Indexers¶

This page is for the owner: how searching works. It covers Prowlarr and how it feeds the other apps, Byparr for sites behind Cloudflare, the two scripts that add and test indexers, the Usenet indexer, and optional trackers that need an account.

Prowlarr: one list for every app¶

Prowlarr (http://<tailscale-ip>:9696) is the arr app for indexers: the sites that are searched for releases. It holds the one list and pushes it to the apps that search. You never add an indexer to Radarr or Sonarr directly; add it once in Prowlarr and it appears in each app whose categories it carries.

flowchart LR
    public["Public torrent sites"] --> P["Prowlarr (:9696)"]
    usenet["NZBgeek, NZBFinder (Usenet)"] --> P
    accounts["Trackers with an account"] --> P
    P -. "indexers tagged byparr" .- B["Byparr (Cloudflare)"]
    P -->|"movie categories"| R["Radarr"]
    P -->|"TV categories"| S["Sonarr"]
    P -->|"audio categories"| L["Lidarr (music)"]
    P -->|"copied by configure-questarr.py"| Q["Questarr (games)"]
    slsk["Soulseek (Tubifarry)"] --> L

• Radarr, Sonarr and Lidarr are linked as Prowlarr applications by configure-arr.py, with full sync: Prowlarr adds, updates and removes their indexers itself. Each gets only the categories it uses (movies, TV or audio), which is why a movies-only site such as YTS appears in Radarr alone and a TV-only site such as EZTV in Sonarr alone. Lidarr is linked only when the music module is on.
• Questarr has no Prowlarr sync of its own. configure-questarr.py copies Prowlarr's indexers into it and removes the ones Prowlarr no longer has, so re-run it after adding an indexer. Questarr reaches Prowlarr at its fixed address, PROWLARR_IP, never by name: Prowlarr writes the host it was called on into its download links, and Questarr refuses links whose host differs from the indexer's address (known issue).
• Soulseek is not a Prowlarr indexer: configure-lidarr.py adds it to Lidarr directly through the Tubifarry plugin, searched first (Music flow).

Does it notice broken indexers?¶

Yes, continuously and on its own:

• Every query and grab is recorded per indexer: successes, failures and response time.
• An indexer that keeps failing is disabled automatically with exponential backoff, and re-enabled when it recovers.
• Persistent problems appear under Prowlarr → System → Health, with a banner in the UI.

You see it in three places: Glance's Lab page counts the indexers backing off and says when each retries; stack-health fails when no indexer passes a test; and indexers-check shows the detail on demand (below).

Sites behind Cloudflare: Byparr¶

Several big public sites sit behind Cloudflare's bot protection. A plain request gets a challenge page instead of results, and the indexer fails, often while it is being added.

The traditional answer, FlareSolverr, no longer works against current Cloudflare (TRaSH Guides reports it broken). This stack runs Byparr instead, its maintained drop-in replacement: a headless browser that solves the challenge and hands the page back.

• It answers only on the Docker network, at byparr:8191; no port is published.
• Prowlarr uses it as a FlareSolverr-compatible proxy tagged byparr, and only indexers carrying that tag go through it. EZTV does; everything else connects directly.
• A search through Byparr takes about 15 to 20 seconds while the challenge clears. That is normal, not a fault.
• It is the heaviest container when it works, being a browser (up to 2 GB of memory), and idles cheaply. Its health probe launches a real browser each time, so it runs often only while the container starts, then every 15 minutes.

No script registers the proxy. Do it once in Prowlarr: Settings → Indexers → Indexer Proxies → + → FlareSolverr, host http://byparr:8191/, tag byparr. Then give the tag to any indexer that fails with a Cloudflare error.

1337x is deliberately left out: its Cloudflare setup often blocks home addresses outright (error 1006), which no solver gets past. Knaben indexes its listings, so little is lost.

Adding indexers: configure-indexers.py¶

scripts/configure-indexers.py

It adds the indexers in its WANTED list to Prowlarr, then tests each one with a real search and prints the result. It is idempotent: indexers already present are left alone. The list:

To add more, edit WANTED at the top of the script (the names are Prowlarr's definition names) and run it again, or add them by hand in Prowlarr → Indexers → Add Indexer, which lists every definition Prowlarr ships. A Cloudflare-protected site commonly fails to add; that is expected, not a configuration error. Give it the byparr tag and try again.

Testing indexers: indexers-check¶

scripts/indexers-check        # or the alias: indexers

It tests every indexer in Prowlarr and prints Prowlarr's own statistics, then any health warnings:

INDEXER            ENABLED  TEST     QUERIES   FAILED  MS
--------------------------------------------------------------
EZTV               True     PASS     0         0       0
Knaben             True     PASS     8         0       477
YTS                True     PASS     6         0       351
--------------------------------------------------------------
  3 working, 0 failing

A failing indexer gets a line with the error below it. The same lives in the UI at Prowlarr → Indexers → Test All Indexers, and the statistics at Prowlarr → System → Status. The script reads Prowlarr's API key from its settings with sudo.

Usenet indexers¶

Usenet is the preferred source: it downloads at the full speed of your line, shares nothing, and runs outside the VPN. It needs two paid accounts, both optional: a Usenet provider (USENET_* in .env) and an indexer. The stack is set up for NZBgeek (NZBGEEK_API_KEY) and NZBFinder (NZBFINDER_API_KEY), which is strong on French releases; set either or both:

• configure-sabnzbd.py sets up SABnzbd with the provider.
• configure-arr.py adds each one to Prowlarr, which syncs it to every linked app, adds SABnzbd to Radarr, Sonarr and Lidarr, and sets their delay profiles: Usenet first, with torrents waiting 60 minutes (in Lidarr, Soulseek comes before both). Sonarr series tagged french don't wait, because their new episodes come from the ratio trackers.
• Questarr's game searches include Usenet results too; picking one sends it to SABnzbd.

Any other Newznab indexer can be added in Prowlarr the same way; it syncs like the rest.

Trackers with an account (optional)¶

Public trackers are patchy. Popular and recent titles are easy; older, foreign or obscure ones often aren't there at all, in any quality. No setting fixes that. The real fix is a tracker with an account (from a sign-up window or an invite), which slots into Prowlarr exactly like the others and syncs to every app.

configure-indexers.py already knows two semi-private French trackers, for French TV and film that the English sites rarely carry:

1. Sign up (free) at draupnirr.xyz or tr4ker.net.
2. Copy the API key from your profile page into .env as DRAUPNIRR_API_KEY or TR4KER_API_KEY.
3. Run scripts/configure-indexers.py again.

Both enforce a minimum ratio, and a tracker that falls below it stops handing out peers. So the script gives them Prowlarr's RSS only app profile: the apps grab new releases from their feeds, where plenty of people are still downloading and seeding pays back, and you can still pick from them in an interactive search. A backlog search skips them, because an old episode with hundreds of seeders and no downloaders costs ratio that seeding never earns back; Usenet and the public trackers cover those. Seeding is not affected: everything already downloaded keeps seeding.

Sonarr lists French shows under their English TVDB name, with "(FR)" added. It still matches French release names through the show's alternate titles, and names new series folders with the TVDB id so Jellyfin finds the right show.

Cheat sheet¶

More in Troubleshooting.