2775` and restarting calibre-web alone (reverted to `755`, no
|
||||
migration). The merge-walk *does* re-assert `2775` during copy; calibre-web re-clobbers it during the
|
||||
flip/redeploy phase (this is also why the source flash calibre was already `755` pre-migration).
|
||||
**migrate.go was therefore NOT modified** — the M3 path works. If the convention must hold for that
|
||||
single-app ingest dir, the fix belongs in the catalog/app layer (an `umask`/entrypoint wrapper for CWA),
|
||||
not in migrate.
|
||||
|
||||
---
|
||||
|
||||
## Part 3 — Beállítások endless-refresh loop fix (v0.68.3) — found during Part 2
|
||||
|
||||
**Symptom (user-reported):** after any migration finished, the settings page reloaded itself every
|
||||
~1.5 s, forever.
|
||||
|
||||
**Root cause:** `MigrationStatus` keeps returning the last `done` job indefinitely (never cleared). The
|
||||
settings page's resume-view IIFE called `migWatch()` for *any* returned job, and `migWatch`'s `done`
|
||||
branch does `setTimeout(()=>location.reload(),1500)`. So every load saw the persisted `done` job → watched
|
||||
it → reloaded → saw it again → endless loop.
|
||||
|
||||
**Fix** (`internal/web/templates/settings.html`): the resume-view now starts the watcher **only for an
|
||||
in-progress job** (`phase!=='done' && phase!=='aborted'`). The one-time post-completion reload still fires
|
||||
from the *active* watcher started by `storageMigrateAll`, so a real migration refreshes drive state once
|
||||
when it finishes — but a stale terminal job no longer triggers the loop.
|
||||
|
||||
**Browser-verified on v0.68.3:** with a persisted `done` job present in the journal, the settings page
|
||||
stayed put for 25 s (`navType:navigate`, planted marker survived, migration panel idle) — no reload.
|
||||
|
||||
---
|
||||
|
||||
## Resting state
|
||||
- Apps healthy on **felhom-flash** (the default drive); **felhom-usb** cleaned/empty.
|
||||
- Controller **v0.68.3** deployed live on guest 9201.
|
||||
- Build/deploy: built+pushed on 192.168.0.180, deployed via the bootstrap mechanism
|
||||
(`docker pull` → `/etc/felhom-controller-image` → restart `felhom-controller-bootstrap.service`).
|
||||
|
||||
Reference in New Issue
Block a user