Skip to content

Prowlarr search timeout is a hardcoded 60 s shared with the download-client timeout, and the retry sweep enqueues up to 50 searches at once #601

Closed

Description

@joeg-git

What happens: an automatic search that is still running when the HTTP client hits its ceiling is aborted at 60 s, and the release that would have been found is never seen. On the request this looks like a search that returned nothing; #522 (released) at least made the failure visible, but the ceiling itself is still there and is not configurable.

Measured on a 1.3.0 instance (self-hosted, 8 Prowlarr indexers, all on category 3030):

  • One aggregated search across 8 indexers took 17–82 s with ~20 searches in flight; 8 of 20 exceeded 60 s.
  • The nightly Retry Missing Torrents Search takes take: 50 (src/lib/processors/retry-missing-torrents.processor.ts:63) and enqueues a search per request, so a sweep puts ~50 searches in the queue in the same second.
  • Result over 72 h: 210 × Failed to search Prowlarr: timeout of 60000ms exceeded, affecting 101 of 140 requests that were waiting for a search. Every one of those is a search that had a chance of a result and was abandoned client-side.

Where it is hardcoded: src/lib/constants/download-timeouts.ts:10 — export const DOWNLOAD_CLIENT_TIMEOUT = 60000, still 60000 on the v1.3.0 tag, used at:

  • src/lib/integrations/prowlarr.service.ts:112 — the Prowlarr axios client, i.e. the search itself
  • src/lib/integrations/prowlarr.service.ts:377 — the RSS fallback fetch
  • plus the qBittorrent / Transmission / Deluge clients, which do not need the same ceiling (their own documented reason is a ~30 s indexer wait before allowing a .torrent download)

So one constant, sized for "download a .torrent file", also governs an aggregated multi-indexer search that legitimately takes over a minute under load.

What I'd expect: any of these would remove the need for a patched image (which today is the only way to raise it — the value is baked into the bundle):

  1. Make the Prowlarr search timeout configurable (setting or env), separate from the download-client timeout, with a default that survives a 20-way burst — something like 180 s.
  2. Bound the fan-out of the retry sweep (a concurrency limit, or a setting for one), so the queue does not put ~50 searches against the indexers at once. That is the root cause; the timeout is just where it surfaces.
  3. Failing that, raise DOWNLOAD_CLIENT_TIMEOUT to 120–180 s. I have run 180000 for a couple of months here and it aborted nothing it should not have, while the 60 s ceiling was visibly cutting off searches that had results.

Happy to test a build if it helps.

Activity

  1. kikootwo commented on Nov 16, 2026

    @kikootwo
    Owner

    Thanks for the detailed report, @joeg-git. This has been fixed and merged in #608. Prowlarr searches now have their own 180 second timeout instead of sharing the download client's 60 seconds, and the nightly retry sweep now starts two searches a minute instead of all of them at once. It will be included in the next release.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions