Core service pipeline

The core service owns the database, exposes the REST API consumed by the frontend, and runs the availability scanner.

Startup

1. Database is initialised and migrations run.
2. Auth defaults (username / password hash) are seeded if absent.
3. The availability poller starts in a background goroutine.
4. The HTTP server starts on port 8080 (configurable via PORT).

Request flow

Browser / frontend
      │
      ▼
  HTTP server (chi router + huma v2)
      │  Bearer token validated on every secured route
      ▼
  Route handler (internal/core/api/handlers/)
      │
      ├── Library routes   → internal/core/db/  (Bun ORM, PostgreSQL)
      ├── Monitor routes   → internal/core/db/
      ├── Download routes  → internal/core/db/
      │                      + proxied calls to downloader service (port 8083)
      ├── Search route     → proxied call to metadata service (port 8081)
      └── Settings routes  → internal/config/  (key-value in DB)

The core service does not talk to qBittorrent or torrent indexers directly — those concerns are delegated to the downloader and indexer services.

Availability scanner

Runs every availabilityCheckInterval seconds (default 60 s). Caches each show folder's directory listing keyed by mtime, so a cycle with no disk changes reads no directories. Each cycle has three phases:

Phase 1 — mark available

Indexes media rows by sanitized media_folder (and title, as a fallback) to a library_id, then walks mediaPath one level deep. Each subdirectory is resolved to a single library; files inside are parsed for an episode number (the SxxExx pattern, then the anitogo parser for absolute numbering). Matching is season-agnostic: the folder already scopes the show, so a monitor resolves via {library_id, episode} and sets available = true. Because each anime season is its own media row and folder, the season digit in the filename (often an unreliable S01 for sequels) is intentionally ignored. Folder names mapping to two libraries are flagged ambiguous and skipped.

Phase 2 — clear stale available flags

For every monitor marked available = true, checks whether its file is still present. If gone and the status was downloaded (kbarr placed it there), the status resets to missing, not pending — so the indexer's missingRetryInterval throttles re-search instead of re-queuing every cycle. Otherwise it is marked unavailable.

Phase 3 — sync season monitors

Counts total vs. available episode monitors per library_id. If all episodes are available, the season monitor marks downloaded. If incomplete, it resets to missing (avoiding an immediate re-queue loop).

Key packages

Package	Role
internal/core/api/	huma v2 router, route registration, handler wiring
internal/core/db/	All Bun ORM queries (media, monitors, downloads, settings)
internal/core/auth/	JWT token issuance and validation
internal/core/clients/	HTTP client that proxies to the metadata service
internal/core/service/	Availability scanner
internal/config/	Settings schema, defaults, get/set helpers