Indexer service pipeline

The indexer service watches the monitor table and searches torrent indexers for pending episodes and seasons. When it finds a match it inserts a download queue entry for the downloader service to pick up.

Startup

1. Database connection initialised (same PostgreSQL as core).
2. Provider packages self-register via init() (kbdex and Prowlarr are blank-imported in cmd/indexer/main.go).
3. PollAndQueue starts in a background goroutine.
4. Health/test HTTP server starts on port 8082.

Poll loop

Runs every prowlarrInterval seconds (default 10 s).

Timer fires
      │
      ▼
Query DB for one pending monitor
  WHERE monitored = true
    AND available = false
    AND (status = 'pending'
         OR status = 'searching' for > 10 min)
  ORDER BY is_season DESC  ← season packs searched first
      │
      ▼
Mark monitor status = 'searching'
      │
      ▼
Build SearchRequest {title, alternate_titles, season, episode, library_id, ...}
      │
      ▼
GetEnabled(db) → active providers (each enabled via its kbdexEnabled / prowlarrEnabled toggle; results from all are merged)
      │
      ▼
Provider.Search(ctx, db, req)   [see provider detail below]
      │
      ▼
buildMatchLog  →  filter by title similarity + season + episode
      │
      ▼
pickBest  →  quality rank + seeders sort, blacklist check
      │
      ├── match found → insert download_queue row, set monitor status = 'queued'
      └── no match    → set monitor status = 'pending' (retry next cycle)

Provider: kbdex

kbdex is an AniDB-aware search service. It handles title resolution and episode filtering server-side.

SearchRequest
      │
      ▼
lookupAnidbID: SELECT source_id FROM media WHERE id = library_id
      │
      ▼
GET {kbdexUrl}/search?anidb_id=N[&season=S][&episode=E]
      │  Results cached on disk keyed by "kbdex:{id}:s{S}:e{E}"
      │  Cache TTL = prowlarrCacheAge
      ▼
[]TorrentResult  (already filtered by kbdex server-side via anitopy)

Provider: Prowlarr

Prowlarr is a generic indexer aggregator. kbarr queries it with multiple title variants and merges results.

SearchRequest
      │
      ▼
Build query variants from title + alternate titles
  ├── Season searches: "Title S01", "Title Season 1", "Title Batch", ...
  └── Episode searches: "Title S01E05", "Title - 05", ...
      │
      ▼
For each variant: GET {prowlarrUrl}/api/v1/search?query=...&apikey=...
      │  Each variant result set cached on disk (prowlarrCacheAge TTL)
      ▼
Merge + deduplicate by DownloadURL
      │
      ▼
[]TorrentResult

Client-side filtering (shared by both providers)

After the provider returns results, buildMatchLog and pickBest apply a second pass:

buildMatchLog — runs each torrent filename through anitogo (Go torrent filename parser), then checks:

1. Title similarity ≥ matchThreshold (default 80%) using Sørensen–Dice bigrams
2. Parsed season matches the monitor's season
3. For season packs: rejects individual episode releases
4. For episodes: parsed episode matches exactly

pickBest — from passing results:

1. Removes blacklisted torrents
2. Drops results above the preferredQuality cap (e.g. drops 4K if preference is 1080p)
3. Sorts by quality rank descending, then seeder count descending
4. Returns the top entry

Key packages

Package	Role
internal/indexer/provider/	Provider interface, registry, shared cache utilities
internal/indexer/provider/kbdex/	kbdex provider implementation
internal/indexer/provider/prowlarr/	Prowlarr provider implementation
internal/indexer/service/	Poll loop, match scoring, download queue insertion
internal/parser/	anitogo wrapper with memoisation and Japanese episode fallbacks