Commit Graph

20 Commits

Author SHA1 Message Date
ShukeBta 4e0929bbca fix organize path mapping and task flow 2026-06-13 16:32:38 +08:00
ShukeBta 1568ae127a perf: 重构 FTS 索引为 rowid 寻址+触发器维护,根治重启 CPU 占满与扫描卡死
- FTS5 普通列(含 UNINDEXED)不支持索引查找,旧版按 media_id 做
  NOT EXISTS/DELETE 全是整表扫描:启动回填 O(N^2) 烧 CPU 数小时,
  扫描时每个 upsert 一次全表扫描,均隔着全局写锁拖死登录(125s 超时)
- v2 布局:FTS rowid 与 media.rowid 对齐,由 INSERT/UPDATE/DELETE
  触发器实时维护;回填只剩 rowid 点查兜底,且去掉 ORDER BY
- 顺带修复刮削直写 Updates() 后新标题搜不到的问题(触发器覆盖)
- 写门改为 context 感知:长写语句不再让登录请求无限期排队
- warmMediaSearchIndex 延迟 30s 错峰启动
- library_scan 周期任务跳过云盘库(由 cloud_sync 夜间窗口负责),
  首轮延迟从 15s 改为等满一个周期,重启不再立即全量扫描风暴
- prune 改为只取 id/path 并按 500 条批量删除,缩短写锁占用
2026-06-12 06:41:14 +00:00
ShukeBta 91df7521ce fix: 资源占用/登录稳定性/QB整理入库/第三方播放404 综合修复
资源占用(Docker 部署 CPU/内存长期居高):
- 云盘探测预算改为按尝试扣减,杜绝队列满时对每个文件反复入队
  并刷出数万条 WARN(实测日志 41165 条)
- 探测队列满时给文件挂 30 分钟退避 + 告警限速为每分钟一条
- 扫描时每个文件的海报/背景图由同步下载(单张最长 20s)改为
  后台预取队列,云盘大库扫描不再串行拉图数小时
- PlaybackInfo 的云盘 ffprobe 探测改异步(原同步最长 8s,
  既拖慢起播又放大云盘流量),带单飞去重
- 访问日志跳过 /api/health 与静态资源;logging.level/format
  配置真正生效(此前是死配置)

登录稳定性(经常登录报错):
- refresh token 未及时落库期间,刷新请求可识别「待落库令牌」,
  不再把用户踢回登录页;轮换/登出后取消后台补写,防止旧令牌复活

QB 下载整理入库:
- 新增 download.path_mappings 设置:自定义下载器→本程序路径映射
  (每行 客户端路径=本地路径),并复用 compose 环境变量映射规则
- 应用重启后补整理最近 24h 内完成的种子(此前重启即永久漏掉)
- 下载客户端初始化失败仍注册并惰性重连(容器启动顺序免疫)
- 硬链接跨文件系统(EXDEV)自动降级为复制,保种语义不变

第三方播放器 404:
- 播放处理器不再把所有错误吞成 404:媒体不存在→404,
  云盘解析失败/STRM 关闭→502+原因
- 存库的云盘播放 URL 规范化为相对路径,免疫扫描时固化的旧 host
- 云盘媒体 SupportsDirectPlay=false,强制走带鉴权的 DirectStream
2026-06-12 04:31:34 +00:00
ShukeBta 23d1fbfddb fix cloud library scale and playback compatibility
Root cause: display-only cloud library filtering was reused by scan jobs, merged cloud mounts could be skipped, and OpenList listing relied on WebDAV/first-page behavior that could cap huge directories around 100 items. Cloud scans also let new items consume probe budget before existing rows with missing track/artwork metadata.\n\nChanges:\n- split display filtering from scannable cloud filtering so merged cloud mounts still scan\n- add OpenList API pagination with WebDAV fallback\n- prioritize existing cloud media missing metadata before new imports\n- restrict automatic cloud sync to one successful 19:00-21:00 daily window and keep manual scan immediate\n- disable startup cloud full-scan by default, with an explicit opt-in setting\n- improve Emby/cloud playback compatibility and cache/search/test coverage from the continued work
2026-06-11 18:31:01 +08:00
ShukeBta c22d8271ae fix: use Get instead of FindByID to lookup storage config by type
The scanner was incorrectly using FindByID(mount.Provider) where mount.Provider
is a type string like 'openlist', not a database ID. This caused all cloud
library scans to fail with 'storage config not found'.

Changed to use Get(ctx, mount.Provider) which correctly looks up by type field.

Fixes cloud library scanning for all providers (openlist/quark/115/clouddrive2)
2026-06-11 13:21:37 +08:00
ShukeBta 95794f7bbc fix: add storage deletion with library cleanup and scan validation
- Delete storage config now properly removes related libraries and media
- Scanner validates storage config exists before starting cloud scan
- Prevents orphaned scans when storage is removed
- Added repository methods: DeleteByType, FindByID for storage config
- Remove unused storageHandler route to fix compilation
2026-06-11 12:48:32 +08:00
ShukeBta 62b204367c fix: auto-scan cloud libraries on boot + allow CORS for media playback
- Add BootCloudLibraries() to auto-scan all cloud libraries on startup
- Delay 3s to avoid conflict with system init, scan without auto-scrape
- Enable CORS for /api/cloud/play/* and /api/img to support 3rd-party players
- Fixes issue where each user triggers separate cloud library scans
- Fixes issue where Infuse/Emby apps cannot play cloud resources
2026-06-11 11:25:43 +08:00
ShukeBta d90b58ba22 fix: cache cloud artwork before library import 2026-06-11 00:54:29 +08:00
ShukeBta 542e85a067 fix: reduce emby load and stabilize strm scan 2026-06-11 00:18:51 +08:00
ShukeBta 859d6e9d24 implement automatic strm generation 2026-06-10 21:36:55 +08:00
ShukeBta fd0428ee7d fix cloud library mounts and artwork caching 2026-06-10 20:43:42 +08:00
ShukeBta 0bf983cec8 fix: improve cloud mount scanning feedback 2026-06-10 18:46:30 +08:00
ShukeBta 0f82cb3084 feat: improve OpenList cloud storage support 2026-06-10 00:08:35 +08:00
ShukeBta 7d7f3cc758 feat: add cloud transfer and cache optimizations 2026-06-09 19:03:28 +08:00
ShukeBta 76fe0dbde9 feat: scrape metadata after organize 2026-06-09 17:00:24 +08:00
soldosluka857 b9b7b7052a feat(organize/scan): transfer modes, seeding-safe relocation, inode dedup, incremental scanning
- Organizer: add move/copy/hardlink/symlink transfer modes (default move),
  per-request target_path/transfer_mode overrides, and honor organize.target_dir
  / organize.transfer_mode settings.
- keep_seeding (default on): escalate move->hardlink (cross-device->copy) so the
  qBittorrent source stays in place and continues seeding after organize.
- qBittorrent SetLocation + POST /downloads/relocate to migrate whole torrents
  while keeping them seeding.
- Scanner: FileID (device:inode) hardlink dedup to avoid duplicate recognition
  and double-counted storage; extract single-file ingest.
- Watcher: recursive watch + incremental per-file ingest/remove instead of full
  re-scan; periodic full library scan now gated behind scan.periodic_enabled
  (default off) to reduce disk wear.
- Telegram bot: handle callback_query in polling, declare allowed_updates in
  webhook, answer callbacks.
- Frontend: settings for transfer mode / keep_seeding / periodic scan; organize
  panel target dir + transfer mode overrides.
- Tests for transfer modes, organizer resolution, SetLocation, inode dedup,
  incremental ingest/remove.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-05-30 12:01:46 +08:00
ShukeBta 32ccb33fed Fix local adult metadata and watch history removal 2026-05-28 14:52:20 +08:00
Kiro 7bd6bcfb1b feat: downloads, RSS, subtitles, Bangumi, watcher, stats, profile, audit
Backend
  - service/bangumi.go: Bangumi (bgm.tv) scraper for anime libraries.
  - service/episode_parser.go: SxxExx / NxE / EPxx / 第NN集 parser with
    unit tests; consumed by the scanner for tv/anime libraries.
  - service/scraper.go: provider chain orchestrator picks Bangumi for
    anime libraries (TMDb fallback), TMDb for everything else; emits
    'no_match' rows so we don't retry forever; AnyEnabled() for the
    scanner kick.
  - service/scanner.go: writes season/episode numbers for tv/anime libs
    and reports a 'probed' counter alongside 'added'.
  - service/transcoder.go (existing): unchanged, stays per-media.
  - service/subtitle.go: discovers external subtitles next to the source
    file (.srt/.vtt/.ass/.ssa) and converts SRT/ASS to WebVTT on the fly.
  - service/qbittorrent.go: thread-safe qBittorrent v2 Web UI client
    (login / add / list / delete) with cookie-jar reuse.
  - service/downloads.go: persists download tasks, reads runtime
    qbittorrent.* settings, polls /torrents/info every 5 s and pushes
    the result to WS subscribers ('download' topic).
  - service/subscription.go: 10-minute RSS poller with regex filter,
    GUID dedup persisted in the settings table, and 'subscription' WS
    events on enqueue.
  - service/watcher.go: fsnotify watcher with 5 s coalescing debouncer
    that triggers per-library rescans on create/rename/remove.
  - service/stats.go: dashboard snapshot — totals, recently added,
    gopsutil-driven CPU/mem/disk readings.
  - service/profile.go: non-credential profile patch + admin role mutator.
  - service/audit.go: best-effort writer for the access_logs table.
  - service/service.go: container wires every new service; Boot() spins
    up watcher / downloads poller / subscription scheduler; Close()
    tears them down on graceful shutdown.

Handlers
  - new files: downloads.go, subscriptions.go, subtitles.go, series.go,
    stats.go, profile.go, util.go.
  - handler.go: registers PATCH /me, /libraries/:id/seasons,
    /media/:id/subtitles, /subtitles/:id, /downloads*, /subscriptions*,
    /stats, /admin/users/:id/role.
  - auth.go / media.go: write audit rows for login + library CRUD and
    refresh the watcher when libraries change.

Frontend
  - api: new helpers for downloads, subscriptions, profile, series, stats,
    subtitles; library helper gained scrape().
  - hooks/useWebSocket.ts: shared connection with 3 s reconnect.
  - components/GlobalEvents.tsx: surfaces scan / scrape / subscription
    completion as toasts (mounted at app root).
  - pages: Library now switches to a season-grouped layout for tv/anime
    libraries; Player attaches WebVTT <track> elements; new pages for
    Downloads (live torrent table), Subscriptions, Profile, Stats.
  - components/Layout.tsx + App.tsx: sidebar groups (媒体库 / 自动化 /
    账号 / 管理) and routes for the new pages; /stats and /admin remain
    admin-only.
  - types/index.ts: new types — Subscription, DownloadTask, QBitTorrent,
    Hardware, StatsSnapshot.

Verified: go build, go vet, go test (incl. ParseEpisode + srtToVTT +
stripASSTags) all pass; tsc -b && vite build emits 17 route chunks plus
the deferred hls chunk; main bundle 247 KB / 83 KB gzipped.
2026-05-14 16:07:49 +00:00
Kiro 0f30c34463 feat: ffprobe + TMDb scrape + HLS transcode + history/favourites/playlists
Backend
  - service/ffprobe.go: thin ffprobe wrapper, parses duration / resolution /
    codecs into a typed ProbeResult. 30s per-file timeout.
  - service/tmdb.go: minimal TMDb provider (search/movie). Disabled when no
    api key; supports tmdb_api_proxy / tmdb_image_proxy overrides for users
    behind a firewall.
  - service/scraper.go: filename cleaner (handles bracketed tags, scene
    noise tokens, year extraction), per-row + per-library enrichment with a
    4 RPS throttle and WS hub progress events. Unit-tested.
  - service/scanner.go: now invokes ffprobe per file and kicks the TMDb
    scraper in the background once a library scan finishes.
  - service/transcoder.go: per-media ffmpeg HLS job manager; outputs
    index.m3u8 + seg_NNNNN.ts under cache/hls/<id>; cancels jobs on
    shutdown; publishes 'transcode' WS events.
  - service/stream.go: serves HLS playlist (with 30s wait-for-ready) and
    .ts segments with path-traversal protection. Adds Probe() helper used
    by the admin 'reprobe' button.
  - service/image_proxy.go: cached, host-allow-listed reverse proxy for
    TMDb / Bangumi / Douban / Fanart / TheTVDB images so the SPA never
    hits a CORS or GFW issue.
  - service/playback.go: history upsert, favourites toggle, playlist CRUD
    + ordered items. RecentHistory joins with model.Media in one extra
    query so the home page can render a 'Continue Watching' row.
  - handler/streaming.go + handler/playback.go: REST endpoints for HLS,
    image proxy, scrape (one + library), reprobe, history, favourites,
    playlists.
  - handler/handler.go: registers /api/hls/:id/{index.m3u8,:seg}, /api/img,
    /api/history, /api/favourites/:id, /api/playlists/* with proper
    auth/admin guards.

Frontend
  - api/client.ts: imageURL() helper; hlsURL() endpoint; reuses the JWT in
    a query parameter for <video src> and <img src>.
  - api/playback.ts: typed helpers for history, favourites, playlists.
  - components/MediaCard.tsx: optional 'progress' prop renders a thin
    bottom progress bar, used by the new Continue Watching row.
  - pages/HomePage.tsx: two rows (Continue Watching + Recently Added);
    falls back to the empty-state hint when both are empty.
  - pages/PlayerPage.tsx: hls.js (lazy-imported) with auto-fallback to
    direct play; ?mode=hls|direct query toggle; resume position written
    every 10s while playing.
  - pages/MediaDetailPage.tsx: heart toggle + admin 'rescrape' / 'reprobe'
    buttons + dedicated 'HLS 转码播放' CTA.
  - pages/FavouritesPage.tsx, PlaylistsPage.tsx, PlaylistDetailPage.tsx:
    new screens.
  - components/Layout.tsx + App.tsx: sidebar links for Favourites and
    Playlists; routes are now lazily code-split via React.lazy + Suspense
    so the initial bundle stays at ~243 KB / 82 KB gzipped (hls.js is
    fetched only on first HLS playback).

Verified: go build, go vet, go test (incl. CleanQuery cases) all pass;
frontend tsc -b && vite build emits 9 route chunks plus a deferred hls
chunk.
2026-05-14 15:44:31 +00:00
Kiro d5cf5fb4b2 feat: bootstrap MediaStationGo (Go + React rewrite of MediaStation)
Adopt the cropflre/nowen-video tech stack and rebuild the project from
scratch:

  - Backend: Go 1.25 + Gin + GORM + SQLite (WAL) + JWT + WebSocket hub.
    Layered packages config / database / model / repository / service /
    middleware / handler. Default admin (admin/admin123) seeded on first
    run; /api routes for auth, libraries, media, stream and admin
    panels. WebSocket scan-progress events at /api/ws.
  - Frontend: React 18 + Vite 5 + Tailwind 3.4 + Zustand + axios +
    react-router 6 + lucide-react + framer-motion + hls.js. Pages for
    Login / Home / Library / Search / MediaDetail / Player / Admin
    (Library, Users, Settings tabs).
  - Distribution: multi-arch Dockerfile (frontend -> backend -> Alpine
    runtime), docker-compose.yml, GitHub Actions for CI and GHCR
    publish, Makefile, env-prefixed config (MEDIASTATION_*).
  - Docs: README, CONTRIBUTING, .env.example, config.example.yaml.

Backend builds, vets and tests pass. Frontend builds via tsc -b && vite
build (250 kB JS / 16 kB CSS, gzipped 84 / 4 kB).
2026-05-14 15:26:29 +00:00