I reinstalled Gladys with the Bullseye 64 image and as soon as I interact with it, for example by asking: « What’s the weather? », in the « Discussion » tab;
Gladys stops responding (no response + an infinite page reload after pressing F5) and I have to run « docker restart gladys » for it to work again.
I should add that the last request displayed and entered in « Discussion » was « Allume lampe yeelight », which is a device that I configured in MQTT + Node-RED and that works without any problem in Gladys.
I’ve edited your message, can you confirm that it’s indeed an issue on Raspberry Pi OS 64-bit and not an x64 system? 
Can anyone else reproduce it?
I couldn’t reproduce it on an x64 system in any case, I’ll test on Raspberry Pi OS..
Yes, my mistake — it’s indeed on Raspberry Pi OS 64-bit…
I confirm that this causes Gladys to bug (tested with Gladys Plus).
config: Pi 3 B+ official image.
restart after a few minutes.
And display of the response in the discussion chat.
Darn! Do you have a log to show us?
I’m setting up a test installation on a Raspberry Pi 3B+ right now to try to see what’s wrong too
I am not local.
It doesn’t happen all the time,
I just tried it again and there was no bug.
I can indeed reproduce this on a Raspberry Pi 3B+!
However, I don’t think it’s a crash — it’s more that the CPU is at 100% and running flat out.
Ok, I added a lot of logs locally on my Raspberry Pi 3B+, and it is indeed the classification that is ultra-slow on the Raspberry Pi 3B+…
The following line remains stuck on the slightest request:
This slowness does not occur on other platforms
What’s surprising is that this dependency hasn’t been updated recently, so it’s supposed to be the same code as in the last release…
The version in node_modules does match the version in the package.json
I think I had that issue, but not knowing how to explain it, I didn’t go any further.
I tried to upgrade to the latest version of the library (node-nlp@4.24.0), without success…
However, in the recent commits of the NLP library’s repo I see something that catches my eye:
Okay, I’m trying to install the previous version on my Pi 3B+ to see if that fixes the issue or not, because in this case nothing has changed in that part (no dependency updates, no code updates)
For your information, the issue is already present in the previous update (v4.8.10)
The issue is already present in Gladys Assistant v4.8.0, released in March 2022
I reverted to Gladys Assistant v4.7.0 (released 7 months ago in November 2021), and I already have the issue
Did this feature work on Raspberry Pi 3B+? ^^
Maybe machine learning simply doesn’t run on this type of device…
I’m going back to Gladys Assistant v4.1.0 (released in February 2021), and the bug is already there.
Already, that’s reassuring in a way: it’s not a bug that appeared out of nowhere — we didn’t touch anything, and it has indeed been there for a long time.
Now we need to see whether it’s related to 64-bit or not..
The bug is also present on the 32-bit image on the Raspberry Pi 3B+
My conclusion is that I think it never worked on the Pi 3B+…
So 2 solutions:
- Find out why this library behaves like this on those Pis, or find another, lighter machine learning technology!
- Disable this feature for those Pis
I can’t tell whether the Pi 3B+ is widely used or not!
Those of you on Raspberry Pi 4, do you have the issue as well?
1 Like
We should run a poll to find out which type of machine people use Gladys on (and why not include this data in Gladys’ stats reports if possible?)
The Raspberry Pi 4 is still relatively new — personally I’ve been on a Raspberry Pi 3B+ for 3 years and as long as it works… ^^’