Multimedia Console indexing PhotoPrism’s own thumbnails → endless re-index loop, count never drops, load pinned

English General
1

Setup

  • TS-253B, 8Gb, QTS 5.2.6.3229
  • Single Port Network
  • Drives (RAID 2):
    • WD80EFAX-68KNBN0
    • WD80EFAX-68LHPN0
  • ~421k actual original media files on disk, ~2.5M photos in Multimedia Console
  • PhotoPrism running in Container Station — it generates and stores its own thumbnail cache
  • QNAP Multimedia Console (media indexer + myconvertserver thumbnail/transcode daemon) + QuMagie (which depends on the Console’s index/thumbnails and adds its own face-recognition AI)

Symptom
After every restart the Multimedia Console starts indexing and never finishes. The image count in QuMagie does not fall — it stays huge and keeps churning. myconvertserver and the media scanner sit at high CPU indefinitely. System load climbs past ~8, to the point where sshd starts refusing new connections.

What I think is happening
The Multimedia Console is indexing the thumbnail/cache files that PhotoPrism generates as if they were new media. Every PhotoPrism thumbnail becomes another “photo” for the Console to index and generate its own thumbnail for — which lands in a folder PhotoPrism/Console then sees again. The result is a self-feeding loop that never converges.

The smoking gun: my media DB holds ~2.5M entries but there are only ~421k actual original files on disk. The ~2M difference lines up with PhotoPrism’s generated cache/thumbnail files being swept into the Console index.

(Context: in a previous session I relocated the Multimedia Console thumbnail cache to a different location, which may have made the Console re-scan and compounded this.)

What I’ve observed / tried

  • Restarting the Multimedia Console / QuMagie — indexing just resumes, count doesn’t drop.
  • Confirmed two thumbnailing engines running in parallel (Console myconvertserver + PhotoPrism), plus two face-recognition passes (QuMagie AI + PhotoPrism) competing for CPU.
  • set AddAllShareFolder to 0 in /etc/config/medialibrary.conf

Questions for the community

  1. What’s the correct way to exclude a folder (PhotoPrism’s storage/cache path) from the Multimedia Console / Media Library scan so it stops indexing generated thumbnails? Is there a supported per-share or per-path exclusion?
  2. Given the DB already has ~2M bogus entries, what’s the safe way to purge missing/invalid entries and re-baseline the media index without a full 2.5M re-scan?
  3. Is relocating the Console thumbnail cache supported, and if so how do I point it back cleanly so it doesn’t trigger a full regenerate?

Goal: get the Console to index only the ~421k real originals, ignore PhotoPrism’s cache entirely, and let the scan complete once.

Thanks — happy to share config, myconvertserver process details, and log excerpts.

2

I think there was an issue with Multimedia Console. There are some other threads on here where users have indicated the indexing never stops. I believe QNAP has issues an update that fixes this.

I would say you likely want to update QTS first. Update to 5.2.9.3499. Update to the latest Multimedia Console.

3

Thanks for the advice @NA9D

That didn’t really miraculously help.

I have updated Multimedia console, Docker storage engine, and removed all the photoprism thumbnails altogether. Otherwise the count of the images being processed in QuMagie was not going down.

Now I just restarted reindexing of all the library, and seems like it has kicked off thumbnail generation for QuMagie too.

4

So after the update, is this issue still occurring? Thanks!

5

The original issue disappeared after I have deleted the thumbnails in question.

Now I need to wait for the qumagie thumbs rebuild, to see if adding stuff to Container station share triggers it again.

I’ll report back if I have something.

6

Hi @beshur the team has confirmed that the Multimedia Console was indexing PhotoPrism’s storage/cache as originals. You can change PhotoPrism’s storage out of the scan by making that folder hidden. The Multimedia Console Console does not index folders whose name starts with a dot.

The PhotoPrism recommends this too, don’t place storage inside originals unless the name starts with “.”

  • Point PHOTOPRISM_STORAGE_PATH to a hidden path like .photoprism-storage and move existing storage there (config/sidecar/backup moved, not deleted; cache regenerates).
  • Leave originals as-is — that’s what QuMagie should index.

Once hidden, new files in the Container Station share won’t restart the loop.