Media pipeline

The shikanime namespace runs a full self-hosted media stack. The pipeline is index → acquire → organize → serve, glued by shared storage.

Stages

Stage	Services
Index	prowlarr (indexer aggregation/proxy)
Request	seerr (requests), sonarr/radarr/whisparr/lidarr (TV/film/adult/music)
Acquire	qbittorrent + xqbit cleanup sidecar
Enrich	bazarr (subtitles), metatube (metadata), maintainerr (library hygiene)
Serve	jellyfin (media), immich (+ml, +postgres, +valkey) for photos

Servarr wiring

• The *arr apps own the library layout; jellyfin only reads it.
• Downloads land on host paths (/mnt/patchouli), libraries on Longhorn PVCs — the import step is the boundary.
• xqbit (qbittorrent-cleanup) runs as a sidecar loop: reaps missingFiles torrents and resumes recoverable errored ones (config via QBT_URL/QBT_USER/QBT_PASSWORD).

Images

Most media images come from the org's own registry — ghcr.io/shikanime/<app> built from nixpkgs via niximgs (jellyfin, radarr, sonarr, whisparr, syncthing, vaultwarden, postgresql, mlflow), built with wharf as the Skaffold custom builder. Manifests pin them by digest; Renovate bumps via image updates.

Photos: immich

immich deploys as a set: the server, immich-ml (machine-learning tagging/face detection), immich-postgres, immich-valkey (cache). Uploads flow app → postgres metadata + object storage on PVC; ML jobs run async through the valkey queue.