Adds an in-process ocrs OCR backend as the active analysis engine
(vision left dormant), enables OCR text search on the page-tag
aggregation endpoints, and reshapes the admin analysis/settings UI to
present the OCR-only surface. Bumps version 0.89.0 -> 0.90.0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The HistoryList clear button hand-rolled the same `.icon-btn danger`
styles that were extracted into the shared IconButton component one
commit earlier — the exact duplication that refactor targeted. Swap it
for <IconButton variant="danger">; data-testid/aria pass through the
component's {...rest} spread so behaviour is unchanged.
Both callers also passed an emptyText equal to the prop's default; drop
the overrides so the default lives in one place.
No behaviour change, so no version bump.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Page tags were a first-class section of the mobile Library tab but had no
home in the desktop /profile tabs, so desktop users couldn't browse their
tagged pages without going through /search. Add a /profile/page-tags route
that reuses PageTagsList (the same component the mobile Library renders) and
a "Page tags" tab between Collections and History.
The loader mirrors the library wrapper's page-tags fetch; 401 → sign-in
prompt, matching the other profile sub-routes.
e2e: the desktop Page-tags tab navigates to /profile/page-tags and renders
the list.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Add-tags and Add-to-collection editors opened as centered Modals on
every form factor, so a mobile user went action-Sheet → centered Modal — an
inconsistent hop, and AddTagsSheet's docstring promised a sheet it never
got. Introduce AdaptiveDialog, which renders a bottom Sheet under 640px and
a centered Modal above it behind one shared contract (open / title / onClose
/ children / testid). Route both editors through it:
- AddToCollectionModal now wraps AdaptiveDialog (both its call sites — manga
detail and the reader — adapt automatically).
- The reader hosts AddTagsSheet in an AdaptiveDialog instead of a Modal, so
its "slots into a Sheet (mobile) or Modal (desktop)" contract is now true.
Modal/Sheet already share an API, so the switch is transparent: the dialog
keeps the same role, title, testid, and close affordance. Desktop is
unchanged.
AdaptiveDialog has its own unit test (Sheet on mobile, Modal on desktop);
an e2e proves the mobile long-press → Add-tag flow now opens a Sheet.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The admin area was reachable only from the desktop header's is_admin link;
on mobile there was no tab or row, so admins had to type the URL and no tab
highlighted. Add a conditional Admin row to the mobile Account hub and
`/admin` to the Account tab's match prefixes, so the tab stays highlighted
while an admin is in the admin area.
e2e: an admin session shows an Admin row linking to /admin; a non-admin
never sees it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The brightness slider lived only in the mobile reader settings sheet, so
desktop users couldn't dim — and could be left stuck dimmed by a value a
phone had persisted (brightness is shared via localStorage and the
--reader-dim effect is form-factor agnostic). Add an inline brightness
slider to the desktop reader nav, marked `desktop-control` alongside the
existing mode toggle and gap select. It binds the same `brightness` state,
so a value set on either form factor stays in sync.
Also mocks `/me/read-progress` in the reader-mode e2e setup — without it the
GET/PUT fell through to the (absent) backend and stalled the reader load,
which is why those specs couldn't run headless. The spec is now hermetic and
covers the new desktop control (it dims/undims, and a phone-persisted value
is recoverable).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The desktop /profile/history page and the mobile Library "History" tab
each hand-rendered reading history with separate markup, which had drifted:
the Library tab was a read-only 2-column list with no date and no
removed-chapter / whole-manga fallbacks, while /profile/history was a
3-column list with a per-row clear button. Extract a shared HistoryList
component so the two can't diverge again.
- New HistoryList.svelte owns row rendering, the optimistic-removal UX
(instant remove, rollback + inline error on failure), and the empty
state. Clear is opt-in via an `onClear` prop so a caller can render
read-only, but both surfaces now pass it.
- The mobile Library History tab gains the clear action, the "Read {date}"
line, and the (chapter removed) / whole-manga fallbacks it was missing.
- Continue label is built in script (was inline) so the " — page N" suffix
keeps its spaces — Svelte trimmed them at the {#if} edge, rendering
"Chapter 3— page 7". Both surfaces now read "Continue Chapter N".
Net -204 lines across the two pages. Uploads parity on the mobile tab is
left as a follow-up (it needs the library loader to fetch uploads).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Completes the IconButton extraction by adding `size` and `radius` props and
converting the two remaining outliers that were left out of the first pass:
the desktop header logout button (+layout) and the home search submit button
(+page), both 36px with the medium radius.
IconButton now drives every icon button in the app (except profile/history's,
which the shared-HistoryList change removes separately). Defaults stay 32px /
small radius, so the four already-converted sites are untouched. Rendered
size, radius, and variant match the old markup — no behaviour change, no
version bump.
This also sets up applying the --tap-min touch floor in one place once the
touch-target change lands.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Five files hand-rolled a near-identical 32px `.icon-btn` (same size, hover,
and primary/danger variants). Extract a single IconButton.svelte component
so the treatment lives in one place. Converts the four sites with the
standard 32px form: collections detail, manga edit, upload, and the
chapter-pages editor.
The component takes a `variant` (plain/primary/danger) and spreads any
button attributes (onclick, disabled, aria-label, title, data-testid)
straight through; `type="button"` defaults but a caller can override. The
rendered button keeps the same class, styles, and DOM position, so layout
and behaviour are unchanged — no version bump.
Three icon-button sites are intentionally left out:
- The header (+layout) and home search button are 36px / different radius —
size outliers that need a size/radius prop before folding in.
- profile/history's copy is being removed in the shared-HistoryList change;
touching it here would just conflict.
A component (not a global class) avoids colliding with those remaining
local `.icon-btn` definitions.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The desktop sort <select> hard-coded its four <option>s while the mobile
sort sheet rendered them from SORT_FIELD_LABELS. The two happened to match,
but a label rename in mangaSort.ts would have silently diverged the desktop
control. Render the desktop options from the same map so they can't.
No behaviour change (the rendered options are identical), so no version
bump. Adds an e2e guard asserting the desktop select's option values and
labels equal SORT_FIELD_LABELS in order — verified red against the old
hard-coded markup (with a label renamed) and green after.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The remove buttons on a collection's cards were revealed only on
hover / focus-within (opacity 0 → 1), so on touch they were invisible and
undiscoverable. Show them persistently at the 640px breakpoint and enlarge
the hit area from 24px to 32px (kept below the full 44px floor because the
button floats over a 4-up cover thumbnail).
Test pins the CSS contract (jsdom can't evaluate @media): the mobile block
makes `.remove` opaque.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
`.resync-msg` (the admin Force-resync result) was display:none under 640px,
so mobile admins — who trigger resync from the overflow sheet — got no
success or error feedback. The element is already a sibling of `.action-row`
(not a child), so it survives the row being hidden; the only thing hiding it
was an explicit mobile rule. Drop that rule.
e2e: a mobile admin triggers Force resync from the overflow sheet and now
sees the "Metadata updated" result (written test-first against the hidden
element, then unhidden).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The desktop/mobile consistency review found primary mobile controls
rendering below the ~44px comfortable touch floor. Address the genuinely
shared primitives in one pass:
- Add a --tap-min: 44px token (documented anchor for the floor).
- tokens.css: text inputs, selects, and submit buttons grow to --tap-min
under the 640px breakpoint (fixes the 36px auth/login form controls).
Scoped to form controls + type=submit so dense icon buttons aren't
inflated.
- SegmentedControl: .seg options reach the floor on mobile (it doubles as
the catalog sort toggle and the Library tab switcher).
- Chip: expand .chip-remove's tappable area to --tap-min via a centered
pseudo-element so the 16px glyph stays compact and tag rows don't grow.
Pure-CSS responsive change — jsdom can't evaluate @media or layout, so the
test pins the stylesheet contract (each shared primitive bumps to --tap-min
inside the mobile breakpoint), guarding against the bump being dropped.
The per-route .icon-btn (copy-pasted across seven files) is left for a
separate refactor.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Sort the manga catalog by created / updated / title / author with an
independent asc/desc direction control. An omitted `order` defaults per field
(dates newest-first, text A→Z), matching the UI, so a bare `?sort=<field>`
means the same thing in the browser and over the API; `sort=recent` remains a
back-compat alias for `created`.
- Backend: SortField/SortOrder parsed with validation (structured 422 on bad
input), per-field default_order, NULLS LAST only on the nullable author key,
and migration 0033 indexing mangas(updated_at DESC, id) to back the default
sort and its id tie-break.
- Frontend: catalog sort field + direction (SegmentedControl) on desktop and a
mobile bottom sheet; pure helpers in $lib/mangaSort; keyboard-accessible
direction control; visible Direction labels and a sheet "Done" button.
- Tests: backend integration coverage (defaults, alias, invalid input,
ascending-id tie-break), frontend unit + e2e.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
0.87.20 narrowed but did not close the force-analyze race:
ack_done / ack_failed / renew matched only on (id, state='running'),
so a worker A whose lease was released mid-flight (force-analyze,
reclaim_orphaned) could still come back later — once a successor B
re-leased the same id and put it back to state='running' — and
clobber B's lease. A's stale result became the row's `done` payload;
B's in-flight work was silently lost. The pre-existing test even
documented this hole: "We can't make ack_done's lease_id distinguish
A from B today".
Add `lease_generation BIGINT NOT NULL DEFAULT 0` to crawler_jobs
(migration 0032). It is bumped by every lease, release / release_unowned,
and reclaim_orphaned, so lease identity is `(id, generation)` rather
than just `id`. Each Lease struct carries its generation. ack_done /
ack_failed / renew / release match on `(id, generation, state='running')`
so a stale ack from a prior generation finds zero rows and falls
back to the existing warn-and-skip branch.
Split the release surface in two:
* release(lease_id, lease_generation) — owner-aware path for
workers (graceful shutdown, session-expired), matches the
caller's specific lease.
* release_unowned(lease_id) — id-only path for force-analyze and
ops tools that drop a running lease without holding a Lease
struct; unconditionally bumps generation so any pending ack
from the in-flight original becomes a no-op.
Force-analyze in repo::page_analysis now uses release_unowned; the
test in api_admin_analysis still passes unchanged (it only asserts
the steady-state outcome). The pre-existing `ack_done_no_ops_when_
lease_was_stolen` test is strengthened: it now mints both A and B
leases, asserts their generations differ, and verifies A's stale
ack does NOT clobber B's running row — the assertion the old test
explicitly couldn't make.
Five new tests in tests/crawler_jobs.rs pin every facet:
* lease_assigns_strictly_increasing_generation_per_release
* ack_done_from_dead_lease_after_release_unowned_is_a_no_op
* ack_failed_from_dead_lease_after_release_unowned_is_a_no_op
* renew_from_dead_lease_is_a_no_op
* release_unowned_bumps_generation_even_when_lease_is_held
Mutation-confirmed: relaxing the ack_done guard to
`lease_generation >= $2` (bind 0) makes the race test fail with
state=done — the exact prior-bug shape.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
0.87.12 added a regression test for backend env wiring, but its
BACKEND_CONSUMED list was hand-maintained and drifted: ~22 vars read
by backend/src/ never made it into compose's backend.environment.
An operator setting ANALYSIS_TEMPERATURE=0.3 in .env got a silent
no-op — exactly the bug class the test was supposed to catch.
Make .env.example the single source of truth:
* docker_compose_wires_every_documented_backend_env_var now reads
.env.example at test time, filters NOT_BACKEND_CONSUMED_IN_ENV_EXAMPLE
(compose-side composed vars, frontend-only vars, sidecar vars),
and requires every remaining key to be wired into compose with a
matching ${KEY...} interpolation OR allowlisted via COMPOSE_RHS_EXEMPT.
* every_backend_src_env_read_is_documented_or_allowlisted scans
backend/src/ for env::var / env_bool / env_i64 / ... reads and
requires each key to be in .env.example OR in INTERNAL_NOT_CONFIGURABLE
with a per-entry rationale. Catches the silent-no-op bug from the
other side.
Wire the previously-missing env vars into docker-compose.yml's
backend.environment (CRAWLER_DAEMON, CRAWLER_DAILY_AT, CRAWLER_TZ,
CRAWLER_IDLE_TIMEOUT_S, CRAWLER_CHAPTER_WORKERS, CRAWLER_JOB_RETENTION_DAYS,
CRAWL_METRICS_RETENTION_DAYS, CRAWLER_RATE_MS, CRAWLER_CDN_RATE_MS,
CRAWLER_USER_AGENT, CRAWLER_PHPSESSID, CRAWLER_COOKIE_DOMAIN,
CRAWLER_TOR_CONTROL_COOKIE_PATH, all ANALYSIS_* tuning knobs).
Document each new var in .env.example with the same default value
shown in compose's `${KEY:-default}` form. Allowlist internal/system
vars (HOME = chromium cache dir fallback, CRAWLER_BROWSER_*,
CRAWLER_SKIP_*, the multi-paragraph ANALYSIS_*_PROMPT seeds) with
per-entry rationale.
Mutation-tested both sides: adding `env::var("MANGALORD_FAKE_NEW_KNOB")`
to main.rs fails the scanner; removing the new compose wiring fails
the documented-var test with all 23 missing keys named.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
CancellableLoader.run() previously swallowed AbortError but propagated
every other rejection. A predecessor whose fetch survived its abort
(timing race, or fn that ignores signal) and then 5xx'd would bubble
the error into the caller's catch, clearing the successor's spinner
and writing its `error` field — exactly the flicker the loader was
supposed to prevent.
Strengthen the contract: a call that no longer owns this.current
returns null regardless of how it died. Symmetric with the existing
late-resolve suppression at the success path. The loadCoverage catch
in admin/analysis/+page.svelte now only fires when this call genuinely
failed AND still owns the loader — so clearing the spinner there is
sound.
Three new unit tests pin all three contract corners: a superseded
predecessor's non-Abort rejection resolves to null; a standalone call
with no successor still throws normally; the AbortError swallow on a
still-owning call (synthetic AbortError from fn with no successor or
cancel()) remains protected, since the new supersede check would
otherwise have eaten every existing AbortError-path test and left
that branch as silently-removable dead code.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
0.87.16 followup. Extract the four debounce-race invariants
(predecessor abort, late-result suppression, AbortError swallow,
ownership-aware finally) from `loadCoverage` into a reusable
`CancellableLoader` helper at $lib/util/cancellable, with 9 unit
tests covering each. Page rewrite uses the helper.
Also corrects: hooks.server.test.ts (added Node-fetch rejection
shape for already-aborted entry; removed redundant tick); admin.test.ts
(AbortError-propagation now uses pre-aborted signal + mockRejectedValueOnce
mirroring real Node fetch).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0.87.14 followup. The 0.87.14 test used MissedTickBehavior::Burst —
any post-prep yield replayed missed ticks and could fake the >= 2
threshold. Switch to Skip and assert on `ticks_post - ticks_pre >= 50`,
which sits well above the inlined-prep ceiling (~5) and far below
the spawn_blocking floor (~700+). Mutation-confirmed.
Also corrects: prepare_analysis doc count ("five fallback paths" → 4).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0.87.13 followup. The Cancelled publish had no asserting test —
mechanical revert of the publish block shipped green. New sqlx test
subscribes before spawn, drives SlowDispatcher into flight, cancels,
asserts both Started and Cancelled frames with breadcrumb fields.
Mutation-confirmed.
Also corrects: cron-test rationale comment (inverted unlock-axis),
Cancelled docstring (self-contradicting clear-triggers sentence).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0.87.11 followup. The UPDATE matched pending+running but the worker
holds the pre-update `force=false` in an in-memory local — so on the
running branch the row flipped but the worker's skip-if-done still
acked done without re-analyzing. Now `jobs::release` any running row
the UPDATE matched, so the worker's ack_done no-ops (state guard) and
a fresh lease picks up the updated payload.
New test mints the running-state shape, asserts state=pending +
force=true + attempts=0 post-click. Mutation-confirmed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0.87.10 followup from the rereview. The fix lacked a test for the
specific axis it changed (admin authority via cookie only). New test
mints a bearer for a promoted admin and asserts PATCH/PUT-cover/DELETE-
cover all 403 on NULL-uploader rows. Mutation-confirmed.
Also: `MultipartBuilder::finalize` now `pub` for bearer-multipart use;
`admin_csrf_guard` rustdoc updated to describe the post-0.87.10
cookie-precedence rule.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two 0.87.15 ship-blockers:
1. mem_check_and_stop test now ALSO greps for the "deferring to next
tick" warning — distinguishes fix from bug. Mutation-confirmed.
2. Extract `refresh_running_after_mem_check` (sets global instead of
echoing so loop-scope flags persist across calls) and drive it
directly from the harness, closing the structurally-untestable
post-mem refresh ERR guard.
Adjacent: log-on-flip gating on both ERR warnings; `log` to stderr;
dangling-TempDir in app::tests reclaim test fixed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two ship-blockers from the rereview:
1. Env var name mismatch: 0.87.12 renamed every operator surface to
ANALYSIS_VISION_MODEL but the reader still called env::var on
ANALYSIS_MODEL — every documented deploy ran analysis with an
empty model. Rename the reader.
2. Compose regression test was substring-only and missed exactly the
ANALYSIS_VISION_MODEL vs ANALYSIS_MODEL RHS mismatch from 0.87.12.
Tightened to require each backend-consumed var either be in a
small hardcoded allowlist (BIND_ADDRESS, STORAGE_DIR) or have its
RHS interpolate `${KEY...}` with the same name. Mutation-confirmed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three 0.87.9 follow-ups:
1. hooks.server.ts: two new tests covering the
already-aborted-at-entry and mid-flight client-disconnect branches
of the abort chain. Mutation-confirmed.
2. lib/api/admin.ts: two new tests pinning the
getAnalysisMangaCoverage(init?.signal) pass-through and the
AbortError-rejection propagation the dashboard's debounce-race
handler relies on. Mutation-confirmed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0.87.7 + 0.87.8 follow-ups:
1. mem_check_and_stop now defers on docker ERR instead of silently
no-op'ing the SIGTERM under pressure.
2. Post-mem_check_and_stop loop refresh now also handles the ERR
sentinel (continues the loop) rather than letting it reset up_for
and defeat MAX_UPTIME.
3. New `vision-manager/test_manager.sh` — 7 smoke tests covering
vision_running's three exit shapes and mem_check_and_stop's ERR
symmetry, via stubbed docker/psql/curl on PATH. Production loop
guarded by MANAGER_TEST_NO_MAIN.
4. New `app::tests::spawn_analysis_daemon_reclaims_orphaned_analyze_leases`
pins the 0.87.7 wire-through: seeds an expired analyze_page lease,
spawns + shuts down, asserts pending + attempts refunded.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two 0.87.5 follow-ups:
1. Pin spawn_blocking dispatch: new current_thread runtime test runs
analyze() on a 6 MP image while a counter task ticks every 5 ms.
Without spawn_blocking the counter is starved (0 ticks); with it
the runtime stays responsive (≥2 ticks). Mutation-confirmed.
2. Warn on Undecodable fallback: five fallback paths in prepare_analysis
now emit a tracing::warn! with dimensions so a JPEG encoder
regression isn't indistinguishable from "the page was just garbage."
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two 0.87.4 follow-ups:
1. Cron tick: existing shutdown-on-cancel test now also asserts the
advisory lock is re-acquirable after cancel-during-tick. Without
this, a refactor moving `pg_advisory_unlock` under the cancel arm
would ship green and break multi-replica safety.
2. Analysis worker: cancel-mid-dispatch left the dashboard banner stuck
on the cancelled page until a later page overwrote it. Add a
`Cancelled` AnalysisEvent variant, publish on the cancel-release
branch, wire the analysis and admin overview dashboards to clear
on it. Per-page chip rolls back from `analyzing` to `queued`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
0.87.1 added 12+ env vars to `.env.example` but the compose backend
`environment:` block forwarded none of them — `.env` interpolation
doesn't pass vars to the container. Operators got
anonymous-serving sites, silent ADMIN_* no-ops, and missing
ANALYSIS_* configuration.
Wire 36 backend-consumed vars; also substitute the hardcoded
`BACKEND_URL` on the frontend; emit a startup warning on half-set
`ADMIN_USERNAME`/`ADMIN_PASSWORD`. Regression-test parses
docker-compose.yml and asserts the wiring.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Four 0.87.6 follow-ups from the adversarial review:
1. **Migration 0031 pre-dedup.** Demote all-but-lowest-id duplicate
`analyze_page` rows in `(pending|running)` to `dead` before
creating the unique index, with a curator-recoverable last_error
marker. Without this, `sqlx::migrate!` would refuse to boot on
any dirty production DB.
2. **`enqueue_for_page(force=true)` collision.** The partial unique
index used to silently swallow force requests when a
`force=false` job was already pending. Repo function now upgrades
the pending row's `force` flag in place (or falls through to
re-INSERT if the sibling drained mid-call), and reports an
`EnqueueForPageOutcome` for accurate auditing.
3. **`record_duration` gate test.** New test seeds a `done` row with
known duration, force-re-analyzes with failing dispatcher +
max_attempts=3 (non-terminal), asserts duration_ms wasn't
overwritten.
4. **`bookmark.rs` PK comment correction.** Use `b.id DESC` instead
of the wrong `manga_id`; PK is actually `id`, and migration 0004
allows both chapter-level and manga-level bookmarks on the same
manga.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two follow-ups to 0.87.2 + 0.87.3 from the adversarial review:
1. **CSRF cookie-ride.** Bearer-bypass branch checked Authorization
header presence only — an attacker page could mint `Authorization:
Bearer junk` and ride the still-attached session cookie. Bypass now
requires bearer AND no cookie. Drive the cookie-name comparison off
`SESSION_COOKIE_NAME` and split on `=` (not prefix).
2. **`require_can_edit` admin path open to bearer.** Now reads admin
authority off an `Option<CurrentSessionUser>` extractor — None on
bearer-only requests. Owner-match path still accepts bearer; admin
path is cookie-gated.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three frontend findings (medium):
1. **`hooks.server.ts` didn't chain client-disconnect into the upstream
fetch.** Closing an admin tab left the backend's `broadcast::Receiver`
+ spawned tokio task running until next backend restart. Listen for
`event.request.signal.abort` and forward to the timeout controller.
2. **`admin/analysis/+page.svelte` missing visibility/pagehide/pageshow
handlers.** Mobile Safari throttles SSE in background tabs; bfcache
leaves a dead connection on back-nav. Apply the close/open-stream
pattern from `admin/crawler/+page.svelte`.
3. **`loadCoverage` had no AbortController.** Debounced search
keystrokes raced. Per-loader controller plumbed through
`getAnalysisMangaCoverage`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three medium vision-manager findings:
1. **`vision_running()` conflated "container absent" with "inspect
failed".** A transient docker-socket-proxy hiccup made the manager
think vision had stopped, which reset `up_for`/`idle_for` and
defeated MAX_UPTIME. Distinguishes: exit 0 → echo true/false;
"No such container" → false; else → ERR (loop skips tick).
2. **`crawl_running()` fail-closed where `analysis_enabled()` fail-opens.**
A psql `ERR` logged a misleading "RAM mutex" line every tick.
Numeric-guard at the call site with a distinct log.
3. **`start_vision` left wedged containers running** on
`START_HEALTH_TIMEOUT` expiry — pinned forever in not-ready state.
Now `stop_vision` so next tick retries from a clean state.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`reclaim_orphaned` previously ran only inside `spawn_crawler_daemon` at
startup. A deploy with the crawler disabled and analysis enabled
(analysis-only mode) never refunded orphaned `analyze_page` leases —
recovery deferred to the lazy lease-expiry path. Wire it into
`spawn_analysis_daemon` too. Safe under multi-replica + idempotent
against the crawler-side call.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three medium correctness fixes:
1. **Pagination ORDER BY missing tiebreakers** — four admin list queries
sorted by a timestamp or `chapters.number` with no stable secondary
sort. Add `id` tiebreakers across `repo::admin_view`,
`repo::admin_audit`, `repo::bookmark`.
2. **`enqueue_pages` NOT EXISTS read-then-insert race** — concurrent
admin clicks could land duplicate analyze_page jobs. Migration 0031
adds a partial unique index mirroring 0014; query relies on
`ON CONFLICT DO NOTHING`.
3. **`record_duration` overwrote a prior done row's duration on a
non-terminal failed retry of a force-re-analyze.** Gate the call
on "did this attempt actually write a row?".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`VisionClient::analyze` ran `image::load_from_memory`, `DynamicImage::resize_exact(Triangle)`,
and per-band JPEG encodes directly on the tokio task. Each call is
hundreds of ms of pure CPU per page. With multiple workers, the runtime
threads got starved — axum handlers, SSE streams, the crawler daemon's
async timers, and other tasks sharing the runtime all stalled while
analysis was active.
Extract a `prepare_analysis` helper that does ALL the CPU work
(decode + plan + width-reduce + slice + JPEG encode) and returns a
`PreparedAnalysis { Single | Sliced | Undecodable }` of already-encoded
byte payloads. `analyze` calls it inside `tokio::task::spawn_blocking`,
then the async HTTP loop only iterates over the prepared bytes — no
image-crate operations happen on the runtime any more.
Single page → one spawn_blocking → one HTTP call.
Tall page → one spawn_blocking → N OCR calls + 1 grounding call.
Memory peak rises slightly for the Sliced path (all bands held
encoded at the same time instead of one-at-a-time) — for a typical
manga page split into 4 slices at ~200 KB each, that's ~600 KB of
extra peak. Negligible vs. the runtime-starvation cost.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two high findings sharing one root cause: long-running daemon work that
ignored the shutdown signal until it finished naturally.
**Cron tick (crawler).** `CronContext::run_tick` ran the metadata-pass
body inside `AssertUnwindSafe(...).catch_unwind().await` with no
cancellation. A fresh-catalog walk (many minutes) wedged
`DaemonHandle::shutdown`, `Supervisors::reload_crawler` (under the
supervisor lock), and SIGTERM responsiveness for the whole pass.
Wrap the body in `tokio::select! { biased; _ = self.cancel.cancelled() => ...; r = body => ... }`.
On cancel the body future drops, which cascades to dropping
`metadata.run()` — exactly what gives Chromium/lease teardown a chance
via the BrowserManager idle reaper. The advisory unlock + conn drop
still run unconditionally.
**Analysis dispatch.** `process_lease` wrapped the vision call only in
`tokio::time::timeout(self.job_timeout, …)` — default 600s. Toggling
analysis off in the dashboard (under the supervisor lock) or stopping
the container blocked on whichever call was in flight. Same fix:
`tokio::select!` against `self.cancel.cancelled()`. On cancel,
`jobs::release` returns the lease to the queue without burning an
attempt — next tick picks it up.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`require_can_edit` matched `Some(owner) != caller → Forbidden` but let
`None` through. Every crawler row (`repo::crawler::upsert_manga`) has
NULL `uploaded_by`, so any signed-in user could PATCH /api/v1/mangas/<id>
and rewrite the catalog — `update`, `put_cover`, and `delete_cover`
were all in scope. With self-registration on by default, the path was:
register → mass-edit catalog + delete cover blobs from storage.
The original carve-out at the comment said "Once an admin role lands the
NULL case can flip to admin-only." The admin role landed in migration
0018; flipping it now. New rule:
* `Some(owner)` and owner == caller → ok (unchanged)
* `Some(_)` and caller is admin → ok (admin moderation)
* `Some(_)` otherwise → 403 (unchanged)
* `None` and caller is admin → ok (NEW — operator can curate)
* `None` otherwise → 403 (was: ok — the bug)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Four high/medium findings from the audit, plus their tests:
1. **SSRF on admin-editable URLs.** `CrawlerSettings::start_url` and
`AnalysisSettings::endpoint` validated only with `Url::parse`. A
hostile or CSRF-able admin could repoint at `169.254.169.254`
(cloud metadata), `127.0.0.1:5432` (postgres), or any RFC1918 host
— and the vision worker bearer-attaches an env-managed API key to
every call. Extract `ensure_public_target` from `safety::is_safe_url`
(allowlist-free public-host check: scheme + private-IP literal +
localhost) and wire it into both fields. Docker DNS names
(`mangalord-vision`) keep passing because they're hostnames, not
IP literals. The analysis check is gated on `enabled=true` so the
dev-default `localhost:8000` stays usable until the worker is
actually turned on (toggling enabled=true later re-runs the gate).
2. **Admin CSRF fail-open default.** `ADMIN_ALLOWED_ORIGINS=` empty
silently skipped the entire CSRF check, so an operator forgetting
the env var shipped an unguarded admin surface. Cookie-auth POSTs
now fail-closed when the allowlist is empty AND when neither
`Origin` nor `Referer` accompanies the request. Two carve-outs:
`Authorization: Bearer …` callers bypass (bots can't be CSRF'd),
and requests with NO session cookie at all bypass to let the auth
extractor return a clean 401 instead of a confusing 403.
3. **Analysis HTTP clients didn't `.no_proxy()`.** The crawler client
already did. Ambient `HTTP_PROXY`/`HTTPS_PROXY` in container env
would exfiltrate page-image bytes + the bearer key through an
upstream proxy. Same for the readiness probe.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three operational gaps tripped a recent review:
1. **Postgres had no `restart:` policy.** tor / docker-socket-proxy /
vision-manager all set `unless-stopped`; pg was the lone exception.
A transient OOM or panic left backend's `depends_on: service_healthy`
blocking every future restart and took the stack offline. Add
`unless-stopped` on postgres and on backend/frontend in the base
compose so the prod overlay is the same shape.
2. **`BACKEND_PROXY_TIMEOUT_MS` was documented but silently a no-op.**
`.env.example` carried it but `docker-compose.yml` never forwarded it
into the frontend service env, so the value never reached
`hooks.server.ts`. Wire it through.
3. **`.env.example` omitted required env vars.** Operators copying the
template got no admin bootstrap, no boot-seed crawler URL, no
analysis configuration. Add: `ADMIN_USERNAME`, `ADMIN_PASSWORD`,
`PRIVATE_MODE`, `ALLOW_SELF_REGISTER`, `CRAWLER_START_URL`,
`CRAWLER_CDN_HOST`, `ANALYSIS_ENABLED`, `ANALYSIS_VISION_URL`,
`ANALYSIS_VISION_MODEL`, `ANALYSIS_WORKERS`,
`ANALYSIS_JOB_TIMEOUT_SECS`, `ANALYSIS_API_KEY` — each with the
comment block that explains seed-vs-runtime semantics.
Also adds `frontend/vite.config.ts.timestamp-*.mjs` to `.gitignore` — a
crashed dev server leaves these transient files behind otherwise.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
