Surface the admin_audit table (written by every mutating admin action but
previously unreadable) as a new /admin/audit tab. New repo::admin_audit::list
(LEFT JOIN users for the actor's username, filter by action/target_kind/actor/
since, at DESC via admin_audit_at_idx) behind GET /v1/admin/audit, mirroring the
crawler history endpoint's paged/clamped idiom.
Frontend: a self-contained AuditTable (target-kind + window + action filters,
expandable JSON payload, AbortController-cancelled fetches, loading/error/empty
states) and shared fmtAgo() in format.ts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The interleaved metadata pass misses mangas to list drift (a title slips a
pagination slot during the slow detail walk). Add a reconcile pass: a cheap,
full, list-only walk (refs only, no detail visit, no early stop) that
set-diffs the walked keys against manga_sources and enqueues the strictly-
missing ones as SyncManga jobs. A set-diff is immune to drift — a manga only
has to appear somewhere in the list.
- Build the previously-dead SyncManga worker by extending RealChapterDispatcher
to run the shared pipeline::process_manga_ref (fetch → upsert → cover →
chapters), refactored out of run_metadata_pass so both paths stay in lockstep.
The crawl worker now leases both sync_chapter_content and sync_manga
(jobs::lease_kinds); both serialize on the single exclusive browser.
- SyncManga payload carries url + title so the worker can rebuild the ref and
the dead-jobs/history UI can label a missing manga that has no manga row yet.
- "Missing" = strict NOT EXISTS (dropped rows count as present). Re-enqueue is
skipped when a pending/running/dead SyncManga job already exists, so a gone
manga (detail 404 → retries → dead) is left dead and not retried.
- New POST /v1/admin/crawler/reconcile (fire-and-forget, shares
manual_pass_lock) + "Reconcile missing" admin button + Reconciling status
phase streamed over SSE.
- dead-jobs/history queries surface payload title/url/key; tables fall back to
them for sync_manga rows; both searches match the payload title.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The admin overview dashboard polls /admin/overview every 30s. Its
`manga_stats` aggregate evaluated MANGA_SYNC_STATE_CASE over the WHOLE
library (14.5k mangas) with no LIMIT, and that CASE runs a correlated
EXISTS over crawler_jobs. With no index on the `sync_chapter_list`
manga_id path, Postgres seqscanned all in-flight jobs *per manga* —
O(mangas x jobs). The disabled-analysis backlog (9.5k pending
`analyze_page` rows that can never match the sync-kind filter) inflated
every scan, so each call took 6-7 min. The 30s poll stacked ~10 of them
concurrently → load 11, Postgres at ~100%.
EXPLAIN ANALYZE on the live DB: the rewrite drops the query from 6-7 min
to ~1.5s (per-manga crawler_jobs seqscan → single index-backed pass).
Fixes:
- Rewrite `manga_stats` to collect the in-flight manga-id set in ONE pass
over the sync jobs (index-backed), then classify via a hash semi-join —
O(mangas + jobs), immune to the analyze_page backlog size.
- Add partial index crawler_jobs_sync_chapter_list_manga_idx (migration
0029) covering the sync_chapter_list manga_id path; mirrors the existing
sync_manga index (0020). Also speeds the per-row Mangas tab listing.
- statement_timeout backstop (5s) on the aggregate so a pathological plan
can never pin a backend for minutes again.
- Single-flight + 10s TTL cache on the /admin/overview handler so
concurrent pollers coalesce instead of stacking.
- Slow the dashboard poll 30s -> 60s (server-cached now anyway).
The in_progress rule in the rewrite is kept in lockstep with the first
arm of MANGA_SYNC_STATE_CASE (documented in both places).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Persist blob sizes (covers + chapter pages) and surface upload storage
usage to admins in two places:
- Admin System tab: total/covers/chapters totals, ratio of disk, average
image sizes, and top-5 largest mangas/chapters leaderboards, behind a
new GET /api/v1/admin/storage endpoint (separate from /admin/system so
the cheap DB aggregates aren't coupled to its 250ms CPU sample).
- Manga detail page: total chapter-content size and per-chapter size,
with an em-dash for uncrawled or not-yet-measured chapters.
Sizes are captured at write time (crawler put_stream return, uploaded
page/cover byte length) into nullable pages.size_bytes /
mangas.cover_size_bytes columns; NULL means "not yet measured", distinct
from a real 0. A one-shot, idempotent admin "Backfill sizes" action
(POST /api/v1/admin/storage/backfill) stats pre-existing blobs in
keyset-paginated, capped batches and reports more_remaining so a large
legacy library can be drained over multiple runs.
Aggregates and leaderboards treat NULL honestly (only fully-measured
entities are ranked); a dashboard banner flags unmeasured rows so partial
figures aren't read as complete. persist_pages now prunes stale page rows
on a shrinking re-crawl so totals and page_count stay consistent.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The crawler's headless Chromium advertised the default HeadlessChrome UA,
which Cloudflare's bot detection challenges on sight — over Tor that meant an
unsolvable 'Just a moment' page on every fetch, so session verification kept
classifying pages as transient and the metadata pass failed at
'acquire browser lease'. (Confirmed empirically: same Tor exit + a normal
Chrome UA returns the real page with the #logo sentinel; the HeadlessChrome
UA returns the challenge.) The existing user_agent setting only fed reqwest,
never the browser. Override the launcher's UA with a realistic non-headless
default (CRAWLER_USER_AGENT to customize); the UA may contain spaces, unlike
CRAWLER_BROWSER_ARGS.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Three failure-mode fixes surfaced by a crawler recovery audit:
- Graceful shutdown mid-dispatch now releases the in-flight job back to
pending without burning a retry attempt. process_lease wraps the
dispatch in a biased tokio::select! cancel arm that aborts the
heartbeat and calls jobs::release; previously a mid-job SIGTERM left
the row 'running' until lease expiry and cost one of max_attempts.
- Boot-time jobs::reclaim_orphaned resets 'running' jobs with an expired
lease back to pending (attempt refunded), run once at daemon startup
before workers start. Crash recovery is now immediate instead of
waiting a full lease window for the lazy lease-expiry path. Safe under
multi-replica: only already-expired leases are touched, which a
healthy heartbeating peer never has.
- Manga-list parsing now warns when listing anchors are dropped for a
missing/empty href or title (split into parse_manga_list_anchors so
the drop count is testable), turning silent source markup drift into
an observable signal. Returned refs are byte-identical to before.
Tests added: shutdown_mid_dispatch_releases_lease_without_burning_attempt,
reclaim_orphaned_resets_only_expired_running_jobs, and a drop-count unit
test. Patch bump 0.81.0 -> 0.81.1 (both manifests).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The private_mode_* config unit tests set then remove_var(DATABASE_URL).
That env is process-global, and the #[sqlx::test] lib tests
(crawler::content, crawler::session_control, ...) read it in parallel,
so on a multi-core runner they intermittently saw it unset and panicked
'DATABASE_URL must be set: NotPresent' (6 failures on the Pi's 4 cores).
Never overwrite/remove it: set only if absent via ensure_database_url(),
leaving CI's real URL stable for the parallel DB tests.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Backend:
- Log on readiness state transitions (ready<->not-ready) so a misconfigured
ANALYSIS_VISION_HEALTH_URL, which otherwise parks the worker silently
forever, is diagnosable.
- Add unit tests for HttpVisionReadiness: 2xx->ready, non-2xx->not-ready,
connection error->not-ready (the production status mapping had no test).
vision-manager:
- Guard the backlog count against non-numeric output before `-gt`, so a stray
value can't exit-2 and kill the loop under `set -e`.
- Throttle the crawl-mutex "deferring start" log to once per episode.
- Only reset the idle/uptime timers when `docker stop` actually succeeds, so a
failed stop retries next tick instead of waiting a full debounce window.
- Decouple the per-probe curl timeout (HEALTH_TIMEOUT) from the warm-up poll
cadence.
Docs:
- Correct the docker-socket-proxy comment: CONTAINERS+POST permits the full
container lifecycle (not just start/stop); state the real trust boundary.
- Document that the externally-defined mangalord-vision container must share
the compose network for name resolution.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The analysis worker leases an analyze_page job (incrementing attempts in
SQL) before calling vision, so a stopped/loading vision burns all retries
and writes a permanent failed page_analysis row. With the vision-manager
autoscaler idle-stopping the container, that would poison the queue on
every cold start.
Add an optional VisionReadiness seam (HTTP GET /health in production) wired
via the new env-only ANALYSIS_VISION_HEALTH_URL. When set, the worker parks
without leasing until the probe answers 2xx; jobs stay pending with
attempts untouched and resume the instant vision is ready. None preserves
today's behavior for always-on endpoints.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move crawler and analysis configuration out of boot-only env vars and into
a DB-backed, admin-editable surface applied live without a restart.
- New `app_settings` table (migration 0026) holds one JSONB row per group;
env vars seed it on first boot, then the DB is the source of truth.
- `settings.rs` adds serializable Crawler/Analysis DTOs with env-base +
DB-overlay conversion and field-level validation; env-only/secret fields
(browser, proxy, TOR, vision API key, PHPSESSID) are never persisted.
- `GET/PUT /api/v1/admin/settings/{crawler,analysis}` (RequireAdmin,
cookie-only, CSRF-guarded): GET returns the editable DTO + a read-only
env-managed view + prompt defaults; PUT validates (422 + per-field
details), persists + audits in one tx, then gracefully respawns the
affected daemon via a Supervisor.
- AppState swaps the crawler control/resync handles and analysis enable
gate behind shared runtime cells so reloads take effect with no restart;
a boot-time spawn failure is logged rather than aborting startup.
- Make the three vision prompts and sampling temperature configurable
(defaults preserved); cookie_domain is admin-editable too.
- Frontend: tabbed /admin/settings page with grouped fieldsets, prompt
reset-to-default, env-managed read-only panel, save-&-apply confirm, and
per-field validation; typed client + ApiError.details.
- Bump version 0.79.1 -> 0.80.0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Small vision models (qwen3-vl-4b) sometimes loop the OCR array, running
the response into the token ceiling (finish_reason: length) and failing to
parse. Layered guardrails:
- Prompts: explicit "transcribe each element once, never repeat/loop, at
most N, STOP when done" in the OCR, grounding and combined prompts.
- Schema: hard maxItems on ocr_results/tagging_results/content_type and
maxLength on text/scene — LM Studio's grammar enforces these, so a loop
is grammar-bounded rather than relying on max_tokens.
- Sampling: a configurable frequency_penalty (ANALYSIS_FREQUENCY_PENALTY,
default 0.3) sent with each request — the decode-time lever that actually
breaks loops.
- Code: sanitize() collapses runaway consecutive duplicates (same
normalized text + kind) so a loop that slips through still can't flood
the row.
Tests: schema caps present, frequency_penalty included only when nonzero,
sanitize collapses consecutive repeats; config default.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Boundary text duplicated across slices (often with slightly different,
cropped transcriptions) survived the text-only merge. The OCR pass now
asks for a per-piece vertical position and the merge uses it:
- OcrResult gains optional `y` (fraction 0..1 of the slice); the Pass-A
OCR schema/prompt request it (combined path leaves it None). Not
persisted.
- merge_ocr takes each slice's working-image y-band and maps `y` to a
page-global position. A seam pair is a duplicate when position-close
(same kind) OR text-similar, so a mis-OCR'd boundary line is caught even
when the text differs. The kept copy is the one more central in its slice
(less cropped); falls back to keep-longer when positions are missing.
- Self-calibrates the model's y values (pixels vs. fraction) and ignores
degenerate columns, so a bad localizer can't over-merge.
Tests: position pairs differing texts and keeps the less-cropped copy;
text-only fallback (dedup/keep-longer/non-adjacent/order) still holds.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Very tall webtoon pages were squashed by the fixed longest-edge downscale,
so OCR was garbage. The vision client now slices to the model's pixel
budget at native resolution and assembles one result:
- plan_slices: native-resolution bands sized to max_pixels (portrait OR
landscape), only when height/width > tall_aspect_threshold; min_slice_
height guards pathologically wide pages (reduce width); max_slices caps
the count (coarser fallback). Normal pages still take one combined call.
- Two-pass, OCR-first: Pass A OCRs each band (near-native), merge_ocr
stitches them with seam-scoped fuzzy dedup (keep the longer transcript,
preserve order/kind, don't collapse non-adjacent repeats); Pass B feeds
the whole downscaled image + the merged OCR text to get tags/scene/safety
grounded in the real dialogue. Only OCR is merged.
- Config: replace max_image_dim with max_pixels / min_slice_height /
slice_overlap / tall_aspect_threshold / max_slices (+ env_f64); bump
job_timeout default 180->600 (N sequential slice calls per page); raise
MAX_OCR_PIECES 60->200. New prompts/schemas: OCR_PROMPT/ocr_json_schema,
GROUNDING_PROMPT/grounding_json_schema.
Tests: plan_slices (single/portrait/landscape/pathological-wide/cap +
coverage/overlap), merge_ocr (seam dedup, keep-longer, non-adjacent
repeats, order), render_whole/render_slice, the new request builders, and
updated config defaults/env.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A text-dense page's full analysis JSON exceeded the 900-token output cap,
so the model stopped mid-object (finish_reason: length) and the truncated
response failed to parse ("EOF while parsing a list"). The page image +
prompt are only a few hundred tokens, so a larger output budget still fits
comfortably in an 8k context window. Still overridable via
ANALYSIS_MAX_TOKENS.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Broadcasts analysis progress so the dashboard updates live:
- analysis::events: AnalysisEvents broadcaster + AnalysisEvent
(Enqueued / Started / Completed / Failed), carrying the
manga/chapter/page breadcrumb.
- The worker daemon resolves each page's breadcrumb (repo::page::locate)
and publishes Started before dispatch and Completed/Failed after.
- The admin reenqueue publishes Enqueued (scoped by manga/chapter).
- GET /v1/admin/analysis/status/stream — SSE (RequireAdmin) forwarding
each event as a named `analysis` frame; broadcast lag emits a `lagged`
frame. AppState carries the always-present events bus.
Tests: worker publishes started+completed (with breadcrumb) and failed;
SSE route is admin-gated (403 non-admin) and returns text/event-stream.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The worker hardcoded response_format={type:json_object}, which LM Studio
rejects with 400 ('must be json_schema or text'); it also leaves
reasoning models (Gemma) with an empty `content`. Fix:
- Default to response_format=json_schema with the real analysis schema
(prompt::output_json_schema), which LM Studio accepts and which forces
clean JSON into message.content. Configurable via ANALYSIS_RESPONSE_FORMAT
= json_schema (default) | json_object | none (config::ResponseFormat).
- build_request_body is now a pure, unit-tested function.
- Vision errors now include the server's response body (status + text)
instead of a bare status, so a 400's reason is visible in logs.
Tests: response-format body shapes (none/json_object/json_schema), schema
shape + round-trip with the DTO, config parsing/defaults.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Generalizes the admin re-enqueue beyond global backfill:
- repo::page_analysis::enqueue_pages(scope, only_unanalyzed) with
ReenqueueScope::{All,Manga,Chapter}. Fixes include-analyzed semantics:
only_unanalyzed=false now enqueues jobs with force=true so the worker
actually re-analyzes already-done pages instead of skipping them.
- POST /v1/admin/analysis/reenqueue accepts optional manga_id / chapter_id
(mutually exclusive, 404 on unknown target) alongside only_unanalyzed.
Tests: manga/chapter scope counts, include-analyzed sets force=true on the
done page, mutual-exclusivity 422, unknown-target 404. Existing global
backfill tests still green (13 total).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Surfaces the analysis worker's NSFW moderation at the manga level:
- repo::manga FILTER_WHERE gains cw_include (AND, page-content-warning
join to mangas) and cw_exclude (NONE) clauses; ListQuery + binds
renumbered ($6/$7, LIMIT/OFFSET $8/$9).
- repo::page_analysis::warnings_for_manga: deduped, alphabetical union
across all the manga's pages.
- domain::MangaDetail gains content_warnings; get_detail populates it.
- api::mangas list accepts cw_include/cw_exclude (reusing the shared
parse_warnings_csv validator).
Tests: detail union (deduped/sorted) + empty case; list include/exclude
filter; unknown-warning 422. Existing manga tests still green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New GET /v1/me/page-search surfacing the analysis worker's output:
- repo::page_analysis::page_search + PageSearchQuery: multi-tag AND across
(user page_tags ∪ global page_auto_tags) via the unnest double-negative
idiom, weighted OCR/scene text ranking (ts_rank over search_doc), and
content-warning include/exclude. One row per page with is_nsfw +
deduped content_warnings + rank.
- domain::PageSearchItem.
- api::page_tags: /me/page-search handler with CSV tag/warning parsing
(parse_tags_csv reuses normalize_tag; parse_warnings_csv validates the
closed vocabulary), requiring at least one positive filter (422 else).
This is where the reserved OCR text search lands for pages.
Tests: multi-tag AND user∪auto, speech>sfx ranking, cw include/exclude
(+ row flags), text-only (no tags), missing-filter 422, unknown-warning
422, auth required.
Note: the /me/page-tags/chapters|mangas aggregations keep single-tag
behavior + the reserved text=501 for now; page-level search is the
primary text surface.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The OpenAI-compatible vision client and the bounded prompt/output handling
for the analysis worker (no DB, fully unit-tested):
- analysis::prompt: terse system prompt carrying the JSON schema, the OCR
kind / content-warning vocabularies, and the output-size caps.
- analysis::vision: VisionClient (downscale via the new `image` dep →
base64 data-URL → chat/completions with an image_url part), plus pure
parse_chat_completion (fence/prose-tolerant) and sanitize (drop empty
OCR, truncate, clamp tags to 10 via the shared page-tag normalizer,
filter unknown warnings).
- config::AnalysisConfig extended with endpoint/model/api_key/timeouts/
max_tokens/max_image_dim/max_image_bytes + from_env.
- Deps: add `image` (jpeg/png/webp), reqwest `json` feature. Expose
api::page_tags::normalize_tag as pub(crate) for reuse.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wires the analyze_page job kind into the page-create paths and adds admin
controls, all gated on a new ANALYSIS_ENABLED flag (AppState.analysis_enabled,
config::AnalysisConfig):
- Chapter upload (api::chapters) enqueues one analyze_page job per page
after the tx commits (best-effort; failures logged, never fatal).
- Crawler content sync (persist_pages → RETURNING id) enqueues per
persisted page when the daemon dispatcher's analysis_enabled flag is set;
threaded through sync_chapter_content (CLI/resync pass false).
- repo::page_analysis::enqueue_all_pages: bulk backfill via INSERT..SELECT,
skipping done pages (only_unanalyzed) and existing pending jobs.
- New admin endpoints (RequireAdmin, 503 when disabled, audited):
POST /admin/analysis/reenqueue and POST /admin/pages/:id/analyze (force).
Tests: upload enqueues N jobs / no-op when disabled; crawler persist_pages
enqueue gate; admin reenqueue (backfill, idempotent, only-unanalyzed, 503,
non-admin 403) and force-analyze (force flag, 404).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds the persistence foundation for the AI content-analysis worker:
- Migration 0025: page_analysis (status + scene + is_nsfw + weighted
search_doc tsvector), page_ocr_text (kind-tagged pieces), global
page_auto_tags (shared tags vocabulary, separate from per-user
page_tags), and page_content_warnings (closed vocabulary).
- domain::page_analysis: AnalysisStatus/OcrKind/ContentWarning enums with
kind->tsvector-weight mapping and lenient model-string parsing, plus the
VisionAnalysis response DTOs.
- JobPayload::AnalyzePage { page_id, force } + KIND_ANALYZE_PAGE, reusing
the existing crawler_jobs queue.
- repo::page_analysis: enqueue_for_page, load, mark_failed, and the single
transactional persist_analysis (delete+reinsert; idempotent; never
touches user page_tags; computes the A/B/C/D-weighted search_doc).
Tests: repo integration (persist/idempotency/user-tags-untouched/enqueue/
mark_failed/speech>sfx ranking) + unit (weights, parsing, DTO, job serde).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
page_tags previously stored the tag inline as free-form text — the lone
un-normalized tag concept, while authors/genres/manga-tags all went
through a lookup table + FK. Fold page tags into the SAME shared `tags`
table that manga_tags uses: `page_tags.tag` becomes `page_tags.tag_id`
referencing `tags(id)`. Manga tags and page tags now share one global
vocabulary.
The HTTP contract is unchanged — the API still speaks tag *names*; the
repo resolves name<->id via `repo::tag::upsert_by_name` (the manga-tag
helper) and keeps the stricter page-tag `normalize_tag` at the boundary.
Frontend, DTOs, and response shapes are untouched. User-visible effect:
page-tag names now appear in the manga-tag autocomplete and vice versa.
Migration 0024 backfills `tags` from existing inline values (deduping
case-insensitively), repoints rows, swaps the unique/secondary indexes
to tag_id, and drops the inline column.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add GET /v1/mangas/:id/similar returning the top 5 mangas ranked by
Jaccard tag overlap (shared / union), excluding self, as MangaCard items
in a plain { items } object. Surface them in a hidden-when-empty
"Similar" section on the manga detail page, reusing MangaCard.
Backend: new repo::manga::list_similar with a manga_tags self-join; a
shared cards_from_rows helper (also used by list_cards) that loads
authors+genres concurrently; single-sourced column list via MANGA_COLS +
manga_cols(alias) to replace the fragile SELECT_COLS string-splitting.
Frontend: getSimilarMangas client fn (guards a malformed body), loader
fetch with non-critical .catch fallback, and the catalog grid hoisted to
a shared global .manga-grid (lib/styles/tokens.css), de-duplicating the
per-route copies.
Bump version 0.62.0 -> 0.63.0 (backend + frontend in lockstep).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add the /search surface (Pages / Chapters / Mangas tabs) backed by
per-user page tags and per-page collections: schema (migration 0023),
backend endpoints for page tags/collections and tagged-page aggregations
(with the OCR text-search param reserved at 501), plus the frontend API
clients, library Page-tags tab, collection page sections, page context
menu / AddTagsSheet, and reader long-press wiring. Includes the
continuous-reader navigation fixes (?page=N handling, chapter-reset
timing, back-button pops history) and tag-normalization hardening
accumulated on the branch.
Bump version 0.60.2 -> 0.62.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Admin-only crawler dashboard backed by an SSE live-status stream,
coordinated browser restart, runtime PHPSESSID refresh, dead-letter
requeue, and a batch of reliability fixes. Closes everything from
the two-pass audit (10 commits' worth) and bumps 0.52.0 -> 0.55.0.
Backend:
- New /admin/crawler/* surface (cookie-auth, RequireAdmin) split
into status / control / dead_jobs / backlog modules. SSE stream
composes in-memory status with DB-derived queue counts, memoizes
the counts for 1s and debounces watch pokes for 250ms (~10x QPS
reduction per subscriber). One-shot GET /admin/crawler shares the
same compose path.
- POST /admin/crawler/run gated by manual_pass_lock try_lock_owned
(409 Conflict on overlapping click); browser restart goes through
the coordinated_restart gate (drain + relaunch + auto-clear of the
sticky session_expired flag on Ok).
- Runtime PHPSESSID refresh via SessionController (allow-list
validation, never logged, audit row carries SHA-256 fingerprint).
Storage layer is repo::crawler::runtime_session_{load,persist}.
- Dead-letter requeue with four scopes (all/manga/chapter/job);
scope=all requires confirm:true; DISTINCT ON dedup keeps the
partial unique index from rejecting requeues for chapters with
multiple dead rows. SQL is four &'static str constants per scope.
- StatusHandle + ChapterGuard / CoverGuard RAII model survives
panics; last-writer-wins on cover so concurrent dispatches don't
clobber each other's slot. Pure functions (should_stop /
should_mark_clean_exit / should_abort_pass) with named regression
tests.
- Reliability bundle: per-lease heartbeat, jitter on retries,
per-job timeout, circuit breaker on consecutive failures, BrowserManager
coordinated restart gate, request fingerprint changes.
- Streaming page download: Storage::put_stream trait method,
LocalStorage impl atomic via temp + fsync + UUID-suffixed rename.
Pages stream through with peak memory ~one HTTP chunk + 64-byte
sniff prefix instead of one full image per dispatch.
- New partial indexes (migration 0022): mangas_missing_cover_idx
and crawler_jobs_dead_idx, both ordered by updated_at DESC to
match the dashboard's LIMIT/OFFSET reads.
- Security hardening: admin_csrf_guard (Origin/Referer allowlist
on /admin/* mutations, opt-in via ADMIN_ALLOWED_ORIGINS),
admin_no_store_guard (Cache-Control: no-store on admin
responses), audit rows carry per-scope target_id.
Frontend:
- /admin/crawler page decomposed into lib/components/crawler/
(11 components: ProgressBar, SearchBar, CrawlerHero,
CrawlerControls, ActiveChaptersCard, ActiveJobsTable,
MissingCoversTable, DeadJobsTable, RestartConfirmModal,
RequeueAllConfirmModal, SessionModal). Page is 532 LOC of
orchestration; each component 22-148 LOC.
- EventSource lifecycle wired to visibilitychange / pagehide /
pageshow (BFCache); after 5 consecutive errors probes the status
endpoint so a 401 routes through the global on401Hook instead of
infinite silent reconnects.
- Backlog $effect refetches debounced 500ms with per-loader
AbortControllers; refresh after a control action only runs when
the SSE stream is dead.
- Inline requeue button on /admin/mangas patches the affected row's
sync_state locally (no full chapter-list refetch); proper
aria-label. Requeue-all gets its own confirm modal; both confirm
modals autofocus Cancel.
- SvelteKit reverse proxy bypasses its 5-minute AbortController
for Accept: text/event-stream; pure shouldBypassProxyTimeout
helper covered by unit tests.
Config / docs:
- New env vars (.env.example): ADMIN_ALLOWED_ORIGINS,
CRAWLER_JOB_TIMEOUT_SECS, CRAWLER_METADATA_MAX_CONSECUTIVE_FAILURES,
CRAWLER_BROWSER_RESTART_THRESHOLD.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The user-facing chapter list ordered by (number ASC, created_at ASC),
which broke the source site's order in two ways: non-numeric entries
("notice. : Officials") parsed to number=0 and clustered at the top,
even though the site placed them mid-list, and variants sharing a
number ("Ch.14 : PH" / "Ch.14 : Official") were torn apart by the
created_at tiebreak.
Capture each chapter's position in the source DOM as `source_index`
(0 = first = newest on this site) on every crawler sync, including the
UPDATE branch so a new chapter prepended on the source shifts every
existing row down by one on the next tick. The list query reverses
this with `ORDER BY source_index DESC NULLS LAST, number ASC,
created_at ASC` so the oldest chapter appears first, variants stay
adjacent in the order the site shows them, and non-numeric entries
land where the site placed them. User-uploaded chapters and pre-
migration rows keep their NULL source_index and fall through to the
prior number/created_at tiebreak via NULLS LAST.
The reader's client-side `[...chapters].sort((a,b) => a.number - b.number)`
is dropped; prev/next now walks the server-ordered array positionally
so it traverses variants and non-numeric entries in display order.
Existing data populates on the next cron tick or via admin force-resync.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both enqueue paths now order by chapters.number so the cron tick and the
bookmark hook insert jobs from chapter 1 upward instead of source-discovery
or random-UUID order. The lease query tiebreaks on created_at so jobs
sharing a batch's scheduled_at come off the queue in insertion order,
propagating the enqueue intent through to dequeue. Concurrent workers
and per-CDN latency can still drift actual completion order.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a per-tick cover-backfill pass to the crawler daemon so mangas whose
cover download failed on first attempt get retried — the metadata pass's
early-stop optimisation otherwise prevents the walk from revisiting them.
Adds admin-only POST /admin/mangas/:id/resync and POST /admin/chapters/:id/resync
that refetch metadata + cover (or chapter content with force_refetch) from the
crawler source synchronously and return the refreshed row. Surfaced in the
UI as "Force resync" buttons on the manga detail and reader pages,
admin-only via session.user.is_admin.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When `PRIVATE_MODE=true`, every API path except a small allowlist
(`/health`, `/auth/{config,login,logout,register}`) requires a valid
session cookie or bearer token — anonymous reads are rejected with
401. Self-registration is force-disabled in private mode regardless
of `ALLOW_SELF_REGISTER`, so a locked-down instance flips with a
single switch (admins still mint accounts via `POST /admin/users`).
The backend gate is a tower middleware that reuses the existing
`CurrentUser` extractor, so the cookie + bearer paths cannot drift
from per-handler auth. `/auth/config` now exposes the flag plus the
effective `self_register_enabled` value so the frontend can render
the navbar correctly on the first paint.
On the frontend, a new universal root `+layout.ts` fetches the
config and redirects anonymous visitors to `/login?next=<path>`
before page-specific loads fire. The redirect is UX only — the
backend middleware is the source of truth, so crafted requests
still 401.
Defaults stay public (`PRIVATE_MODE=false`); existing deployments
need no env change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The CLI binary already capped runs at CRAWLER_LIMIT mangas, but the
daemon's RealMetadataPass passed a hardcoded `0` (no cap) to
`pipeline::run_metadata_pass`, so the env var was silently ignored once
the daemon took over the metadata pass.
Adds `manga_limit` to `CrawlerConfig`, reads it from `CRAWLER_LIMIT`
(default 0 = no cap), and threads it through `RealMetadataPass::run`
so a daemon-driven sweep stops at the same boundary as a CLI run.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Chromium doesn't know the socks5h scheme (curl/reqwest convention)
and bails navigations with ERR_NO_SUPPORTED_PROXIES. It does, however,
send destination hostnames over SOCKS5 by default, so stripping the
`h` is a pure scheme rename — remote-DNS behaviour is preserved.
reqwest keeps the user's original CRAWLER_PROXY string (`socks5h://...`
remains valid and meaningful for it).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The startup log line in app.rs and bin/crawler.rs `?t`-debug-formats
the TorController, which through the derived Debug on TorAuth would
expand TorAuth::Password(p) and leak the plaintext password to logs.
Implement Debug manually on TorAuth — None / Password(<redacted>) /
Cookie(<path>) — and lock the redaction with a regression test.
Drop the per-NEWNYM success log from info to debug: a busy crawl
rotates circuits many times per minute. Failed NEWNYMs already log
at warn — those stay loud.
Tightens the closed_connection_mid_reply_is_an_error assertion which
was tautological (`closed connection` OR `AUTHENTICATE`) by driving
the mock to read the AUTH line then drop, exercising only the
EOF-mid-reply path.
Audit ref: #7, #9, nit on tautological test.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The three retry-with-recircuit sites disagreed: detect.rs's
retry_on_transient_with_hook used "N = total attempts" (3 → 3
fetches), but session.rs's unauth branch and content.rs's chapter
loop used "N = recircuits" (3 → 4 fetches). At the same wall-clock
"max=3", different sites hit the upstream a different number of times.
Unify on N = total attempts (matching the existing
retry_on_transient convention). The CRAWLER_TOR_RECIRCUIT_MAX_ATTEMPTS
env var now means exactly what its name suggests. Disabling the
recircuit feature collapses to max_attempts=1 (single attempt, no
retry) — bit-for-bit pre-TOR behavior preserved.
Adds a debug_assert!(max >= 1) on both helpers and a new
content.rs test exercising the mixed Transient → Unauth → Ok
sequence to lock in the shared-counter invariant.
Audit ref: #5.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- target.rs swaps retry_on_transient → retry_on_transient_with_hook,
signaling NEWNYM via ctx.tor between attempts when configured.
- session.rs gains verify_session_with_recircuit; the bare
verify_session is now a one-line wrapper passing tor=None,
unauth_max_recircuit=0. The inner run_session_probe_loop is
pure-over-IO and unit-tested with closure-based fakes.
- content.rs extracts fetch_chapter_html_once + the closure-driven
fetch_chapter_html_with_recircuit, used by sync_chapter_content to
retry on Transient or Unauthenticated up to a recircuit_budget.
Budget = 0 (no TOR) preserves original behavior bit-for-bit.
- app.rs and bin/crawler.rs construct the controller before on_launch
and pass it into verify_session_with_recircuit, so a transient
hiccup at startup no longer requires PHPSESSID rotation.
Recircuit budget defaults to CRAWLER_TOR_RECIRCUIT_MAX_ATTEMPTS (3).
Errors from NEWNYM are logged and swallowed — failing to recircuit
should not take down the crawl.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds CRAWLER_TOR_CONTROL_URL / _PASSWORD / _COOKIE_PATH /
_RECIRCUIT_MAX_ATTEMPTS to CrawlerConfig and to bin/crawler.rs's
env reads. Constructs an Option<Arc<TorController>> at daemon /
CLI startup and threads it through FetchContext,
pipeline::run_metadata_pass, and content::sync_chapter_content as
Option<&TorController>.
Pure scaffolding — the controller isn't used yet; behavior is
unchanged. Next commit wires the retry hooks and session-probe
recircuit.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a sibling fn that fires a caller-supplied async hook between a
transient failure and the next attempt. The existing
retry_on_transient becomes a thin wrapper over it (no-op hook), so
no call sites churn yet.
Hook contract: fires only between attempts (N-1 times for N
attempts), never after a non-transient error or after the final
attempt. Designed for TOR NEWNYM, but the signature is generic.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Minimal client over tokio::net::TcpStream — AUTHENTICATE then
SIGNAL NEWNYM, one-shot connection. Supports cookie-file and
password auth (cookie preferred when both provided); covers the
multi-line `250-...\r\n250 OK` reply form so future torrc tweaks
won't confuse the parser.
Not yet wired into the crawler — that lands in the next commits.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Operators whose sources shard images across numbered CDN subdomains
can't pre-enumerate every host in CRAWLER_DOWNLOAD_ALLOWLIST. The new
flag short-circuits the host check in DownloadAllowlist::contains
while leaving scheme, localhost, and private-IP defenses in
is_safe_url untouched — scraped URLs pointing at 10.x /
169.254.169.254 / file:// stay refused. Default is false; fail-closed
posture is preserved unless the operator opts in. Wired into both the
server (config::build_download_allowlist) and the bin/crawler.rs
one-shot.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Handle::close aborts its chromiumoxide driver task when another
Arc<Browser> outlives the call, so shutdown returns instead of
hanging on a stream that never terminates. Generic close_or_abort
helper with regression tests covering both Arc paths.
- daemon.shutdown() is wrapped in a 5s timeout in main as defense
in depth.
- Default RUST_LOG silences chromiumoxide::conn / chromiumoxide::handler
WS-deserialize ERROR spam.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pairs with the ALLOW_SELF_REGISTER toggle from 0.42.0: admins can mint
accounts regardless of the toggle state, so a closed-membership
deployment still has a working enrollment path. The endpoint accepts
{ username, password, is_admin? } so admins can mint co-admins in one
call (avoiding a separate promote + extra audit row for the common
"invite a co-admin" flow).
Implementation:
- POST /api/v1/admin/users guarded by RequireAdmin
- Reuses validate_username / validate_password from api::auth (made
pub(crate)) so the admin path can never produce an account self-
register would reject and vice versa
- repo::user::admin_create_user wraps INSERT + admin_audit insert in
a single tx — same "audit reflects what committed" semantics as the
existing admin_safe_* fns
- Audit row: action="create_user", payload={username, is_admin}
Frontend:
- createAdminUser() in lib/api/admin.ts
- /admin/users grows a collapsible "Create user" form above the table
(username, password, "Make admin" checkbox). Errors surface inline;
the list reloads on success.
Backend tests: 7 new, including the headline
`create_user_works_even_when_self_register_disabled` that pins the
admin-create path is NOT gated by the public toggle.
Lets operators run a closed-membership deployment by setting
ALLOW_SELF_REGISTER=false (default true, so existing deploys are
unaffected). When off, POST /auth/register returns 403 forbidden. The
rate-limit token is consumed BEFORE the disabled check so the timing
doesn't distinguish enabled-but-rejected from disabled — closes the
toggle-state probe channel.
New public GET /auth/config returns { self_register_enabled: bool }
so the frontend can render its register affordances correctly
without conflating "disabled" with "rate-limited" (which a probe
attempt would).
Frontend: a lightweight reactive `authConfig` store loads the flag
once on root-layout mount (and again on /register direct navigation,
which bypasses the layout's onMount). Header hides the Register link
when the toggle is off; /register renders a "self-registration is
disabled — ask an administrator" notice instead of the form.
Admin-create endpoint that pairs with this toggle is intentionally
not in this PR — it lands as the next branch (feat/admin-user-create).
The toggle alone is independently useful for deployments that want
to lock down enrollment without yet wiring an admin UI.
Addresses the security-audit findings on top of the admin feature stack:
M1: /admin/mangas/:id/chapters now paginates (default limit 200, max 500).
A long-runner with thousands of chapters would otherwise produce a multi-MB
response with that many scalar subqueries per row — admin-only but a real
stall risk on one expand-click. Adds explicit pagination tests for the cap
and offset; frontend renders a "Showing first N of M" hint when the cap
clips the result.
L1: repo::user::set_is_admin renamed to set_is_admin_unchecked with a
doc-comment pointing at admin_safe_set_is_admin for production use. The
short name was a footgun — a future contributor reaching for it would
silently bypass self-protection, the last-admin invariant, and the audit
log. Used only by integration-test setup; production code goes through
the admin_safe_* paths.
CSRF posture: build_session_cookie carries a comment that the
SameSite=Lax default is the project's CSRF defense for state-changing
mutations and breaks the instant anyone adds a side-effecting GET under
/admin/*. Spells out what to do then (Strict + explicit token check).
Test counts: 43 backend admin tests + 12 vitest admin tests all green;
svelte-check 0/0 across 446 files.
- admin_safe_set_is_admin: short-circuit when target.is_admin == value,
before writing audit. PATCH {is_admin: true} on someone already admin
previously wrote a misleading "promote_user" row even though the UPDATE
was a no-op.
- list_chapters (/admin/mangas/:id/chapters): explicit exists() check on
manga_id, returns 404 instead of 200 [] for a typo'd / deleted manga.
- ChapterSyncState priority: the Failed branch now requires page_count = 0,
so a chapter with pages on disk AND a historical dead job (from a
re-download attempt that crashed) stays Synced. The old order
contradicted Synced's documented "downloaded at some point" contract.
Doc comments updated alongside the SQL.
Three new regression tests pin the behaviour.
