Deduparr¶

Readiness probe calibration¶

deduparr runs a dedup cycle every DEDUPARR_SERVICE__INTERVAL (2m), during which its single-threaded HTTP server stalls /health/ready for about 13s.

The readiness probe originally used a 5s timeout, which timed out on roughly every dedup cycle: about 51 Unhealthy events an hour, never enough consecutive failures to drop the endpoint, so pure noise. Raising the timeout to 15s outlasts the stall, and the 30s period keeps readiness from adding load to the single-threaded server.

The liveness probe tolerates the same stall differently: its 10s timeout can still catch the stall once, but 3 consecutive failures 30s apart (about 90s) are needed before it actually restarts a hung process.

Liveness uses /health/live¶

/health/ready also checks that Sonarr and Radarr answer. While liveness pointed at it, every Sonarr or Radarr outage restarted deduparr, which was healthy and only waiting: 29 restarts on 2026-10-20 during the Sonarr v5 upgrade, and about one a day otherwise. Liveness now uses /health/live, which answers whenever the process does; readiness keeps /health/ready.