- Add composite indexes on userEpisodeWatches and userMovieWatches for hot queries
- Batch episode tracking: wrap season/batch watches in single transaction (~8 queries vs 8*N)
- Fix N+1 patterns in credits, recommendations, and filmography with batch prefetch+insert
- Stream TV season hydration via Suspense instead of blocking page render
- Optimize webhook logs (per-connection LIMIT 10) and system health queries
- Merge genre filter waterfalls into single Promise.all fetch
- Migrate deprecated Jotai loadable() to unwrap()
- Replace isolated createStore()+Provider with useHydrateAtoms on root store
- Remove unnecessary atomWithStorage SSR guards (handled by Jotai internally)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Coalesces concurrent imports via an in-flight promise map, and catches
SQLITE_CONSTRAINT_UNIQUE as a safety net to return the existing row.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Motion's initial="hidden" rendered opacity:0 in server HTML, making content
invisible until JS hydrated — causing a multi-second blank gap after skeleton
disappearance. Replaced with CSS @keyframes stagger animation that plays
immediately from SSR. Also made color extraction fire-and-forget, merged
duplicate stats queries, and removed redundant dashboard loading.tsx.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Integrate TMDB credits data into the app: cast carousels on title detail
pages, person detail pages with biography and filmography, and person
search results in the command palette. Includes new persons/titleCast
DB tables, profile image caching, and a nightly credits refresh cron job.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fetch trailer video keys from TMDB videos API during title import/refresh,
store in DB, and display a "Trailer" button on title pages that opens a
media-chrome Sutro-themed YouTube player in a fullwidth dialog.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add try/catch around `request.json()` in all POST routes to return a
400 instead of crashing on malformed bodies. Validate `type` as a
strict `"movie" | "tv"` enum in search, discover, import, and resolve
routes. Validate `tmdbId` as a positive integer. Add a
`SORT_BY_PATTERN` regex and page-range check (1–500) to the discover
route. Wrap all outbound TMDB calls in try/catch and return 502 on
failure so clients get a structured error rather than an unhandled
rejection.
Fix optimistic-update rollbacks in `use-title-actions`: capture
`prevStatus` and `prevWatches` before each mutation and restore both
atoms in the catch block for catchUp, handleMarkSeason,
handleUnmarkSeason, and single-episode toggle.
Fix a bug in `getContinueWatchingFeed` where the watchDateMap could
hold a stale date for episodes watched more than once; the map now
keeps the most-recent `watchedAt` per episode.
Add createLogger to tmdb client, image-cache, availability, and
webhooks services. Replace all silent .catch(() => {}) in metadata.ts
with debug-level error logging. Add progress and success logs in cron
job bodies, colors extraction, and webhook processing. Fix pre-existing
unused parameter lint warning in title-card.tsx.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace nested loop queries with batch fetches using inArray() across
discovery, tracking, metadata, settings, and system-health services.
The biggest win is getContinueWatchingFeed() going from ~375+ queries
to 5. Add a cached getSession() wrapper via React.cache() to deduplicate
auth lookups shared between layouts and pages. Parallelize independent
async operations in importTitle(), cacheImagesJob(), and the explore page.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
drizzle-orm/bun-sqlite is fully synchronous — all queries return values
directly, not promises. Remove await from all db calls, drop async from
functions that no longer need it, simplify Promise.all patterns that
wrapped sync operations, and fix setSetting() which was missing .run()
(previously masked by await triggering execution via thenable).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Override CSS custom properties (--primary, --ring, --primary-foreground,
--status-watching) on the title detail page wrapper using the vibrant
color from the poster's node-vibrant palette. All Tailwind classes
referencing these variables automatically pick up the new color.
Also await lazy color extraction in getTitleWithChildren() so palettes
are available on first visit to never-before-opened titles.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace fully client-side dashboard and title detail pages with server
component orchestrators that fetch data directly via service functions,
eliminating extra network round-trips through API routes. Each section
streams independently through its own Suspense boundary.
- Extract getUserStats(), getTitleWithChildren(), getRecommendationsForTitle()
into service layer; update getNewAvailableFeed() to include tmdbId/voteAverage
- Split dashboard into server sections (stats, continue watching, library,
recommendations) with client children for animations
- Split title page into server hero + client interaction provider with shared
context for optimistic mutations across actions and seasons
- Add generateMetadata with OG tags, server-side TMDB ID resolution via
redirect(), loading.tsx and not-found.tsx for both pages
- Add per-section skeleton components, fix ContinueWatchingSkeleton dimensions
- Remove 6 unused API routes (feed/*, titles/[id] GET, titles/[id]/recommendations)
- Remove components/stats-summary.tsx, add lib/types/title.ts for shared types
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Navigate instantly to /titles/tmdb-{id}-{type} and show a skeleton
while the new /api/titles/resolve endpoint handles the full import
(TMDB fetch, availability, recommendations, colors). Existing titles
resolve in ~5ms; the URL is replaced with the UUID once resolved.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Recommendations were fetched fire-and-forget, causing a race condition
where the title detail page would load before recommendations were
populated in the database, making the section appear intermittently.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Introduce a local disk cache for TMDB images served through a proxy API
route (/api/images/[...path]), eliminating direct client-side CDN
dependencies. Images are cached by category (posters, backdrops, stills,
logos) and served with immutable cache headers.
Move all tmdbImageUrl() calls from client components to API routes and
server components so clients receive ready-to-use URLs. This removes the
need to expose TMDB_IMAGE_BASE_URL and IMAGE_CACHE_ENABLED via
next.config.ts env block. The landing page is split into a server
wrapper (app/page.tsx) and client component (components/landing-page.tsx).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extract dominant colors from movie/TV poster images server-side and use
them to create per-title atmospheric effects on detail pages — tinted
backdrop gradients, ambient glow orbs, and colored poster shadows.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Docker self-hosting support:
- Dockerfile (multi-stage Alpine build with tini init)
- docker-compose.yml with named volume for SQLite persistence
- /api/health endpoint for container health checks
- Auto-migration on startup via drizzle-orm/libsql/migrator
- Graceful shutdown (SIGTERM stops scheduler, closes DB)
- Next.js standalone output mode for minimal image size
Database driver migration (better-sqlite3 → @libsql/client):
- Eliminates native C++ compilation, enabling Alpine Docker images
- All DB queries converted from sync to async across services and routes
- DATABASE_URL now uses libsql file: prefix format
- drizzle.config.ts dialect changed to turso for libsql support
- Initial migration files generated in drizzle/
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
TV episodes were missing when: (1) a title existed as a recommendation
"shell" with no episode data, (2) a prior refreshTvChildren() call
failed partway through, or (3) the title detail page was loaded before
episodes finished importing. importTitle() now re-fetches episodes for
existing TV titles with zero seasons, the GET endpoint hydrates shell
titles on access, and refreshTvChildren() handles per-season errors
gracefully instead of aborting entirely.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add all 10 milestones: Drizzle ORM + SQLite database with WAL mode,
Better Auth email/password authentication, TMDB API integration for
search and metadata import, TV season/episode caching, user tracking
(watchlist/status/watches/ratings with auto-transitions), discovery
feeds (continue watching, library, recommendations), US streaming
availability via TMDB providers, background job scheduler with
instrumentation hook, and dark cinema-themed frontend with DM Serif
Display + DM Sans typography and amber accent design system.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>