Kometa¶

Kometa (kubernetes/apps/media/kometa) runs as a nightly CronJob in the media namespace, building Plex collections and overlays for Movies and TV. Titles it adds through Radarr and Sonarr carry the kometa tag there (plus kometa-top250, kometa-bestpicture or kometa-trending); review them with each app's tag filter.

Failed runs¶

Kometa exits 0 even when collections or playlists fail, and reports them only as Kometa Failure rows in its run log. The CronJob greps /config/logs/meta.log for that string and exits 1, so a failure surfaces as KubeJobFailed rather than a green Job. The review playlists failed on every run for days unnoticed this way: Kometa rejects radarr_taglist and sonarr_taglist inside playlists. For any other doubt, read the Error Summary block at the end of the Job's log.

The 02:00 schedule¶

MDBList's free tier resets its 1,000-request daily quota at midnight UTC. Kometa 2.5 honours Retry-After on an HTTP 429 with no cap, so a run that starts before the reset, while the previous day's requests are still spent, sleeps until the reset instead of failing. @daily in Europe/London runs at 23:00 UTC in summer, an hour before the reset, so every run used to start already out of quota.

The fix, in #5344, moved the schedule to 0 2 * * *. 02:00 local is 01:00 UTC in summer and 02:00 UTC in winter, always after the reset. 02:00 also survives the UK's spring DST transition: clocks jump from 01:00 to 02:00 GMT that night, so a 01:00 local schedule would be skipped.

The 20-hour run deadline¶

On 2026-09-23, a run stalled in OMDb and MDBList rate-limit sleeps and was still running three days later, silently skipping the 24^th, 25^th and 26^th under concurrencyPolicy: Forbid (#5429). activeDeadlineSeconds: 72000 now kills any run still going 20 hours after it starts, well past the roughly 4-hour length of a normal run, and before the next 02:00 schedule.