Automating Media Management with the Arr Stack
How my Sonarr and Radarr setup is wired, what a quality sync tool did to it halfway through a 30TB re-download, and the two bugs that only showed up under load.

If you run your own media server, you've probably spent more time than you'd like babysitting downloads. The Arr stack (Sonarr for TV, Radarr for movies and the tools around them) takes most of that off your hands. It took me a fair bit of tuning before it stopped annoying me, though, and the worst lesson came in the middle of re-downloading my whole library.
What I run
The automation lives on one VM: Sonarr for TV, a second Sonarr just for anime because it needs different indexers, Radarr for movies, Bazarr for subtitles and Prowlarr for indexers. SABnzbd and qBittorrent do the downloading from behind a VPN container. Jellyfin is on a separate VM and only reads the finished files.
The one compose detail I'd copy into any setup is the volume mapping. Every container sees the NAS at exactly the same path it has on the host:
yamlsonarr: image: lscr.io/linuxserver/sonarr:latest volumes: - /opt/apps/sonarr/:/config - /mnt/nas/data:/mnt/nas/data radarr: image: lscr.io/linuxserver/radarr:latest volumes: - /opt/apps/radarr/:/config - /mnt/nas/data:/mnt/nas/data
Downloads and the library sit in the same dataset, under torrents/, usenet/ and media/. When SABnzbd reports a finished file, Sonarr finds it at the path it was given and moves it within the same filesystem. There are no path mappings to keep in sync, and an import doesn't copy 20GB.
The 30TB re-download
Earlier this year I had to rebuild my media pool from scratch (the ZFS side is its own post), so everything came back down again through this stack. It's a good stress test. A normal week is a handful of episodes. During the backfill Sonarr had over 3,500 items in its queue.
Around the same time I installed Recyclarr, which syncs the TRaSH Guides quality settings into Sonarr and Radarr. The TRaSH values are aimed at the best quality you can get, and they came with high minimums and no upper limit on anything 720p or above. I didn't catch it for a couple of days, and in that time the stack happily pulled 20GB movies and 100GB-plus series onto a pool that was being refilled from empty.

Removing Recyclarr didn't undo anything, because it had already written its values into each app's quality definitions. Those are stored as megabytes per minute of runtime, which is why the numbers look odd at first. I set them back by hand through the API, the same on every tier from 480p to 2160p:
json{ "quality": { "name": "Bluray-1080p" }, "minSize": 8, "preferredSize": 17, "maxSize": 34 }
For a two hour movie that's 1GB to 4GB, aiming for 2GB. Sonarr got 11, 27 and 67, which works out to roughly 500MB to 3GB for a 45 minute episode. Then I cleared about 1,650 Usenet grabs out of Sonarr's queue so they'd be searched again at the new sizes. I kept the torrents that were already running, since those have to seed.
So I don't use Recyclarr now. That's not a knock on the TRaSH Guides, they're written for people who want the best copy of everything. I want a library that fits on the disks I have.
Two things that broke under load
Sonarr backs off an indexer that keeps failing, and the bulk searches tripped rate limits on a few of them. They got pushed to the longest backoff, 24 hours, and stayed there after the indexers were fine again, because the state lives in Sonarr's own database. Deleting the rows from its IndexerStatus table got them searching again.
SABnzbd started rejecting grabs with an empty error. Sonarr logged Error response received from SABnzbd: with nothing after the colon. SAB's own log said it was ignoring a duplicate NZB. It remembers what it has seen in history1.db, separately from the history you can see in the UI, and turning duplicate detection off in the config didn't help. Removing the stale row and restarting the container did.
Checking on it
I wrote an MCP server that exposes the Sonarr, Radarr, SABnzbd, qBittorrent and Jellyfin APIs as tools, so I can ask for the queue, unstick a download or search for missing episodes in plain language. It's overkill, and I still use it all the time. I rarely open the Sonarr UI anymore.
The pool that got refilled is at 96% now, and that's with the caps in place.