Server load explodes and crashes the server when trying to "review face groups"

[amo13]

Unfortunately, it's been a while since I am unable to use the face recognition app. When I go to my personal settings and choose face recognition, I can see the recognized people. Then I click on one of them. Loading takes a bit of time because I have many many photos, but that's okay. When I click on "review face groups" to sort out false positives for that person, I can see a few thumbnail pictures appear but it's mostly the grey shapes staying that way:

If I take a look at htop on my server, I can see the load explode:

and some time later:

Even if I close the browser tab, my server stays unresponsive for a few hours. Before setting CPUQuota=300%, MemoryHigh=50% and MemoryMax=66% for the systemd service, even htop would freeze and I would not be able to ssh into the server for the few hours.

I suspected that some php bad configuration is the culprit but I couldn't find what to change. What's interesting to me is that with htop it looks like redis is the bottleneck. But I have no idea what is actually going on in the background when loading that page.
I'd appreciate any help with dealing with that issue.

[matiasdelellis]

Hi @amo13
Sorry for delay.. I'm very busy.. 😞
Please, try Memories. It should be much more efficient. 😉

[amo13]

Thanks for your response despite your lack of time, I appreciate it.
I tried and failed because of a Memories app issue. But even if it gets fixed, does Memories allow for easily sorting out false positives for a given person with a lot of photos?

[matiasdelellis]

About this issue, (I should fix it when I can), there is a trick. You can assign the name directly. (Unfortunately without autocompletion), but if you assign the name the clusters are automatically merged.

Memories allow for easily sorting out false positives for a given person with a lot of photos?

You can use the "Mark person in the preview" option and in any case fix it with the person sidebar

[amo13]

Ah, great that there is a workaround for the time being. Thank you very much for your help with this!

[36grad]

Coming a bit late to the party, but a workaround for the OP's issue for people who use Nextcloud with apache2 is to set MaxClients to a suitable number in mpm_prefork.conf (probably there is a similar solution with nginx).
Essentially the web UI will spawn a process for each of the pictures that need to be loaded. On my machine, each process would consume about ~100MB RAM, so with the default of 256 max client connections (effectively processes), that overloads my machine and leaves me with a lot of gray shapes.
Limiting the MaxClients to e.g. 40 will cause my user to create a web server RAM consumption of not more than 4GB which is for my family use case very effective. I have not noticed any negative side effects.

I guess somewhere in the javascript of the UI there should be a limit / queue so that not for each visible frame the data is fetched immediately from the web server, but only a limited number of items at a time.