New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Background processing stops very often #412
Comments
Same here, and if I issue
So probably related and fixed by #398 ?
|
Same, would be nice if i could actually use those 16 cores dedicated to this container to speed things up |
same. 6 cores set, but i get 30% cpu util. in intervals |
same here with 4 cores and 8 GiB RAM. ~30% utilization for 2-5 min in 10min intervals. |
I am also having issues. There are sometimes half to whole hours between classifications and i have a queue of 52360 faces. |
Try manually updating to 3.1.1. Immediately after the update my server constantly used one core to classify files. Only bug I found so far is that the appinfo.xml is out of date and still display 3.1.0. |
I too am having this issue. Face recognition: 198618 Queued files, Last classification: 5 minutes ago [cron] Error: DivisionByZeroError: Division by zero at <>
at 2022-11-03T06:25:03+00:00 |
v3.1.2 should fix all of this. Let me know if it does. 🙏 |
Seems to have improved from hours but I still experience minutes (noticed up to 20) between identifications
…________________________________
From: Marcel Klehr ***@***.***) ***@***.***>
Sent: Thursday, November 3, 2022 6:58:52 PM
To: nextcloud/recognize ***@***.***>
Cc: ollioddi ***@***.***>; Comment ***@***.***>
Subject: Re: [nextcloud/recognize] Background processing stops very often (Issue #412)
v3.1.2 should fix all of this. Let me know if it does. 🙏
—
Reply to this email directly, view it on GitHub<#412 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABY656GDVIEE264JSMSBPH3WGP4NZANCNFSM6AAAAAARLBLVNE>.
You are receiving this because you commented.Message ID: ***@***.***>
|
I guess a setting is in order here, to allow everyone to tweak it to their liking. |
@marcelklehr is the interval hard-coded? I understand why, but when the queue is so huge, I'd rather have it use more resources and finish early. The pattern @derekakelly has looks very similar to what I experience. Docker stats show an idle of around 5% usage spiking towarda 600-700%? when identifying. |
In my case, the app seems to take 5mins break very 10 or so items. |
The cpu cores option is only applicable for a single job, though, and we don't want a single job doing all the work because this needs to scale to huge instances as well, with multiple worker nodes. It's not entirely trivial to make something scale up and down seamlessly, but we're trying to accomodate everyone :) |
Another issue that influenced this bug was that recognize would always assume people were using WASM mode, and scaled down the batch size accordingly, but erroneously, which causes a lot of pauses on fast machines. |
@marcelklehr would it assume WASM mode even when toggled in the UI? This could perhaps explain my experience. I am running on an i5 11400 with 12 threads so it would categorize it as a "faster" machine |
Yes, even when WASM mode is off the latest release uses a smaller batch size currently. |
v3.2.0 is out now which should fix the always-assume-WASM bug and allows setting custom batch sizes for all classifiers. Let me know about your experience with the latest version :) |
Hi, thanks for the update. How are you supposed to change the batch size? If I type a different number in the field in the settings menu, it will fall back to the old value after reloading the page. There doesn't seem to be a button to save/apply the changes. |
Sadly i get now an error in recognize and in the logs. Recognize:
|
The error stacks you posted are unrelated to recognize I'm afraid. |
That would be a bug :/ v3.2.1 should fix this. Thanks for the feedback! |
Describe the issue
I installed Recognize with NC 25 at the 19.10, and it's working on my images since this time. In the CPU usage from my nextcloud VM, it shows that the background process only works for some minutes and stops then:
In the settings of Recognize it displays the same:
It always just processes a small amount of images, and then it stops for a while.
My cron jobs are running every 15 mins, it looks like image processing also starts with every cron job and then runs for some minutes.
Currently, there are 11k images in the queue, and it is still increasing. This process would take multiple week to finish if it continues like this.
Expected behavior
I would expect the image recognition to run the whole time in the background, without stopping that often.
Recognize (please complete the following information):
Server (please complete the following information):
The text was updated successfully, but these errors were encountered: