nextcloud
Face clustering stops while face recognition finishes
Which version of recognize are you using?
8.2.0
Enabled Modes
Face recognition
TensorFlow mode
Normal mode
Downstream App
Memories App
Which Nextcloud version do you have installed?
Nextcloud Hub 9 (30.0.5)
Which Operating system do you have installed?
ZorinOS / NixOS
Which database are you running Nextcloud on?
MySQL
Which Docker container are you using to run Nextcloud? (if applicable)
No response
How much RAM does your server have?
32GB
What processor Architecture does your CPU have?
x86x64
Describe the Bug
Within face recognition, face clustering stopps, as it seems randonly after a while. Face recognition itself finished operation but face clustering says 6.652 faces left to cluster; Scheduled background jobs: 0. It remains that way.
Expected Behavior
The run face clustering till the end and start again after new pictures appear.
To Reproduce
Resets do not solve the problem. After a while, it stops again.
Debug log
No response
(Sysadmin of the aforementioned server here)
The nextcloud.log didn't contain anything that caught my eye, so I did some further research and noticed that almost all of the background jobs seem to be missing:
[I] root@aidoskyneen /v/l/nextcloud# nextcloud-occ background-job:list | grep Recognize
| 444912 | OCA\Recognize\BackgroundJobs\MaintenanceJob | 2025-02-05T17:47:08+00:00 | null
So the issue probably is that the clustering background job simply doesn't run - and the initial face clustering seems to have happened because we manually ran nextcloud-occ recognize:cluster-faces.
Is there some way to manually re-register the mssing background jobs?
I have the same issue. Just finished classifying a day ago and have 38000 waiting to be clustered. occ recognize:cluster-faces does not do anything either.
mmh
Is there some way to manually re-register the mssing background jobs?
That's currently not possible I think.
occ recognize:cluster-faces does not do anything either.
Why is that? What happens when you run it?
command
sudo docker exec -u www-data nextcloud-app-1 php occ recognize:cluster-faces
result
ClusterDebug: Retrieving face detections for user User1
ClusterDebug: Not enough face detections found
Clustering face detections for user User 2
ClusterDebug: Retrieving face detections for user User 2
ClusterDebug: Not enough face detections found
Clustering face detections for user User 3
ClusterDebug: Retrieving face detections for user User 3
ClusterDebug: Found 13340 fresh detections. Adding 0 old detections and 0 sampled detections from already existing clusters. Calculating clusters on 13340 detections
Now that I ran it again it is skipping 2 users. It was not skipping them last night when I ran it and it did classify all of their files. User 1 and User 2 do not have any pictures to scan.
A little background,
This has worked fine for a couple years. I recently had a database issue where something was opening connections and not closing connections to the database. Resulting in nextcloud crashing after about 12 hours. I troubleshot this to the Recognize app, Memories App, or Face Recognition APP. Uninstalled all three, droped the tables from the database. (Mistake) After that nextcloud would not work at all. I had to manually reinstall all three apps to get the tables back. Then removed them using the App store. Then reinstalled recognize and got that set up. It scanned and classified all of the file but wont cluster for some reason.
Another question is how and which tables to remove from the the database for a complete uninstallation?
And it ends with this message? It may be running out of RAM, because there's a lot of face detections to cluster. I always recommend to use the cluster-faces command with the batch size parameter set to 10000 or even lower to allow it to fit in RAM.
yes then nothing further happens. I just tried with batch size 500 same result except this time not sure why but under the video tagging i now have
There are queued files but no background job is scheduled to process them.
That did not pop up until i ran the above command and it worked just this morning. Not sure if they are related or not.
I turned off GPU mode and the clustering started working with that command. The video also started working and scheduling background jobs. No background jobs are scheduled though for clustering. Is there a way to add that? The others schedule background jobs as soon as a picture or video is uploaded. The GPU ran all night last and completed the whole classify process but does not work this morning.
yes then nothing further happens. I just tried with batch size 500 same result
Well, clustering takes time, how long did you wait?
I turned off GPU mode and the clustering started working with that command.
That's peculiar. Clustering doesn't use GPU
It was about 20 minutes that I waited and nothing changed. But its working now and faces are showing with GPU off. Weird thing is now it wont detect my GPU if I turn it back on. Which is fine. Now that the initial classification is done I don't really need to go through the hassle of troubleshooting that. It was hard enough just to get working the first time. I will give it a couple days and see if the background job starts working. If it does not i will just make a cron job that runs that command until I hear if this is a recognize issue or just a me issue.
It was about 20 minutes that I waited and nothing changed.
That's likely not long enough, clustering takes quite some time.
The MaintenanceJob that runs every 12h should create missing ClusterFacesJobs actually.
@alyaeanyx If it doesn't you should have errors about that in your nextcloud log.
Ok thank you for the help! Just to note when i run the job manually I can see the numbers go down in the admin settings pretty instantly if that is what you were referring to.